Add files using upload-large-folder tool

00c2b94550129fb8f084bb495841a196a9f5afe6d5c14e29f461a9abeaa8a98e/peer_review/peer_review.md

@@ -0,0 +1,283 @@
+Peer Review File
+
+Enhancing Combinatorial Optimization with Classical and Quantum Generative Models
+
+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/.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+The authors propose a framework that incorporates quantum or classical generative models to tackle optimization problems.
+The main focus of the work is problems related to portfolio optimization and a quantum-inspired instantiation of the framework.
+
+The paper is generally well-written and presents an interesting approach and results.
+However in its present form I have difficulty assessing the significance, scope, and impact of the results, which are paramount to my recommendation.
+
+Hence at present I recommend additional revision of the manuscript.
+I provide some further comments below.
+
+Comments:
+
+-What is scope and impact of the results? A general framework is described, but what specific instantiations and cases should we expect to provide greatest advantage?
+(I clearly see, at minimum, a classical-ML approach to portfolio optimization, and would appreciate more clarity regarding more general settings)
+
+-Results are shown for portfolio optimization, but wasn't clear to what other important problems the approach is best suited for or should be expected to be advantageous.
+
+-In particular, how common are problems where "cost function evaluation can be very expensive" ?
+Can this be quantified? For instance, what is the resource tradeoff between query cost and runtime? Is your approach still competitive if this is relaxed?
+
+-The results rely on a quantum inspired method, what evidence or support do we have that
+using actual quantum devices will improve further? How should we expect GEO to perform for other quantum or classical models?
+
+-For the quantum case, is the proposed method not still severely limited by the underlying quantum hardware and quantum model resource requirements? (in contrast, for instance, to problem decomposition approaches where the goal may be to accommodate fewer quantum resources)
+
+-Regarding "GTS" optimizer in Section III.C and the subsequent reported results for it, I was unclear what was meant. Each of G,T,S refers to an independent optimization strategy, are you running each and reporting the best?
+
+-Similarly, I was left with follow up questions such as to what degree were these (G,T,S) as well as the other nine "leading SOTA optimizers" tailored/optimized for the problem at hand? (In regards to truly 'fair' comparison in the reported numbers)
+
+-Table I is not so easy to read with the many entries. Is it worth including first 4 optimizers here? They don't seem competitive (no to mention their columns are mostly "-"s)
+
+-RCABC appeared to preform comparably to TN-GEO, it would be nice to see more discussion of this or even a more detailed comparison
+
+Reviewer #2 (Remarks to the Author):
+** Key results **
+
+This paper proposes to use a quantum inspired generative model to help more efficiently explore the space of feasible solutions for combinatorial optimization problems. The generative model is used to sample new candidate solutions: from a set of already explored solutions with their associated cost, the generative model is trained to learn a distribution over the solution space for which the probability of each seen solution is proportional to their associated cost, thus potentially making it possible to sample new promising candidate
+solutions. The authors focus on using a quantum inspired generative model based on tensor networks (matrix product states), which have been previously introduced in [15].
+
+The proposed approach is evaluated on the task of portfolio optimization and compared with state of the art optimization algorithms for this task. The experiments reveal that the approach is competitive with these state of the art solvers (which have been fine tuned for decades), even outperforming some of them on this particular task.
+
+** Validity **
+
+The results presented in the paper are valid. Both the methodology and the experiment setup is sound.
+
+** Data & methodology **
+
+The paper is overall clearly written and the experiments demonstrate well the potential benefits and usefulness of the method. One concern that I have is that the advantage of using a quantum approach is not clearly demonstrated. In particular, it is not clear how the fact that quantum inspired models where chosen for the generative part is key to obtaining the experimental results. At the very least I believe the proposed approach should be compared with replacing the MPS model with a simple HMM learned using the Baum-Welch (i.e. EM) algorithm. Other generative models should be considered as baselines as well (e.g. a simple RNN trained using backpropagation through time or a more complex model such as NADE [Uria et al]). It may be the case that replacing the MPS model by such an alternative non-quantum inspired model would lead to similar result. To sum up, I believe the authors should experimentally investigate and discuss more in depth to which extent the quantum part of their approach is necessary and beneficial.
+
+Uria, Benigno, et al. "Neural autoregressive distribution estimation." The Journal of Machine Learning Research 17.1 (2016): 7184-7220.
+
+** Appropriate use of statistics and treatment of uncertainties **
+Yes, the results are reported appropriately using classical statistical tools and treatment of uncertainties.
+
+** Conclusions: Do you find that the conclusions and data interpretation are robust, valid and reliable? **
+
+Overall the conclusions and interpretation of the experimental results are robust and reliable. Though, as I mentioned above, I believe there is a set of experiments missing which would demonstrate more the relevance of a quantum approach to the generative model part of the proposed method.
+
+** References **
+
+The manuscript references previous literature appropriately.
+
+** Clarity and context **
+
+The paper is very well written and structured, and easy to follow for someone familiar with tensor network models.
+
+** Suggested improvements **
+
+- Include an experiment to compare the proposed approach with non-quantum generative models
+
+- It could be a nice addition to have experiments on another combinatorial optimization problem than portfolio optimization.
+
+- Related to the previous point, the authors may consider presenting their approach in a more general context rather than specifically for the portfolio optimization problem. As I understood it, the proposed approach can be applied to many different kind of optimization
+problems but the current presentation can suggest that the method is tailored specifically for portfolio optimization. A more general presentation of the method, as well as a clear explanation of the different kind of problems the approach can be applied to, could make for a more impactful paper (by reaching a wider audience). For example, can the approach be applied to any MIP?
+Reply to Reviewer #1:
+
+The authors propose a framework that incorporates quantum or classical generative models to tackle optimization problems. The main focus of the work is problems related to portfolio optimization and a quantum-inspired instantiation of the framework.
+
+The paper is generally well-written and presents an interesting approach and results. However in its present form I have difficulty assessing the significance, scope, and impact of the results, which are paramount to my recommendation.
+
+Hence at present I recommend additional revision of the manuscript. I provide some further comments below.
+
+Reply: We thank the reviewer the careful revision our manuscript, and for the positive feedback about the current form of the manuscript. We address each of the questions below with the hope this helps to clarify the scope and significance of our work.
+
+-What is scope and impact of the results? A general framework is described, but what specific instantiations and cases should we expect to provide greatest advantage?(I clearly see, at minimum, a classical-ML approach to portfolio optimization, and would appreciate more clarity regarding more general settings)
+
+-Results are shown for portfolio optimization, but wasn't clear to what other important problems the approach is best suited for or should be expected to be advantageous.
+
+Reply: We would like to mention that beyond our proposal being a “a classical-ML approach to portfolio optimization”, the baseline of our contribution is a ML framework that leverages classical, quantum-inspired, or quantum generative models for solving any combinatorial problems. Although not a requirement for using the framework, having the possibility to easily generate bitstrings in the valid solution space (as the case studied here, i.e., bitstrings with a specified cardinality) can be useful since these synthetic data points can be passed as a warm start for the generative model. This in turn can help guide the trained generative model towards the right support of the probability distribution to be explored.
+
+Using ML techniques for combinatorial problems is not a straightforward and rather a new research domain. More specifically, although other reinforcement learning and deep learning approaches had been proposed (see e.g., Ref. [1] and [2]), to the best of our knowledge our work is the first proposal using generative models for combinatorial optimization tasks. This is also the first proposal flexible enough to easily incorporate and explore quantum-inspired or quantum generative models. At first, we thought of adapting or enhancing existing reinforcement learning and deep learning proposals with quantum models, but this was not practical or at least not obvious how to do it. Instead, we noticed using generative models could be a very promising approach, given its prominence as candidates for practical quantum advantage. But this required to go beyond the incremental adaptation and to design the algorithm in a completely different framework, which resulted in the GEO proposal here.
+Here are more general settings that make this approach advantageous over other available solvers:
+
+1) The entire approach is data-driven: what this implies is that the more data is available, either from previous attempts to solve the problem with other state-of-the-art solvers, the better the performance is expected. In the example of GEO as a booster we used data explored by Simulated Annealing (SA) but if we had previous observations from any or many other solvers, we could combine it and give it as a starting point to GEO.
+2) It leverages the power of generative models: The essence of the solver is that it is aiming to unveil non-obvious structure in the data, and once it has captured those correlations, it suggests new outstanding candidates with features similar to the top ones seen until that iteration phase.
+3) the model is cost function agnostic, i.e., it is a black-box solver. This is paramount since any cost function can be solved with our approach. Most of the proposals for quantum or quantum inspired optimization require the cost function of the problem to be mapped to a quadratic or polynomial expression. This black-box feature open the possibility to tackle any discrete optimization problem, regardless of how complicated or expensive it is to compute the cost function. This is possible within GEO since the only information passed to the generative model are the bitstrings who have been explored and their respective cost value.
+4) Nothing is special about the portfolio optimization problem. This follows from the item above. The main motivation for selecting this specific instantiation of portfolio optimization was the availability of concrete benchmarks and an extensive literature of solvers which have been fine-tuned for over the past decades. Every time a new metaheuristic is proposed, chances are this cardinality-constrained portfolio optimization problem is used to benchmark. Other recent independent works have considered other real-world applications of GEO. For example, in Ref. [3], the authors considered an industrial case related to a floor planning NP-hard problem. This black-box feature is one of the most prominent ones which render our approach advantageous compared to other quantum heuristics, such as the quantum approximate optimization algorithm (QAOA), which relies on the cost function to be a polynomial in terms of the binary variables.
+
+Although the items above only point to features which make the approach valuable and different when compared to optimizers proposed to date, it is still to be explored if similar advantages can be observed in other family of important problems. Although this question is an active research question, there has been some progress in understanding which datasets or problems might benefit the most. In a recent publication, (see Ref. [4]), we showed how the search space of problems with equality constrains, can be represented efficiently in quantum-inspired generative models, but this is not the case for deep-neural-network-based generative models. Besides the demonstration presented here pointing that quantum and quantum-inspired models are a promising route to be explored, this more recent paper poses a more concrete differentiator between the capabilities of quantum-inspired models with respect to traditional deep learning models. These equality constraints appear in a broad family of problems, including the cardinality constrained portfolio optimization studied here, but also it extends to problems with any linear equality constrains.
+
+Revisions to the manuscript: The new Introduction has been significantly rewritten and restructured for clarity and to indicate more explicitly these salient highlights indicated above. We also revised the Outlook section adding some material related to the recent developments towards the identification of data sets which might benefit from quantum-inspired and quantum models over state-of-the-art generative models
+-In particular, how common are problems where "cost function evaluation can be very expensive" ? Can this be quantified? For instance, what is the resource tradeoff between query cost and runtime? Is your approach still competitive if this is relaxed?
+
+Reply: We thank the referee for bringing this point since some clarifications can be added to the manuscript. Note that this desirable condition of having an expensive cost function applies mainly to GEO as a standalone, although it is important to note that even in that setting it is not a necessary condition. The reason is because the overhead of training the generative model might have a larger impact for that setting than in the booster mode, where the initial available data can be seen as a warm start. It can be noted that even in problems with cost function which might be quick to evaluate, as long as the problem is hard and a solution is not reached with a competitor solver, it might be still worth to use GEO as a standalone which might yield a completely different set of solutions, or even better ones, since its search strategy is completely different.
+
+This condition of an expensive cost function was relaxed altogether in the setting of GEO as a booster (described in Sec. II A) where the comparison criteria was moved from number of cost function evaluations to runtime. In that study, we used a similar criteria as that suggested by the referee, where we used the trade off between query cost and runtime by using the more generic and practical criteria of fixed total wall clock runtime used for each of the algorithmic strategies. There, the time it takes to evaluate the cost function and the training of the generative models are considered, since these are part of the total time it takes to run each algorithm. In that comparison, between Simulated Annealing (SA) and GEO, although the training of the generative model within GEO takes longer, compared to the quick updates in SA, we still see an advantage in adopting GEO as a booster, which consisted of GEO being initialized with partial solutions obtained from SA. We show in that section that it is more efficient to change the strategy to GEO, than to insist and continue performing runs just with SA, if both algorithmic strategies are given the same total wallclock time.
+
+Revisions to the manuscript: We have added further clarification in the new paragraph right before subsection IIA.
+
+-The results rely on a quantum inspired method, what evidence or support do we have that using actual quantum devices will improve further? How should we expect GEO to perform for other quantum or classical models?
+
+Reply: The question of whether a significant advantage can be obtained by using quantum devices is an active research topic and certainly needs to be explored further. One proposal to reach a more systematic and incremental enhancement from the best quantum-inspired solution to an enhanced quantum-hardware realization was recently proposed in Ref. [5]. There, one starts from the best available quantum-inspired tensor-network solution and maps it to a quantum circuit. This can be subsequently modified by adding gates beyond those from the decomposition to increase the plausible correlations beyond those accessible with the quantum-inspired tensor-network-based solution. The access to longer-range correlations enhances, in turn, the expressibility of the quantum generative model while taking it beyond the capabilities of classical simulation. In that work, the specific case of generative models was illustrated, and therefore, these novel decomposition techniques can be directly applied to extend the capabilities of TN-GEO explored here, and, as the technologies mature and the level of noise is reduced, explore these enhanced models directly on quantum devices. Additionally, in Ref. [6] a comparison of quantum
+generative models with state-of-the-art classical generative models was presented, and the results were very encouraging in the data sets studied.
+
+To the comment of the performance of TN-GEO against other classical models, and as part of the request from the second referee, we added a classical version of GEO, based on the Neural Autoregressive Density Estimation (NADE) model [7]. In this revised version of the manuscript, we showed that although NADE-GEO is competitive with the early solvers (from about a decade ago), it is underperforming compared to TN-GEO and the other state-of-the-art solvers within the past five years. Although we don’t expect the results to be universal, in this specific TN-GEO versus NADE-GEO comparison we still see an advantage for the quantum-inspired over the classical generative neural network model, with the advantage as well that the TN-base model has less hyperparameters to fine tune. Each of these comparisons is an extensive amount of work, and we feel that despite being beyond the scope of this particular work, our work opens the possibility to explore quantum and quantum-inspired generative models toward solving arbitrary combinatorial optimization, and as a concrete framework to study quantitatively practical quantum advantage with future quantum technologies (see e.g., Ref. [6] whose framework is inspired in this work)
+
+Revisions to the manuscript: We complemented the Outlook where we had already mentioned about the opportunities and challenges that arise from thinking of an implementation in quantum devices. In particular, we added some of the discussion above around the new references that appeared after we submitted our work. We have also expanded the Results & Discussion subsection, Sec. IIC, with the respective discussion related to NADE-GEO and TN-GEO.
+
+-For the quantum case, is the proposed method not still severely limited by the underlying quantum hardware and quantum model resource requirements? (in contrast, for instance, to problem decomposition approaches where the goal may be to accommodate fewer quantum resources)
+
+Reply: If we correctly understand the proposal from the Reviewer, the reviewer is thinking of problem decomposition techniques which divide the problem into many smaller subproblems to accommodate the number of qubits in near-term technologies. From a practical point of view, for these techniques to be promising for quantum technologies, one still needs the subproblems still be intractable or hard to solve by conventional classical solvers, otherwise, it would not be worth to submit them to a quantum hardware. If the instances happen to be indeed intractable, then GEO can still be used to solve these subproblems, since in the cases of decomposition techniques we are familiar with, it is still required that solutions to the subproblem needs to be gathered before they are combined or used to solve the larger problem (e. g., see Refs. [8], [9], and [10] below and that have been added to the revised manuscript). In other words, we don’t see it necessarily as one strategy versus the other, since we think of GEO as a standalone strategy to solving hard optimization problems or subproblems. It is important to note that, with now gate-based devices reaching the level of hundreds of qubits, such techniques and demonstrations can become more relevant. But we can see how GEO can be incorporated as well in the solution of the smaller but intractable partitions which are sent to these quantum devices.
+
+To the question of the limitation of quantum hardware resources, not only in terms of number of qubits, but also in terms of connectivity, this is one is an important one and we will be addressing that in a theoretical and experimental ongoing work in our team. In that study, noise is also factored as part of the experimental demonstration.
+Revisions to the manuscript: We have complemented the Outlook addressing both, the potential interplay with other decomposition techniques, and the open questions and challenges related to hardware limitations.
+
+-Regarding "GTS" optimizer in Section III.C and the subsequent reported results for it, I was unclear what was meant. Each of G,T,S refers to an independent optimization strategy, are you running each and reporting the best?
+
+-Similarly, I was left with follow up questions such as to what degree were these (G,T,S) as well as the other nine "leading SOTA optimizers" tailored/optimized for the problem at hand? (In regards to truly 'fair' comparison in the reported numbers)
+
+Reply: It is important to note that we have not fine-tuned all the solvers ourselves. The Results section of the papers is broken into three subsections, each highlighting different features from GEO. Sec. IIA focuses on GEO as a booster and how it can build from results obtained with other solvers. Sec. IIB focuses on GEO as a standalone and compares its performance to SA and the Bayesian optimization library GPyOpt. Finally, Sec. IIC focuses on a comparison of GEO with state-of-the-art solvers. While in Secs. IIA and IIB we implemented and fine-tuned each solver, in Sec. IIC, we leverage the state-of-the-art results from nine other solvers reported in the literature in the last two decades. In the latter case, each non-GEO solver was thoroughly fine-tuned by the researchers of each reference. This portfolio optimization problem is so canonical that when a new solver is proposed, researchers can compare their results by taking the results from the new proposed solver, as long as the benchmark problems are run in identical conditions. This was one of the main motivations for us to choose this well-established benchmark problem. The "rules of the game" for reporting each market index and performance indicator are reported in Appendix A 2. In contrast, for the other two subsections, the criteria of evaluation are different, and it emphasizes the performance of GEO when one imposes a limit on the total wall-clock time (Sec. IIA) and when there is a limited number of calls to the cost function (Sec. IIB). The latter is a potential scenario when the bottleneck or expensive step is the cost function evaluation itself (e.g., as it is the case of drug discovery where each evaluation (each candidate molecule) might require synthesis in the lab and an expensive and long process towards its Food and Drug Administration (FDA) approval).
+
+To the specific question of (G,T,S) the Reviewer is correct that the authors from the original reference back in 2000 ran each strategy (Genetic algorithms, Tabu search and Simulated annealing) and reported their best results for each. Since this paper was one of the first ones adopting the current metrics and benchmarking procedure, when the community started creating more sophisticated metaheuristic strategies, they continue referring to the results in this paper as GTS, implying the values correspond to the best result from either G, T, or S
+
+Revisions to the manuscript: To clarify this to the reader, we have emphasized further in Sec. IIC that the results from the state-of-the-art competitors have been independently fine-tuned and are available in the literature. We have also added more context at the beginning of the Results section to contextualize all the different results subsections and strategies presented in the paper.
+-Table I is not so easy to read with the many entries. Is it worth including first 4 optimizers here? They don't seem competitive (no to mention their columns are mostly "-"s)
+
+Reply: We have followed the Reviewer’s suggestion and we have simplified the table accordingly (please see new Table I). The main reason for including all the solvers was to follow the format from previous papers which compared to these early metaheuristics from the early 2000’s. Since this cardinality constrained problem has been used for benchmarking solved since the late 1990s, these were the solvers which have been competitive since the last two decades or so ago. As time has passed, and in particular in the last decade, other researchers have proposed other figure of merits to assess the performance of the solvers. Since these were not known in the first papers, they have been conventionally included in the most recent papers with the ‘-’ since the data is not available. Others have adopted these other metrics as valuable and that is the reason the more recent papers all evaluate them, and we decided to include them here as well.
+
+The original intention from having the complete table was to have the whole spectrum of solvers and help position our quantum-inspired GEO in the historical progress of optimizers for this specific task, clearly showing GEO outperforms these solvers from the early 2000’s up to a decade ago, and being on par with the current state-of-the-art optimizers. Note that with the suggestion from the other Reviewer to include a classical version of GEO, called NADE-GEO in the revised version, we can see that NADE-GEO is competitive with solvers from about a decade ago, but not as competitive as TN-GEO or the new solvers.
+
+Revisions to the manuscript. We have collapsed the table appearing in the main text as suggested by the reviewer. The original table, expanded now with the NADE-GEO results, has been moved to Appendix C.
+
+-RCABC appeared to preform comparably to TN-GEO, it would be nice to see more discussion of this or even a more detailed comparison
+
+Reply: We have expanded the discussion on RCABC at the end of the Results section (Sec. IIC). Note in the current version we have replaced the name RCABC to ABC-HP to match the name of the best performing variant in that 2021 paper cited, and which corresponds to the results reported here.
+
+References:
+
+[1] Bello, I., Pham, H., Le, Q.V., Norouzi, M. and Bengio, S., Neural combinatorial optimization with reinforcement learning. arXiv preprint arXiv:1611.09940 (2016).
+
+[2] Bengio, Y., Lodi, A. and Prouvost, A., Machine Learning for Combinatorial Optimization: a Methodological Tour d’Horizon. CoRR abs/1811.06128 (2018). arXiv preprint arXiv:1811.06128 (2018).
+
+[3] Banner, W.P., Hadiashar, S.B., Mazur, G., Menke, T., Ziolkowski, M., Kennedy, K., Romero, J., Cao, Y., Grover, J.A. and Oliver, W.D., Quantum Inspired Optimization for Industrial Scale Problems. arXiv preprint arXiv:2305.02179 (2023).
+
+[4] Lopez-Piqueres, J., Chen, J. and Perdomo-Ortiz, A., Symmetric tensor networks for generative modeling and constrained combinatorial optimization. Machine Learning: Science and Technology. 4, 035009 (2023).
+[5] Rudolph, M.S., Miller, J., Motlagh, D., Chen, J., Acharya, A. and Perdomo-Ortiz, A., Synergy between quantum circuits and tensor networks: Short-cutting the race to practical quantum advantage. arXiv preprint arXiv:2208.13673 (2022).
+
+[6] Hibat-Allah, M., Mauri, M., Carrasquilla, J. and Perdomo-Ortiz, A., A Framework for Demonstrating Practical Quantum Advantage: Racing Quantum against Classical Generative Models. arXiv preprint arXiv:2303.15626 (2023).
+
+[7] Uria, B., Côté, M.A., Gregor, K., Murray, I. and Larochelle, H., Neural autoregressive distribution estimation. The Journal of Machine Learning Research, 17(1), pp.7184-7220 (2016).
+
+[8] Ponce, M., Herrman, R., Lotshaw, P.C., Powers, S., Siopsis, G., Humble, T. and Ostrowski, J., Graph decomposition techniques for solving combinatorial optimization problems with variational quantum algorithms. arXiv preprint arXiv:2306.00494 (2023).
+
+[9] Ushijima-Mwesigwa, H., Shaydulin, R., Negre, C.F., Mniszewski, S.M., Alexeev, Y. and Safro, I., Multilevel combinatorial optimization across quantum architectures. ACM Transactions on Quantum Computing, 2(1), pp.1-29 (2021).
+
+[10] Zhou, Z., Du, Y., Tian, X. and Tao, D., QAOA-in-QAOA: solving large-scale MaxCut problems on small quantum machines. Physical Review Applied, 19(2), p.024027 (2023).
+Reply to Reviewer #2
+
+** Key results **
+This paper proposes to use a quantum inspired generative model to help more efficiently explore the space of feasible solutions for combinatorial optimization problems. The generative model is used to sample new candidate solutions: from a set of already explored solutions with their associated cost, the generative model is trained to learn a distribution over the solution space for which the probability of each seen solution is proportional to their associated cost, thus potentially making it possible to sample new promising candidate solutions. The authors focus on using a quantum inspired generative model based on tensor networks (matrix product states), which have been previously introduced in [15].
+
+The proposed approach is evaluated on the task of portfolio optimization and compared with state of the art optimization algorithms for this task. **The experiments reveal that the approach is competitive with these state of the art solvers (which have been fine tuned for decades), even outperforming some of them on this particular task.**
+
+** Validity **
+The results presented in the paper are valid. Both the methodology and the experiment setup is sound.
+
+** Data & methodology **
+The paper is overall clearly written **and the experiments demonstrate well the potential benefits and usefulness of the method**.
+
+Reply: We thank the referee for the very positive feedback on our proposed framework. Below we address the concerns raised in the comments.
+
+One concern that I have is that the advantage of using a quantum approach is not clearly demonstrated. In particular, it is not clear how the fact that quantum inspired models where chosen for the generative part is key to obtaining the experimental results. At the very least I believe the proposed approach should be compared with replacing the MPS model with a simple HMM learned using the Baum-Welch (i.e. EM) algorithm. Other generative models should be considered as baselines as well (e.g. a simple RNN trained using backpropagation through time or a more complex model such as NADE [Uria et al]). It may be the case that replacing the MPS model by such an alternative non-quantum inspired model would lead to similar result. To sum up, I believe the authors should experimentally investigate and discuss more in depth to which extent the quantum part of their approach is necessary and beneficial.
+
+Uria, Benigno, et al. "Neural autoregressive distribution estimation." The Journal of Machine Learning Research 17.1 (2016): 7184-7220.
+
+Reply: We agree with the Reviewer this is a critical point that deserves further study since it is in general an open question. One of the highlights for GEO is the flexibility of swapping classical, quantum inspired, and quantum models in the future as quantum technologies mature. We have followed the suggestion from the Reviewer and implemented NADE, which is the most complex model of the ones suggested. In the new results summarized in Table V (computed from the explicit values in Table III), it can be seen that although NADE-GEO is a competitive solver, it is only statistically on par with the solvers which were the best performers up to about a decade ago (GTS, IPSO, IPSO-SA, and PBILD), but statistically different from its quantum-inspired version (TN-GEO), and the solvers proposed from 2015
+and onwards (GRASP, ABCFEIT, AAG, VNSQ, and ABC-HP). Although this does not constitute a proof in any way of the superiority of quantum or quantum-inspired models, this corresponds to the experimental investigation which hints to the value and promising directions we highlight in this work from this flexible framework.
+
+To the milestone of understanding when quantum or quantum-inspired models could be key to obtaining an advantage over classical models, several subsequent publications have attempted to address this point. For example, in Ref. [1], we have recently published a version of TN-GEO which uses symmetries which are relatively easy in these quantum-inspired TN models. We showed how the search space of problems with equality constraints can be represented efficiently in quantum-inspired generative models, but this is not necessarily the case for classical deep-neural-network-based generative models due to their non-linear activation units. That recent work poses a more concrete differentiator between the capabilities of quantum-inspired models with respect to traditional deep learning generative models. In general, the question of practical quantum advantage in quantum machine learning is still wide open, but in this recent work, Ref. [2] below, the proposed framework there leverages our GEO proposal to establish a clear-cut criteria from optimization problems to concretely test the performance of both classical and quantum generative modeling proposals. From a practical point of view, and in case of real and relevant real-world applications, this question would need to be approached on a case-by-case basis, but that subsequent paper proposes a framework where the performance of quantum generative models can be tested, following similar metrics inspired from this work.
+
+** Appropriate use of statistics and treatment of uncertainties **
+Yes, the results are reported appropriately using classical statistical tools and treatment of uncertainties.
+
+Reply: We thank the referee for the positive feedback on our statistical analysis.
+
+** Conclusions: Do you find that the conclusions and data interpretation are robust, valid and reliable? **
+Overall the conclusions and interpretation of the experimental results are robust and reliable. Though, as I mentioned above, I believe there is a set of experiments missing which would demonstrate more the relevance of a quantum approach to the generative model part of the proposed method.
+
+Reply: We thank the referee for the positive feedback on our overall conclusions. We hope the new numerical experiments, including the simulations of NADE-GEO strengthens our claims around the potential benefit of quantum inspired generative models, and in general, of quantum models as an alternative to be explored in the near term.
+
+** References **
+The manuscript references previous literature appropriately.
+
+Reply: We thank the referee for the positive feedback on our literature coverage.
+
+** Clarity and context **
+The paper is very well written and structured, and easy to follow for someone familiar with tensor network models.
+Reply: We thank the referee for the positive feedback on the content and the structure of the paper.
+
+** Suggested improvements **
+
+- Include an experiment to compare the proposed approach with non-quantum generative models
+
+Reply: Following the advice from the referee, we have implemented NADE-GEO and the comparison with TN-GEO and the other SOTA solvers are included throughout Tables I-V. As previously mentioned, the classical ML model is competitive in the sense that is comparable with models from about a decade ago. But still underperforming TN-GEO, which is on par with the state-of-the-art solvers for this application. This is also included in the new summary Table II in the main text, where it can be seen the superior performance of TN-GEO over NADE-GEO in this specific application.
+
+- It could be a nice addition to have experiments on another combinatorial optimization problem than portfolio optimization.
+
+- Related to the previous point, the authors may consider presenting their approach in a more general context rather than specifically for the portfolio optimization problem. As I understood it, the proposed approach can be applied to many different kind of optimization problems but the current presentation can suggest that the method is tailored specifically for portfolio optimization. A more general presentation of the method, as well as a clear explanation of the different kind of problems the approach can be applied to, could make for a more impactful paper (by reaching a wider audience). For example, can the approach be applied to any MIP?
+
+Reply: We agree with the reviewer that the current presentation hints to GEO as a solution tailored only for the portfolio optimization problem, where in reality, as the referee mentioned, GEO is more in the category of a generic solver. Although we had mentioned in a couple of places its black-box nature, many readers might not immediately associate this concept with a solver that can deal with any cost function, and therefore any combinatorial problem. This black-box feature is one of the most prominent ones which render our approach advantageous compared to other quantum heuristics, such as the quantum approximate optimization algorithm (QAOA), which relies on the cost function to be a polynomial in terms of the binary variables. To mitigate missing these essential points raised by the referee, the introduction has been significantly revised and restructured adding clarity to the main features of our framework.
+
+Although we did not explore explicitly in this work the case of problems containing both discrete and continuous variables (e.g., MIP) or only continuous variables , we discussed in the Outlook how an approach as the one in Ref. [3] could allow to generalize to these mixed variable cases by using such hybrid quantum-classical generative models within GEO. In a recent publication, Ref. [4], our team addressed this question of how to treat generative models with continuous variables (or a mixture of discrete and continuous) directly with TN models, which would be directly applicable to MIP and other related problems.
+
+To the question of including other applications, the main motivation for selecting this specific instantiation of portfolio optimization was the availability of concrete benchmarks and an extensive literature of solvers which have been fine tuned for over the past decades. In particular, we wanted to make sure we could make stronger claims in a well-established benchmark problem, and assess in this
+way where this new quantum-inspired solution TN-GEO fits within the suite of SOTA algorithms. Although this could be done for other well-known benchmarking problems, we felt the paper was extensive enough since it presents the introduction of this new framework plus demonstrating some of its highlights reflected in the three Results subsections: 1) the possibility to leverage data observed with other solvers (Sec. IIA), 2) its high performance in the regime of very limited calls to the cost function evaluation (Sec. IIB) and 3) its performance compared to SOTA algorithms under the rules established by those specific benchmarks (Sec. IIC). Although we considered extending to other application domains is outside of the scope of this work, other recent independent works have considered other real-world applications of GEO. For example, in Ref. [5], the authors considered an industrial case related to a floor planning NP-hard problem. In Ref. [6] the authors considered quantum-inspired generative models similar to the ones we propose here, to explore the solution space of candidates in molecular discovery. We hope the referee considers as well that the work is complete as is, given the new developments supporting the range of applications of our work.
+
+Revisions to the manuscript: We hope the new Introduction brings more clarity to the essential points raised by the referee. To emphasize the point from the referee that even MIP or problems with continuous variables could be addressed within GEO, we have added some comments in the Outlook about recent work extending and addressing this challenge, and how it fits within our framework.
+
+References:
+
+[1] Lopez-Piqueres, J., Chen, J. and Perdomo-Ortiz, A., Symmetric tensor networks for generative modeling and constrained combinatorial optimization. Machine Learning: Science and Technology. 4 035009 (2023)
+
+[2] Hibat-Allah, M., Mauri, M., Carrasquilla, J. and Perdomo-Ortiz, A., A Framework for Demonstrating Practical Quantum Advantage: Racing Quantum against Classical Generative Models. arXiv preprint arXiv:2303.15626 (2023).
+
+[3] Rudolph, M.S., Toussaint, N.B., Katabarwa, A., Johri, S., Peropadre, B. and Perdomo-Ortiz, A., Generation of high-resolution handwritten digits with an ion-trap quantum computer. Physical Review X, 12(3), p.031010 (2022).
+
+[4] Meiburg, A., Chen, J., Miller, J., Tihon, R., Rabusseau, R., Perdomo-Ortiz, A. Generative Learning of Continuous Data by Tensor Networks. arXiv preprint arXiv: 2310.20498 (2023).
+
+[5] Banner, W.P., Hadiashar, S.B., Mazur, G., Menke, T., Ziolkowski, M., Kennedy, K., Romero, J., Cao, Y., Grover, J.A. and Oliver, W.D., Quantum-Inspired Optimization for Industrial Scale Problems. arXiv preprint arXiv:2305.02179 (2023).
+
+[6] Moussa, C., Wang, H., Araya-Polo, M., Bäck, T. and Dunjko, V., 2023. Application of quantum-inspired generative models to small molecular datasets. arXiv preprint arXiv:2304.10867 (2023).
+REVIEWERS' COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+The authors have considered and reasonably addressed the comments of the referees, resulting a significant number of changes and improvements to the manuscript.
+Hence I am happy to recommend for publication.
+
+Reviewer #2 (Remarks to the Author):
+I am overall satisfied by the answers from the authors to my initial review and to the consequent revisions that have been made to the manuscript.
+
+I find that stylistically speaking, the quality of the introduction has decreased in some of the new paragraphs added in the revision (see detailed comments below). I also give below some relatively minor stylistic comments about the new content that has been added to the revision:
+
+p. 1, second item: "*either* from previous..." there is a missing "or" and second part of the sentence.
+
+p. 2 first item: This open -> This opens
+
+p. 2 The last item is stylistically very odd and reads more than a rebuttal to the first round of reviews than an actual paragraph from an introduction. More precisely, this list is presented as a list of "salient highlights" of the proposed approach; is "Nothing is special about the portfolio optimization problem" a "salient highlight" of the contribution. I understand the aim but this is done in a very clumsy way, in my opinion. I would suggest rewriting this item and rephrasing in the direction of emphasizing the "versatility" of the method, while justifying the focus on portfolio optimization.
+p.2 last item: "... *as* the NP hard problem *as* the workhorse..." is a bit stylistically clumsy
+
+p.2 last item: line starting with a comma after "Ref [8]"
+
+p.6 to be consistent, the item 10)'s sentence in the before last paragraph should end with "," instead of "'.
+
+p. 10 before last paragraph of first column: "it is to use it as" -> "is to use it as"
+
+p.10 "DATA AVAILABILITY": "have been deposited" -> e.g. "is available"
+
+p.14 sec 4: "introduced by Uria [37]" -> "introduced by Uria et al. [37]"
+"is that it models" -> "is to model"
+Reply to Reviewer #1
+
+We thank Reviewer #1 for their feedback throughout this review process and for their recommendation to publish this significantly revised version.
+
+Reply to Reviewer #2
+
+We thank Reviewer #2 for their recommendation to accept this latest version for publication. In this resubmission, we have incorporated all the stylistic changes suggested by the reviewer. We thank the reviewer for this valuable feedback throughout this review process.

00d0f482762f2f37431ca49a939480fc54bdd5eb053d5ac8ce0b474c9dacda22/preprint/preprint.md

@@ -0,0 +1,417 @@
+An integrin-targeting AAV developed using a novel computational rational design methodology presents improved targeting of the skeletal muscle and reduced liver tropism
+
+Ai Vu Hong
+avuhong@genethon.fr
+
+Genethon   https://orcid.org/0000-0002-0872-4295
+Laurence Suel
+Genethon
+Jérôme Poupiot
+Genethon
+Isabelle Richard
+Genethon
+
+Article
+
+Keywords:
+
+Posted Date: October 27th, 2023
+
+DOI: https://doi.org/10.21203/rs.3.rs-3466229/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+Additional Declarations: Yes there is potential Competing Interest. A.H.V. and I.R. are inventors on PCT application EP2023/065499 for the integration of RGDLxxL/I motif in AAV capsid for enhanced muscle transduction efficiency. I.R. is a part-time employee of Atamyo Therapeutics. The other authors declare no competing interests.
+
+Version of Record: A version of this preprint was published at Nature Communications on September 11th, 2024. See the published version at https://doi.org/10.1038/s41467-024-52002-4.
+Abstract
+
+Current adeno-associated virus (AAV) gene therapy using nature-derived AAVs is limited by non-optimal tissue targeting. In the treatment of muscular diseases (MD), high doses are therefore often required, but can lead to severe adverse effects. To lower treatment doses, we rationally designed an AAV that specifically targets skeletal muscle. We employed a novel computational design that integrated binding motifs of integrin alpha V beta 6 (αVβ6) into a liver-detargeting AAV capsid backbone to target the human αVβ6 complex – a selected AAV receptor for skeletal muscle. After sampling the low-energy capsid mutants, all in silico designed AAVs showed higher productivity compared to their parent. We confirmed in vitro that the enhanced transduction is due to the binding to the αVβ6 complex. Thanks to inclusion of αVβ6-binding motifs, the designed AAVs exhibited enhanced transduction efficacy in human differentiated myotubes as well as in murine skeletal muscles in vivo. One notable variant, LICA1, showed similar muscle transduction to other published myotropic AAVs, while being significantly more strongly liver-detargeted. We further examined the efficacy of LICA1, in comparison to AAV9, in delivering therapeutic transgenes in two mouse MD models at a low dose of 5E12 vg/kg. At this dose, AAV9 was suboptimal, while LICA1 transduced effectively and significantly better than AAV9 in all tested muscles. Consequently, LICA1 corrected the myopathology, restored global transcriptomic dysregulation, and improved muscle functionality. These results underline the potential of our design method for AAV engineering and demonstrate the relevance of the novel AAV variant for gene therapy treatment of MD.
+
+One Sentence Summary
+
+We developed a novel computationally AAV design method resulting in a new myotropic AAV, which allows low-dose AAV treatment for muscular dystrophies.
+
+INTRODUCTION
+
+Over 50 years since their discovery, adeno-associated viruses (AAVs) have shown great promise as an effective viral vector for gene delivery and gene therapy, leading to recent approval of therapeutic products \(^{1,2}\). Due to unmet medical needs and natural AAV tropism, many AAV-based gene therapies focus on treating muscle diseases (MD) \(^{3}\). Systemic treatment in such diseases aims to primarily target skeletal muscle, which accounts for more than 40% of body mass, and therefore often requires very high doses (\( \geq 1E14 \) vg/kg) to achieve meaningful therapeutic efficacy \(^{3-6}\). In addition, most recombinant AAVs built on natural-occurring variants lack specificity and often accumulate in the liver, with the concomitant risk of hepatotoxicity \(^{7}\). Other key challenges of rAAV use persist, including manufacturing, immunological barriers, and associated toxicity \(^{1,2,8,9}\).
+
+AAV is a small non-pathogenic single-stranded DNA parvovirus. Multiple open reading frames (ORFs) were identified in its genome, including *Rep*, *Cap*, *AAP* and *MAAP* \(^{1,10}\). The single *Cap* ORF expresses three capsid proteins - virion protein 1 (VP1), VP2 and VP3, which assemble into an icosahedral 60-mer capsid. Structurally, the VP3 monomer core contains a highly conserved eight-stranded β-barrel motif \(^{11}\).
+Inserted between the β-strands, nine surface-exposed variable regions (VR1-9) result in local topological differences between serotypes and dictate virus-host interaction. Consequently, genetically modifying VRs can drastically change the AAV, transduction, antigenic profile, and fitness \(^{10, 12, 13}\). VR4 and VR8, that cluster together spatially, forming the most prominent protrusion at the 3-fold axis, have been widely subjected to modifications, notably by inserting short peptides into the loop apices \(^{14}\). This resulted in some highly efficient capsid variants for transducing a variety of cell types and tissues \(^{1, 12}\). Among these, remarkably, AAVMYOs \(^{15, 16}\) and MYOAAVs \(^{17}\) transduce skeletal muscles, deliver therapeutic transgenes efficiently, and were shown to correct dystrophic phenotypes in MD mouse models at relatively low doses (2E12 – 1E13 vg/kg).
+
+Importantly, the myotropic AAVs \(^{15-17}\) identified by muscle-directed high-throughput screening (HTS) were shown to share an Arg-Gly-Asp (RGD) motif, presumably targeting the integrin complex \(^{17-20}\). Integrins are a group of heterodimeric proteins composed of an α- and a β subunit that serve various cellular functions, including cell adhesion, cell migration, and cell signaling \(^{21}\). As adhesion molecules, integrins also mediate cell-pathogen interactions, and are therefore exploited by many viruses, including natural AAV, to infect cells \(^{22-24}\). Indeed, many of these viruses use an RGD motif on their viral envelope glycoproteins or capsids for cell attachment, endocytosis, entry, and endosomal escape \(^{18, 22, 25}\). The discovery that RGD-dependent integrin-targeting AAV variants can acquire myotropism therefore represents a novel potential candidate approach for a rational design to target skeletal muscle.
+
+This study introduces a novel computational method for a rational AAV design targeting skeletal muscle, which resulted in a novel myotropic vector for MD gene therapy. First, the human skeletal muscle-enriched integrin complex alpha V beta 6 (αVβ6) was selected as the target receptor. Inspired by one-sided protein design \(^{26, 27}\), we computationally designed a previously developed liver-detargeting hybrid capsid between AAV9 and AAVrh74 (Cap9rh74) as an αVβ6 binder. The VR4 loop was completely modified, in which new sequences were iteratively selected to simultaneously optimize for free energy, while hosting αVβ6-binding RGDLXXL/I motifs. All designed AAVs were well-produced, at higher titers than their parent. The designed AAVs were confirmed to require αVβ6 binding for cellular transduction. The most promising variant, renamed LICA1, was selected for further analysis and showed superior transduction in human differentiated myotubes and strong myotropism in several mouse models. We evaluated this variant by delivering therapeutic transgenes in two MD mouse models at a very low dose of 5E12 vg/kg, in comparison to AAV9. In both cases, LICA1 presents higher efficacy than AAV9 in correcting dystrophic phenotypes, global transcriptomic changes and restoring muscle function, thanks to improved transduction and transgene expression in skeletal muscles. Collectively, the study provides a proof-of-concept for a new rational AAV design pipeline leveraging protein design tools, which resulted in a novel myotropic AAV with high potential for gene therapy for muscle diseases.
+
+RESULTS
+
+1. Selection of the cellular receptor for rational design
+Several myotrophic AAVs have recently been developed, notably, the insertion into the AAV9 VR-VIII loop of P1 peptide (RGDLGLS) \(^{15,16}\), and a series of RGD-containing sequences identified by directed evolution \(^{17}\). Importantly, these modified capsids shared a common RGD motif, which suggested their affinity to integrin (ITG), cell-surface heterocomplexes that interact with the extracellular matrix \(^{28}\). Using publicly available datasets, we aimed to select relevant integrin subunits for a subsequent rational AAV design targeting skeletal muscle.
+
+Chemello and colleagues previously performed single-nucleus RNA sequencing, comparing gene expression of all cell types in the skeletal muscle of wild-type (WT) and Duchenne muscular dystrophy mouse models (D51) \(^{29}\). We extracted RNA levels of all integrin alpha and beta genes from these data (**Figure S1A**). Among all subunits, only the α- subunits *Itgav*, *Itga7* and the β-subunits *Itgb6*, *Itgb1*, and *Itgb5* show relatively high expression in the myogenic nuclei. Of interest is the fact that the expression level of *Itgb6* is highly enriched in myonuclei, and significantly upregulated in the dystrophic condition, whereas *Itgb1* and *Itgb5* expression are ubiquitous in all cell types, and significantly lower than the *Itgb6* level in all myonuclei. Among the two expressed α-subunits, only *Itgav* was known to associate with *Itgb6* to form αvβ6 heterocomplexes – a member of the RGD-binding integrin family \(^{30}\). Furthermore, bulk RNA sequencing data from multiple human tissues confirmed high expression of *Itgav* and *Itgb6* in skeletal muscle, and low expression of *Itgb6* in the liver and spleen, two preferred targets of natural AAV (**Figure S1B**, GTEx V8, dbGaP Accession phs000424.v8.p2). We therefore hypothesize that AAV transduction in skeletal muscle can be improved by rationally designing an AAV capsid that specifically binds to αvβ6.
+
+**2. Rational design of a hybrid capsid, Cap9rh74, with a high affinity to the αVβ6 complex**
+
+As we aim to specifically target the skeletal muscle, we selected a hybrid capsid that we previously developed and that has a liver-detargeting property as the parental capsid in our design (Patent Number: EP18305399.0). This hybrid capsid of AAV9 and AAV.rh74 (AAV9rh74) was constructed by replacing the AAV9 sequence of VR4 to VR8 with that of AAV-rh74. The hybrid capsid showed similar infectivity in skeletal and cardiac muscles but was strongly de-targeted from the liver. The latter property is of particular interest in skeletal muscle gene transfer since the majority of administrated viral vector will not accumulate in the liver, as is the case for natural AAVs \(^{31,32}\).
+
+After selection of the cellular receptor of interest and capsid backbone, AAV capsids were computationally engineered (**Fig. 1A**). First, the 3D structure of the parental capsid, of with structure was unknown, was modeled using AlphaFold2 \(^{33,34}\). The structural prediction of the Cap9rh74 aa 219–737 monomer performed using AlphaFold2 was at a high level of confidence, with predicted local distance difference test (IDDT-Cα), a per-residue measure of local confidence, of 97.04 and low predicted aligned error (PEA) of 4.32 (**Fig S1C-D**). This structure is thus suitable for the next steps in the design.
+
+Second, we extracted the 3D structure or sequences of binding motifs of the human integrin complex from PDB. Importantly, αVβ6 was previously shown to bind with high affinity to the RGDLXXL/I motif
+found in the human TGF-β1 and TGF-β3 prodromains \(^{35, 36}\). Binding peptides with eight amino acid residues, aa214-221 in TGF-β1 (PDB: 5ffo) and aa240-247 in TGF-β3 (PDB: 4um9), were isolated from the corresponding crystal structures before grafting into the Cap9rh74 VR4 loop. Both motifs bind to αVβ6 dimer at a very similar position (**Fig S1E**).
+
+Third, the defined binding motifs were then grafted into the VR4 loop (residues 453–459) of the capsid protein based on the RosettaRemodel protocol \(^{37}\). In the grafting-remodel process, many rounds of backbone optimization and sequence design iteratively search for low-energy sequence–structure pairs (**Fig. 1B**). The lowest-energy designs in grafting experiments of each TGF-β motif showed convergence in both structure and sequence (**Fig. 1C-D, S1F-G**). The new VR4 loops include the binding peptide and two flanking 2-amino acid linkers and retain the LXXL/I motif as an α-helix, which is important to bind in the β6 subunit’s pocket \(^{36}\).
+
+Retrospective docking simulations of the two AAV_ITGs with the best scores, namely Cap9rh74_5ffo and Cap9rh74_4um9, on the αVβ6 complex showed highly similar binding positions of the new VR4 loop to its corresponding inserted motifs (**Fig. 1E-F**). This suggests that the new capsids can bind to αVβ6 thanks to VR4-included RGDLXXL/I motif. Sequences with the best scores, which reflect the thermodynamic stability of one static protein conformation \(^{38}\), were subjected to experimental validation.
+
+**3. All designed AAV_ITGs showed higher productivity and enhanced cellular transduction via αVβ6 binding.**
+
+The two AAVs with the best design were then tested for productivity and the effectiveness of using αVβ6 as a cellular receptor. They were produced by tri-transfection with pLTR-CMV-GFP-Luciferase as the expression cassette. Thanks to energy optimization, all the designed AAV-ITG variants significantly increase their titers compared to their parental hybrid capsid, to levels similar to those for AAV9 (**Fig. 2A, S2A**). In addition, all modified AAV-ITG variants retain proportions of VP1, VP2, VP3 capsid proteins with a similar ratio of AAV9 (**Fig. 2B**). This suggests that the designed sequences result in more stable AAV capsid complexes thanks to their estimated low energy structure, and therefore better production efficacy.
+
+Next, we examined whether these AAV-ITGs can effectively use αVβ6 as a cellular receptor upon infection. First, a HEK293 cell line (293_αVβ6) constitutively overexpressing both integrin subunits, αV and β6, was created using the PiggyBac system (**Fig S2B-C**). The designed AAVs were then tested for their infectivity in this cell line. As expected, infection of AAV_ITGs in 293_αVβ6 cells, as defined by vector copy numbers (VCN), was higher than for AAV9 and AAV9rh74 (**Fig. 2C**). Both AAV_ITGs dramatically improved the luciferase activity (\(FC_{9rh74_4um9/AAV9}=60.50\), \(FC_{9rh74_5ffo/AAV9}=25.99\), \(FC_{9rh74_4um9/9rh74}=63.99\), and \(FC_{9rh74_4um9/9rh74}=27.49\), **Fig. 2D**). To investigate how specific AAV_ITGs used αVβ6 as a cellular receptor, we tested their infectivity under binding competition conditions. The number of AAV_ITG viral vectors entering the cells was significantly reduced when blocked by the recombinant protein αVβ6 before viral infection, but no change occurred with AAV9 or AAV9rh74
+(Fig. 2E). This result suggests that efficient transduction of AAV_ITGs requires specific binding to a αVβ6 complex.
+
+During myogenesis, αVβ6 is only expressed in late differentiation, but not in the myoblast stage (Fig S1A, S2D). We therefore hypothesized an enhanced transduction of AAV_ITGs in differentiated myotubes, but not myoblasts. We infected both human myoblasts and myotubes with AAV_ITGs. Low levels of luciferase activity were observed in all AAVs tested in human myoblasts (Fig. 2G,I). On the other hand, in human differentiated myotubes (hMT), VCN and luciferase activities in both AAV9rh74_4um9 and _5ff0 were significantly higher than for AAV9 or AAV9rh74 (Fig. 2F,H,K). In particular, variant AAV9rh74_4um9 showed a 16.56 (p < 0.0001) and 25.02-fold (p < 0.0001) improvement in luciferase activity compared to AAV9 and AAV9rh74, respectively, which is in agreement with its superior transduction efficiency and transgene expression seen in 293_αVβ6 cells.
+
+In summary, the two designed AAV_ITGs were both well-produced and function via αVβ6-specific binding, thus enhancing their transduction efficiency in 293_αVβ6 and human differentiated myotubes.
+
+4. AAV_ITGs enhanced transduction in skeletal muscle following systematic administration
+
+AAV_ITGs, together with AAV9 and AAV9rh74, were administrated systematically via intravenous injection (transgene: CMV_GFP-Luciferase, dose: 1E13 vg/kg, age at injection: 6 weeks, n = 4) in C57Bl6 mice to examine their biodistribution 3 weeks post-injection (Fig. 3A).
+
+In agreement with a previous study, AAV9rh74 slightly reduces transduction in skeletal muscle compared to AAV9 but accumulates much less in the liver (Fig. 3B-D). Thanks to the liver-detargeting capsid and in accordance with the fact that αVβ6 is weakly expressed in the liver, we expected poor entry into the liver for designed AAV_ITGs. Indeed, AAV_ITGs is strongly detargeted from the liver, both at VCN and mRNA levels, even further than the parental capsid (Fig. 3C-D). In contrast, enhanced transduction was observed in all skeletal muscles that were tested, including the tibialis anterior (TA), quadriceps (Qua) and diaphragm (Dia) (Fig. 3B-D). The two AAV_ITGs both showed a substantial increase in VCN and luciferase activity compared to both AAV9 and AAV9rh74. Similar to the results obtained in *in vitro* models, AAV9rh74_4um9 is the best transducer among the two AAV_ITGs. Compared to AAV9, the variant 9rh74_4um9 significantly increased VCN 5.31/7.21/2.48-fold and increased luciferase activity 15.2/13.2/23.57-fold in Qua, TA, and Dia (p < 0.05), respectively. Compared to the original backbone AAV9rh74, this variant even magnified the difference by increasing VCN 5.53/2.85/7.69-fold and increasing luciferase activity 152.35/106.68/60.43-fold (p < 0.05). Furthermore, AAV9rh74_4um9, but not AAV9rh74_5ff0, significantly increased transduction in the heart (FC\(_{VCN}\)=4.15, FC\(_{LUC}\)=15.43, p < 0.05). All AAVs that were tested showed poor delivery and transgene expression in the lungs and kidneys. No alteration of TGFβ and integrin signaling was observed at one-month post-injection in all AAVs being tested (Fig S2F-G). Overall, these data indicate that AAV_ITGs, especially the 9rh74_4um9 variant, are strongly liver-detargeted and exhibit enhanced tropism towards skeletal and cardiac muscles.
+5. AAV9rh74_4um9 transduced skeletal muscle similarly, but detargeted the liver more strongly than other myotropic AAVs
+
+Several engineered myotropic AAVs (mAAVs), including AAVMYO \(^{15}\), MYOAAV-1A and – 2A \(^{17}\), have demonstrated superior efficacy for *in vivo* delivery of muscle compared to natural AAVs. To evaluate the properties of these AAVs compared to ours, we performed *in vitro* and *in vivo* experiments. Viral preparations were produced using the same reporter transgene (CMV_GFP-Luc). All mAAVs were well-produced in 400ml suspension, with higher titers than AAV9rh74. However, MYOAAV productivity was significantly lower than 9rh74_ITGs and MYOAAVs (**Fig S3A**). Since all investigated mAAVs shared a common integrin-targeting RGD motif, these AAVs were then evaluated for their transduction via integrin complexes in myotubes and in cell lines where integrin complexes were stably overexpressed by the PiggyBac system. In 293_aVβ6 cells as well as in hMT, where αVβ6 is highly expressed, AAV9rh74_4um9 showed the highest transduction among the tested myotropic AAVs, with the sole exception that luciferase activity of MYOAAV2A was higher in hMT (**Fig S3B-C**). We also tested AAV transduction efficiency in two other cell lines, 293_WT, where αVβ6 expression is low, and 293_a7β1 that stably overexpresses a non-RGD-targeting α7β1 integrin. In both conditions, MYOAAV2A and AAV9rh74_4um9 showed the highest transduction (**Fig S3D-E**). These results suggest that, as intended with the rational design, AAV9rh74_4um9 uses αVβ6 more preferentially for cellular transduction than others, yet it can also efficiently use other integrin(s) similar to MYOAAV2A.
+
+Following *in vivo* injection in the same setting as described above (6-week-old WT mice, dose: 1E13 vg/kg, n = 4), the three mAAVs and 9rh74_4um9 all showed strong liver-detargeting, high enrichment in both skeletal and cardiac muscles, and negligible transduction levels in other organs that were tested (kidneys, lungs, and brain) (**Fig. 3G-H**). No significant difference was observed in either VCN or luciferase activity between all three mAAVs and 9rh74_4um9 in the skeletal muscles that were tested. In heart muscle, MYOAAV2A showed a significant increase in VCN compared to other myotropic vectors, but no difference in luciferase activity, in agreement with the original observation \(^{17}\). The most striking difference is the level of liver-detargeting between these vectors. The VCN for 9rh74_4um9 in liver is 3.34/22.05/13.85 times lower than for AAVMYO (p = 0.0022), MYOAAV-1A (p = 0.0013) and – 2A (p = 0.033), respectively (**Fig. 3G**), and is therefore the only vector that accumulates less in liver than skeletal muscles (**Fig S3F-G**). These data indicate higher muscle specificity for the 9rh74_4um9 variant compared to other myotropic vectors that have been investigated to date.
+
+In summary, the 9rh74_4um9 variant, hereafter referred to as LICA1 (linked-integrin-complex AAV), consistently showed enhanced transduction and strongest liver-detargeting. Therefore, we then attempted to evaluate LICA1 as a delivery vector for muscular dystrophies, in comparison with AAV9. Two different setups will be investigated: the transfer of microdystrophin (\(μ\)Dys) – an incomplete transgene - in mdx, a mild mouse model of Duchenne muscular dystrophy (DMD) and of the full-length human α-sarcoglycan (SGCA) in a severe mouse model of limb-girdle muscular dystrophy R3 (LGMD-R3).
+6. Low-dose LICA1-μDys gene transfer is effective in specifically overexpressing microdystrophin in dystrophic muscle but not sufficient to fully correct the underlying pathology
+
+DMD is caused by mutations in the DMD gene, which encodes for dystrophin protein – a key player in the dystrophin-glycoprotein complex (DGC), which is critical for the structural stability of skeletal muscle fibers \( ^{39} \). Lack of dystrophin can result in progressive loss of muscle function, respiratory defects, and cardiomyopathy. The most commonly used DMD animal model is the mdx mouse, with a lifespan reduced by 25%, milder clinical symptoms than those seen in human patients, with the exception of the diaphragm muscle \( ^{40} \). Among many therapeutic strategies to restore dystrophin expression, high-dose AAV-based gene transfer of shortened functional forms of the dystrophin ORF provided excellent results in animal models, but unsatisfactory conflicting data in current clinical trials \( ^6 \). Severe toxicities, even patient death, have been reported from these trials (NCT03368742, NCT04281485), assumed to be related to the dose of \( \geq 1E14 \) vg/kg. We therefore explored the possibility of low-dose μDys gene transfer \( ^{41} \) in mdx mice using LICA1 in comparison to AAV9 (**Fig S4A**, age at injection: 4 weeks, dose: 5E12 vg/kg, treatment duration: 4 weeks, n = 5). Three muscles with increasing levels of severity – TA, Qua, and Dia – were used to study AAV transduction and treatment efficacy.
+
+LICA1 showed better μDys gene transfer than AAV9 in this model. LICA1-treated mice exhibited a significantly higher VCN in all 3 muscles that were tested, 1.85/2.02/1.07 times higher in TA (p < 0.0001), Qua (p < 0.0001), and Dia (p = 0.020), respectively (**Fig. 4A**). RNA levels indicated even greater differences and were 4.56–7.57 times higher in the LICA1-treated group (**Fig. 4B**; TA: FC = 4.56, p < 0.0001; Qua: FC = 5.46, p = 0.0001; Dia: 7.57, p = 0.05). Consequently, LICA1 can transduce almost 100% in TA and Qua, and 49.98% in Dia, while substantially lower numbers were seen in AAV9-treated muscles, at 73.22% (p = 0.0001), 57.8% (p < 0.0001), 10.34% (p < 0.0001) in TA, Qua, Dia, respectively (**Fig. 4C**, **Fig S4B**).
+Furthermore, while infection levels and expression of the transgene in liver were high for the AAV9 vector (despite the use of muscle-specific promoter), the VCN and mRNA levels in LICA1-treated liver were extremely low (**Fig. 4A-B**, \( FC_{VCN-AAV9/LICA1}=36.8, \ p = 0.0002; \ FC_{mRNA-AAV9/LICA1}=64.7, \ p < 0.0001 \)). These data again confirmed the transduction efficiency and specificity towards skeletal muscle for the LICA1 vector, even with low-dose treatment.
+
+The histological features and muscle functionality after AAV treatment were restored accordingly. The centronucleation index (percentage of centronucleated fibers) – an indicator of the regeneration/degeneration process – did not change with AAV9 (except in TA) but was significantly reduced upon LICA1 treatment (reduction of 21.68%, 19.05%, 22.88% in TA, Qua, Dia, respectively) (**Fig. 4D**, **Fig S4C**). Similarly, the fibrosis level in two severely affected muscles, Qua and Dia, only exhibited a significant reduction with LICA1, but not AAV9 (**Fig. 4E**, **Fig S4D**). The serum biomarker MYOM3 level, an indicator of muscle damage \( ^{42} \), showed a reduction for both AAV treatments, with a considerable further reduction seen in the LICA1-treated group (**Fig. 4F**, \( FC_{AAV9/KO}=0.75, \ FC_{LICA/KO}=0.43, \ p_{AAV9-LICA}>0.0001 \)). More importantly, AAV9 treatment did not affect any muscle functionality being tested (**Fig. 4G-I**), while significant improvements with LICA1-μDys treatment were observed in escape
+test – a measure of global force (Fig. 4G, \( FC_{LICA1/mdx}=1.19, p_{LICA1/mdx}=0.02 \)) and *in situ* TA mechanical force measurement (Fig. 4H, \( FC_{LICA1/mdx}=1.14, P_{LICA1/mdx}=0.0006 \)). However, none of the treatment normalized to the WT functional levels. These data indicate that LICA1 is better than AAV9 at restoring dystrophic histological features and muscle functions.
+
+We also investigated the molecular alteration in Qua upon AAV treatment using RNA-seq. On the two first principal components (PCs) of the PCA, a clear distinction between four transcriptome groups (WT, mdx, AAV9, LICA1) was observed, while LICA1-treated muscles were clustered closer to the WTs than others (**Fig S4E**). To our surprise, despite excellent transgene expression by LICA1, global transcriptomic restoration was relatively modest (**Fig. 4K**). Nevertheless, a substantial improvement can still be seen for LICA1 compared to AAV9. Among 4216 down- and 4501 upregulated differentially expressed genes (DEGs) identified in mdx muscle, 1515 (35.9%) and 1728 (38.4%) were restored by AAV9, while LICA1 was able to correct 1736 (41.2%) and 1980 (44.0%), respectively (**Fig. 4L-M**). In addition, a greater number of genes were either not or insufficiently corrected by AAV9 than by LICA1 (**Fig. 4N**). A total of 2572 genes were downregulated (61.0%) and 2620 (58.2%) incompletely restored, while significantly lower numbers were seen for LICA, with 2094 (49.67%) down- and 2019 (44.86%) upregulated. Interestingly, some known dysregulated pathways, including α- and Y-interferon responses and oxidative phosphorylation, were significantly better normalized by LICA1 than by AAV9 (**Fig S4F**).
+
+In summary, at 5E12 vg/kg, LICA1-μDys, but not AAV9, was efficient in transducing close to 100% myofibers, except in the diaphragm. This effective improvement in transduction can significantly reduce some dystrophic features in all muscles that were tested, yet restoration in the global transcriptome remains modest. However, greater improvements in functional, histological, and transcriptomic restoration were achieved with LICA1 compared to AAV9.
+
+**7. Low-dose LICA1-SGCA treatment restored the muscle functionality, dystrophic phenotypes, and transcriptomic dysregulation in a severe SGCA mouse model.**
+
+LGMDR3 is caused by mutations in the SGCA gene \(^{43}\) – another component of the DGC complex. Defects in the SGCA protein therefore lead to muscle weakness and wasting. A LGMDR3 mouse model has been established, which closely represents patient’s clinical phenotypes \(^{44}\). Similar to the setting in mdx mice, low-dose AAV treatment with 5E12 vg/kg was investigated in this mouse model. AAV9 or LICA1 encoding human SGCA (hSGCA) under control of a muscle-specific human Acta1 promoter were injected into 4-week-old SGCA-KO mice (**Fig. 5A**). Analysis was performed 4 weeks post-treatment.
+
+In all three muscles that were tested, TA, Qua, Dia (in order of increasing severity), transduction in various measures, VCN, mRNA level, and percentage of SGCA + myofibers, was significantly greater in the LICA1-treated group than for AAV9 (**Fig. 5B-D**, **Fig S5A**). Of note is the fact that the differences in transduction efficacy (%SGCA + myofibers) between LICA1 and AAV9 are greater in more severely affected muscles (**Fig. 5D**). At such a low dose, AAV9 was able to transduce > 80% myofibers in TA while LICA1 can reach close to 100% (\( p < 0.0001 \)). While LICA1 still transduced almost 100% of fibers in Qua (the muscle
+affected with intermediate severity), only 58.1% fibers were transduced by AAV9 on average (p < 0.0001). In the most severely affected muscle, Dia, both vectors displayed reduced efficiency; however, LICA1 continued to demonstrate much better transduction (\( \mu_{AAV9} = 22.1\%, \mu_{LICA1} = 59.5\%, p < 0.0001 \)).
+
+The differences in transgene delivery and expression positively correlated with levels of histological and functional restoration. Different dystrophic histological features, including percentage of centronucleated fibers (**Fig. 5E**, **Fig S5B**), percentage of fibrosis area (**Fig. 5F**, **Fig S5C**), and fiber size distribution (**Fig. 5G**), were all significantly better normalized by LICA1 than AAV9, especially in more severely affected muscles. Importantly, no significant improvement was observed in the AAV9-treated group in centronucleation index and fibrosis level in Dia, while LICA1 reduced these parameters by half (**Fig. 5E-F**). Fiber sizes were also restored to near-WT distribution by LICA1 in this muscle (**Fig. 5G**). No difference in body weight was seen between groups with or without AAV treatment (**Fig S5D**). At the functional level, however, the escape test – a measure of global force - showed a significant increase in AAV9-treated mice (FC = 1.42, p = 0.0072) and was even higher in LICA1-treated group (FC = 1.72, p < 0.0001) (**Fig. 5H**). On the other hand, *in situ* TA mechanical forces were both improved in the two AAV groups at similar levels (**Fig. 5I**), possibly due to > 80% transduction rate by both vectors. Similar to the global force, the serum MYOM3 level was greatly reduced in the LICA1-treated group but not for AAV9, indicating less muscle damage (**Fig. 5K**). No difference was seen in the anti-capsid antibody between the two AAV treatments (**Fig S5E**). These results indicate that better and significant functional and histological restoration in the LICA1-treated mice was achieved, even at low-dose treatment, thanks to superior transduction efficacy.
+
+We further investigated the molecular alterations following AAV treatment by transcriptomic profiling of the quadriceps muscle. The first principal component (PCs) of the PCA was able to separate a group including WT and LICA1 with a group including SGCA-KO and AAV9, suggesting close proximity between elements within these 2 groups (**Fig S5F**). A heatmap of all 8591 significant DEGs (4035 downregulated and 4556 upregulated) further highlighted the restorative effect of LICA1 on gene expression levels (**Fig. 5L**). LICA1-treated muscles, in particular, demonstrated a significant correction of 69.9% (2821/4035) and 66.5% (3028/4556) of down- and upregulated DEGs, respectively, compared to 12.4% (500/4035) and 9.21% (420/4556) corrected by AAV9 treatment (**Fig. 5M-N**). Conversely, not all DEGs were significantly restored or returned to WT levels. The number of such transcripts in AAV9-treated muscles was much higher than in the LICA1-treated group (**Fig. 5O**): 2541 (63.0%) downregulated DEGs and 3045 (66.8%) upregulated DEGs for AAV9, with only 483 (12.0%) downregulated DEGs and 1038 (22.8%) upregulated DEGs in the LICA1-treated group. These data illustrate that low-dose LICA1 treatment can effectively normalize the majority of the dysregulated transcriptome and is much more efficient in correcting gene expression dysregulation than AAV9 at the same dose.
+
+In summary, low-dose (5E12 vg/kg) AAV gene transfer using LICA1 in the LGMDR3 mouse model is effective in restoring muscle function, dystrophic histology, and the dysregulated transcriptome. The efficacy was much greater than for AAV9 at the same dose due to enhanced transduction.
+DISCUSSION
+
+Given the severe complications observed with very high dose AAV treatment, lowering the dose by increasing vector specificity via capsid modification is one way to address these issues. This study investigated the possibility of altering AAV tropism towards skeletal muscle by targeting integrin. We designed an AAV as a αVβ6 binder, which resulted in a novel myotropic AAV variant, namely LICA1. LICA1 showed greatly enhanced transduction in skeletal muscle in WT and two MD mouse models. Consequently, by improving the delivery of therapeutic transgenes (hSGCA and μDys) in these MD mouse models, LICA1 was able to correct dystrophic phenotypes, global transcriptional dysregulations and significantly restore muscle function.
+
+AAV capsid sequence design method that ensures high AAV production
+
+AAV tropism is commonly altered by inserting a small peptide into the VR4 or VR8 loop without any sequence constraints. Since no consideration regarding AAV capsid stability is included in this method, the resulting AAV can suffer from instability, reduced productivity, and increased AAV genome fragmentation \(^{17,45}\)(ASGCT 2023). In the current study, a physics-based protein sequence design method was used to graft the binding motifs from TGFβ-1 and – 3 into the VR4 loop of the hybrid capsid AAV9rh74. The major differences to the classical peptide insertion method are that the entire VR4 loop was modified to include a new binding motif and the amino acids around this motif (linkers) were selected to minimize the potential energy. Low-energy sequences ensure the stability and intended folding of the designed proteins, presumably leading to improved stability of the AAV particle \(^{38}\). Six AAVs designed using this method were tested experimentally and all showed better productivity than their parent, Cap9rh74, and similar levels to well-produced AAV9. This suggests that low Rosetta energy correlates with high stability of capsid protein, and thereby high AAV production.
+
+Integrin αVβ6 as a myotropic AAV receptor for skeletal muscle
+
+Virus-host interaction is the foundation for improved viral vectors, yet skeletal muscle receptors that allow effective AAV transduction are poorly defined. However, top hits from two independent studies with different screening schemes identified myotropic AAVs with a common RGD motif, \(^{15,17,19}\). In addition, it has previously been described that integrin functions as cellular receptor for natural AAV \(^{23,24}\). Coincident with our screening for possible integrin receptor, only αVβ6 is highly expressed and enriched in skeletal muscle (**Fig S1**). By including αVβ6 binding motifs, AAV_ITGs efficiently utilized αVβ6 for cellular infection. Enhanced transduction was observed in conditions with high (either ectopic or natural) αVβ6 expression, including human differentiated myotubes and murine skeletal muscles of WT and two other MD mouse models. In most cases, the improved transduction was evident at the VCN level, indicating better cell entry via αVβ6 binding.
+
+In addition, we conducted a study comparing LICA1 and three other published myotropic AAVs. No significant differences in skeletal muscle transduction were observed on either VCN or transgene expression levels. However, the liver infection rate was significantly lower with LICA1 compared to the
+other mAAVs, presumably due to the use of a liver-detargeted backbone and the low expression level of αVβ6 in liver. As a result, the LICA1 vector exhibited the highest muscle/liver transduction ratio among all AAVs tested, suggesting increased specificity towards skeletal muscle. This finding highlights the importance of selecting an appropriate targeting receptor for rational design and further supports αVβ6 as a promising candidate for targeting skeletal muscle.
+
+LICA1 is a potential vector for muscular diseases
+
+AAV gene therapy in muscle diseases typically requires very high doses (≥ 1E14 vg/kg) for functional benefits \(^{41, 46}\), yet can result in severe and even fatal adverse events \(^{7}\). In this study, we explored low dose (5E12 vg/kg) treatment using the LICA1 vector in two MD mouse models, DMD and LGMDR3. Of note is that this dose is at least 20 times lower than the doses currently used in clinical trials for neuromuscular diseases \(^{3}\). In both models, LICA1 was significantly better than AAV9 in delivering and expressing therapeutic transgenes, consequently restoring better histological dystrophic phenotypes. In TA and Qua, LICA1 was able to transduce more than 80% of fibers. It was still a challenge to effectively transduce diaphragm muscle at this dose, yet more than 50% of Dia fibers were positive for transgene expression with LICA1 in both models while AAV9 transduced very poorly. This improvement in transgene expression translates directly into improved histological restoration, including centronucleation index and fibrosis level. In particular, with only more than 50% successfully transduced fibers, LICA1 was able to reduce diaphragm fibrosis by 42.8–47.0% (mdx and SGCA\(^{-/-}\) models respectively), whereas no change was seen in AAV9-treated groups. The biomarker for muscle damage level, MYOM3, was reduced by 57.5–67.2% (mdx and SGCA\(^{-/-}\) models respectively) by LICA1 and significantly greater than AAV9. Similarly, global muscle force was significantly restored to a higher level with LICA1 than with AAV9 in SGCA-KO mice. These data confirmed superior muscle transduction by LICA1 and resulting therapeutic benefits were obtained even at low-dose treatment in two MD models.
+
+However, treatment efficacy varies between two disease models at molecular levels. We profiled transcriptomic changes in Qua following AAV treatment in both MD models. Despite similar transduction efficiency of LICA1 in the two models, restoration of dystrophic transcriptional changes in SGCA-KO was significantly greater. It is noteworthy that μDys is an incomplete form of dystrophin. The μDys used in the present study lacks several functional domains, including multiple spectrin-like repeats that bind to nNOS, F-actin, sarcomeric lipid and microtubules, and a dystrobrevin- and syntrophin-binding C-terminus \(^{41}\). This might explain the inadequate efficacy in restoring global gene expression in μDys gene therapy trials, in spite of highly effective gene transfer. Similarly, despite excellent functional restoration by microdystrophin gene transfer in various animal models, outcomes from these clinical trials are unsatisfactory \(^{6}\). Therefore, careful assessment of molecular restoration should be included for evaluating gene therapy efficacy.
+
+In summary, this study presents an alternative computational method that aids rational AAV design and ensures high-production AAV variants. The proof-of-concept design targeting skeletal muscle resulted in a high-productivity myotropic AAV, thereby effectively delivering therapeutic transgenes and restoring
+dystrophic phenotypes in two MD mouse models at a low dose. This work contributes to the ongoing efforts to reduce AAV treatment doses and further advance AAV engineering, paving the way for more effective and accessible gene therapies in the future.
+
+MATERIALS AND METHODS
+
+Study Design
+
+The primary objective of the study was to design a novel myotropic AAV capsid with a high production yield by using a computationally rational design. The secondary aim was to investigate the possibility of low-dose AAV treatment using a designed AAV in animal models of muscular dystrophies, which typically require an alarmingly high dose (≥ 1E14 vg/kg). We used publicly available datasets to identify possible receptors for skeletal muscle and protein design tools to engineer AAV capsid protein. Resulting variants were characterized for their productivity and transduction efficiency in various *in vitro* cell lines and multiple mouse models. Experiments were performed at least three times, unless noted otherwise. The AAV injection and infection experiments were conducted in a nonblinded fashion. The blinding approach was used during dissection, histological validation, immunostaining analysis, *in vivo* functional tests, and biomarker analysis. No data were excluded. Details on experimental procedures are presented in Supplementary Materials and Methods.
+
+Animal care and use
+
+All animals were handled according to French and European guidelines for human care and the use of experimental animals. All procedures on animals were approved by the local ethics committee and the regulatory affairs of the French Ministry of Research (MESRI) under the numbers 2018-024-B #19736, 2022-004 #35896. C57Bl/6, B6Ros.Cg-Dmdmdx-4Cv/J mice were obtained from the Jackson Laboratory. A knockout mouse model of *α*-sarcoglycan was obtained from the Kevin Campbell laboratory (University of Iowa, USA) 44. Mice were housed in a SPF barrier facility with 12-h light, 12-h dark cycles, and were provided with food and water *ad libitum*. Only male mice were used in the present study. Well-being and weights of the animals were monitored for the duration of the study. The animals were anesthetized with a mix of ketamine (100 mg/kg) and xylazine (10 mg/kg), or with isoflurane (4%) for blood samples. For AAV intravenous injections, a maximum volume of 150 μl containing AAV vectors was injected via the sinus route after the animals had been anesthetized with isoflurane. The AAV intravenous doses used in the present study were 5E12 or 1E13 vg/kg.
+
+Cell culture and in vitro study
+
+Adherent HEK293-T cells were maintained in the proliferating medium containing DMEM (Thermo Fisher Scientific), supplied with 10% fetal bovine serum and 1X gentamycin at 37°C, 5% CO2. Human immortalized myoblasts (AB1190 cell line) were maintained in Skeletal Muscle Cell Growth Medium (PromoCell, C23060) and differentiated in Skeletal Muscle Differentiation Medium (PromoCell, C23061).
+In vitro AAV infection was performed by directly adding AAV into culture medium at the dose of 1E9 or 1E10 vg per 24-well plate well. After 48h post-infection, cells were washed and subjected to VCN and gene expression analysis.
+
+To inhibit AAV infection, AAVs were incubated with recombinant hITGAV-hITGB6 protein (Bio-Techne, 3817-AV-050) at 37°C for 30 minutes, at a concentration of 1μg protein per 5E9vg AAV before addition to the cells (1E4 vg per cell). The same condition treated with recombinant hSGCA protein served as a control for the comparison.
+
+Statistical Analysis
+
+Results are presented as mean ± SEM, unless noted otherwise. Significance of differences in multiple pairwise comparisons of more than two groups was determined by one-way ANOVA. The significance of differences in pairwise comparisons of multiple groups with multiple treatments was determined by two-way ANOVA. To account for multiple testing and control the false discovery rate (FDR) across the numerous pairwise comparisons, the Benjamini-Hochberg (BH) procedure was applied with an FDR threshold of 0.05. Statistical tests were performed using GraphPad Prism 9. Results were considered significant when p-values or adjusted p-values were less than 0.05.
+
+DECLARATIONS
+
+Acknowledgments: The authors are Genopole's members, first French biocluster dedicated to genetic, biotechnologies and biotherapies. We are grateful to the “Imaging and Cytometry Core Facility” and to the in vivo evaluation, services of Genethon for technical support, to Ile-de-France Region, to Conseil Départemental de l’Essonne (ASTRE), INSERM and GIP Genopole, Evry for the purchase of the equipment. We would like to acknowledge the technical help of Carolina Pacheco Algalan and Alejandro Arco Hierves. The Genotype-Tissue Expression (GTEx) Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health, and by NCI, NHGRI, NHLBI, NIDA, NIMH, and NINDS.
+
+Funding: This work was supported by the “Association Française contre les Myopathies” (AFM), and “Institut National de la Santé Et de la Recherche Médicale” (INSERM, FranceRelance N°221513A10).
+
+Author contributions: The project was conceptualized by A.H.V. and I.R. A.H.V., L.S.P., and J.P. conducted experiments and performed data analysis. Funding supporting this project was obtained by I.R. A.H.V. and I.R. supervised the project. The manuscript was written by A.H.V. and I.R.
+
+Competing interests: A.H.V. and I.R. are inventors on PCT application EP2023/065499 for the integration of RGDlxL/I motif in AAV capsid for enhanced muscle transduction efficiency. I.R. is a part-time employee of Atamyo Therapeutics. The other authors declare that they have no competing interests.
+
+Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. All transcriptomic data will be deposited in the NCBI Sequence Read Archive
+(SRA) upon publication. Processed data including differential gene expression analysis are available in data file S1 and S2. The plasmid constructs and reagents generated as part of this study are available under a material transfer agreement from the corresponding authors.
+
+REFERENCES
+
+1. Wang, D., Tai, P.W.L. & Gao, G. Adeno-associated virus vector as a platform for gene therapy delivery. Nat Rev Drug Discov **18**, 358-378 (2019).
+2. Pupo, A. et al. AAV vectors: The Rubik's cube of human gene therapy. *Molecular therapy : the journal of the American Society of Gene Therapy* **30**, 3515-3541 (2022).
+3. Crudele, J.M. & Chamberlain, J.S. AAV-based gene therapies for the muscular dystrophies. *Hum Mol Genet* **28**, R102-R107 (2019).
+4. Duan, D. Systemic AAV Micro-dystrophin Gene Therapy for Duchenne Muscular Dystrophy. *Molecular therapy : the journal of the American Society of Gene Therapy* **26**, 2337-2356 (2018).
+5. Mack, D.L. et al. Systemic AAV8-Mediated Gene Therapy Drives Whole-Body Correction of Myotubular Myopathy in Dogs. *Molecular therapy : the journal of the American Society of Gene Therapy* **25**, 839-854 (2017).
+6. Mercuri, E., Bonnemann, C.G. & Muntoni, F. Muscular dystrophies. *Lancet* **394**, 2025-2038 (2019).
+7. Ertl, H.C.J. Immunogenicity and toxicity of AAV gene therapy. *Front Immunol* **13**, 975803 (2022).
+8. Verdera, H.C., Kuranda, K. & Mingozzi, F. AAV Vector Immunogenicity in Humans: A Long Journey to Successful Gene Transfer. *Molecular therapy : the journal of the American Society of Gene Therapy* **28**, 723-746 (2020).
+9. High-dose AAV gene therapy deaths. *Nature biotechnology* **38**, 910 (2020).
+10. Ogden, P.J., Kelsic, E.D., Sinai, S. & Church, G.M. Comprehensive AAV capsid fitness landscape reveals a viral gene and enables machine-guided design. *Science* **366**, 1139-1143 (2019).
+11. DiMattia, M.A. et al. Structural insight into the unique properties of adeno-associated virus serotype 9. *Journal of virology* **86**, 6947-6958 (2012).
+12. Li, C. & Samulski, R.J. Engineering adeno-associated virus vectors for gene therapy. *Nat Rev Genet* **21**, 255-272 (2020).
+13. Tseng, Y.S. & Agbandje-McKenna, M. Mapping the AAV Capsid Host Antibody Response toward the Development of Second Generation Gene Delivery Vectors. *Front Immunol* **5**, 9 (2014).
+14. Buning, H. & Srivastava, A. Capsid Modifications for Targeting and Improving the Efficacy of AAV Vectors. *Molecular therapy. Methods & clinical development* **12**, 248-265 (2019).
+15. Weinmann, J. et al. Identification of a myotropic AAV by massively parallel in vivo evaluation of barcoded capsid variants. *Nature communications* **11**, 5432 (2020).
+16. El Andari, J. et al. Semirational bioengineering of AAV vectors with increased potency and specificity for systemic gene therapy of muscle disorders. *Science advances* **8**, eabn4704 (2022).
+17. Tabebordbar, M. et al. Directed evolution of a family of AAV capsid variants enabling potent muscle-directed gene delivery across species. Cell **184**, 4919-4938 e4922 (2021).
+18. Ruoslahti, E. & Pierschbacher, M.D. Arg-Gly-Asp: a versatile cell recognition signal. *Cell* **44**, 517-518 (1986).
+19. Bauer, A. et al. Molecular Signature of Astrocytes for Gene Delivery by the Synthetic Adeno-Associated Viral Vector rAAV9P1. *Adv Sci (Weinh)* **9**, e2104979 (2022).
+20. Zolotukhin, S., Trivedi, P.D., Corti, M. & Byrne, B.J. Scratching the surface of RGD-directed AAV capsid engineering. *Molecular therapy : the journal of the American Society of Gene Therapy* **29**, 3099-3100 (2021).
+21. Hynes, R.O. Integrins: a family of cell surface receptors. *Cell* **48**, 549-554 (1987).
+22. Hussein, H.A. et al. Beyond RGD: virus interactions with integrins. *Arch Virol* **160**, 2669-2681 (2015).
+23. Asokan, A., Hamra, J.B., Govindasamy, L., Agbandje-McKenna, M. & Samulski, R.J. Adeno-associated virus type 2 contains an integrin alpha5beta1 binding domain essential for viral cell entry. *Journal of virology* **80**, 8961-8969 (2006).
+24. Summerford, C., Bartlett, J.S. & Samulski, R.J. AlphaVbeta5 integrin: a co-receptor for adeno-associated virus type 2 infection. *Nat Med* **5**, 78-82 (1999).
+25. Stewart, P.L. & Nemerow, G.R. Cell integrins: commonly used receptors for diverse viral pathogens. *Trends Microbiol* **15**, 500-507 (2007).
+26. Strauch, E.M. et al. Computational design of trimeric influenza-neutralizing proteins targeting the hemagglutinin receptor binding site. *Nature biotechnology* **35**, 667-671 (2017).
+27. Cao, L. et al. Design of protein-binding proteins from the target structure alone. *Nature* **605**, 551-560 (2022).
+28. Ruoslahti, E. RGD and other recognition sequences for integrins. *Annual review of cell and developmental biology* **12**, 697-715 (1996).
+29. Chemello, F. et al. Degenerative and regenerative pathways underlying Duchenne muscular dystrophy revealed by single-nucleus RNA sequencing. *Proceedings of the National Academy of Sciences of the United States of America* **117**, 29691-29701 (2020).
+30. Pang, X. et al. Targeting integrin pathways: mechanisms and advances in therapy. *Signal Transduct Target Ther* **8**, 1 (2023).
+31. Shen, X., Storm, T. & Kay, M.A. Characterization of the relationship of AAV capsid domain swapping to liver transduction efficiency. *Molecular therapy : the journal of the American Society of Gene Therapy* **15**, 1955-1962 (2007).
+32. Ballon, D.J. et al. Quantitative Whole-Body Imaging of I-124-Labeled Adeno-Associated Viral Vector Biodistribution in Nonhuman Primates. *Human gene therapy* **31**, 1237-1259 (2020).
+33. Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. *Nature* **596**, 583-589 (2021).
+34. Mirdita, M. et al. ColabFold: making protein folding accessible to all. Nature methods **19**, 679-682 (2022).
+35. Dong, X. et al. Force interacts with macromolecular structure in activation of TGF-beta. *Nature* **542**, 55-59 (2017).
+36. Dong, X., Hudson, N.E., Lu, C. & Springer, T.A. Structural determinants of integrin beta-subunit specificity for latent TGF-beta. *Nature structural & molecular biology* **21**, 1091-1096 (2014).
+37. Huang, P.S. et al. RosettaRemodel: a generalized framework for flexible backbone protein design. *PloS one* **6**, e24109 (2011).
+38. Alford, R.F. et al. The Rosetta All-Atom Energy Function for Macromolecular Modeling and Design. *Journal of chemical theory and computation* **13**, 3031-3048 (2017).
+39. Duan, D., Goemans, N., Takeda, S., Mercuri, E. & Aartsma-Rus, A. Duchenne muscular dystrophy. *Nat Rev Dis Primers* **7**, 13 (2021).
+40. Stedman, H.H. et al. The mdx mouse diaphragm reproduces the degenerative changes of Duchenne muscular dystrophy. *Nature* **352**, 536-539 (1991).
+41. Bourg, N. et al. Co-Administration of Simvastatin Does Not Potentiate the Benefit of Gene Therapy in the mdx Mouse Model for Duchenne Muscular Dystrophy. *Int J Mol Sci* **23** (2022).
+42. Rouillon, J. et al. Serum proteomic profiling reveals fragments of MYOM3 as potential biomarkers for monitoring the outcome of therapeutic interventions in muscular dystrophies. *Hum Mol Genet* **24**, 4916-4932 (2015).
+43. Eymard, B. et al. Primary adhalinopathy (alpha-sarcoglycanopathy): clinical, pathologic, and genetic correlation in 20 patients with autosomal recessive muscular dystrophy. *Neurology* **48**, 1227-1234 (1997).
+44. Duclos, F. et al. Progressive muscular dystrophy in alpha-sarcoglycan-deficient mice. *The Journal of cell biology* **142**, 1461-1471 (1998).
+45. Bryant, D.H. et al. Deep diversification of an AAV capsid protein by machine learning. *Nature biotechnology* **39**, 691-696 (2021).
+46. Israeli, D. et al. An AAV-SGCG Dose-Response Study in a gamma-Sarcoglycanopathy Mouse Model in the Context of Mechanical Stress. *Molecular therapy. Methods & clinical development* **13**, 494-502 (2019).
+
+**Figures**
+Figure 1
+
+Computational rational AAV capsid design to bind to αVβ6 integrin
+
+A. Overview of the design pipeline, including three steps: 1. Capsid 3D structures were obtained either from the PDB database or predicted by AlphaFold2. 2. The capsid VR4 loop was completely replaced by integrating the binding motif, which was extracted from receptor’s natural binder, using RosettaRemodel
+protocol. 3. Top scored designs from the previous grafting step were docked onto the intended receptor in silico to verify the binding potential of the designed capsid. B. An illustration of the sampling for low-energy sequence-structure pairs during motif-grafting process. Capsid VR4 after removing the loop was colored in blue, extracted binding motif was colored in red. The sampled linkers and sequences (Fig. S1F) were labeled in green. C-D. The three lowest energy designs after grafting TGFβ3 (C) and TGFβ1 (D) into the capsid VR4. All top designs showed convergence in structures and sequences, suggesting sampling approached the global optimum. E-F. Retrospective docking of motif-grafted capsids (E. Cap9rh74_4um9 and F. Cap9rh74_5ffo) onto the αVβ6 structure. The left panels are illustrations of the structures with the lowest energy at the interface of capsid and integrin proteins (dG_separated: difference in free energy of two proteins). Both two newly designed VR4s (colored in green) were predicted to bind to the αVβ6 complex at very similar positions to natural binding motifs (colored in red). The right panels are scatter plots of dG_separated energy versus root-mean-square deviation (RMSD) from the lowest energy structure of all sampled docking positions.
+Fig. 2
+
+A  AAV productivity
+
+B
+
+C  Vector Copy Number
+   293_aVβ6
+
+D  Luciferase activity
+   293_aVβ6
+
+E  Vector Copy Number
+   Protein Inhibition
+   AAV + r.SGCA
+   AAV + r.ITGAV-B6
+
+F
+
+G  Vector Copy Number
+   Myoblasts
+
+H  Vector Copy Number
+   Myotubes
+
+I  Luciferase activity
+   Myoblasts
+
+K  Luciferase activity
+   Myotubes
+
+Figure 2
+
+Designed AAV_ITGs were well-produced and improved transduction via αVβ6 binding.
+
+A. AAV titers of different AAV variants in bulked small-scale production in suspension three-day post-triple-transfection (2ml production, n=6, one-way ANOVA). B. Western blot of VP proteins from purified AAVs showed similar VP ratios for designed AAV_ITGs capsids compared to AAV9 and AAV9rh74,
+suggesting successful capsid assembly. **C-D.** VCN (**C**) and luciferase activity (**D**) of 293_αVβ6 after AAV infection (n=3-4, one-way ANOVA). Both the two designed AAV_ITGs showed enhanced VCN and luciferase activities compared to AAV9rh74 and AAV9. **E.** Inhibition of cell entry of designed AAV_ITGs, but not for AAV9 or AAV9rh74, in 293_αVβ6 cells by αVβ6 recombinant protein. AAVs were preincubated with αVβ6 recombinant protein (r.ITGAV-B6) for 30 minutes at 37°C before infection (n=3, two-way ANOVA). SGCA recombinant protein (r.SGCA) was used as the control. **F-K.** Enhanced transduction of AAV_ITGs in *in vitro* human differentiated myotubes, but not in myoblasts. **F.** Representative images of the GFP signal of myotubes 48 hours post-infection (scale bar: 400μm). **G-K.** VCN and luciferase activities of AAV_ITGs in comparison with AAV9 and AAV9rh74 in myoblasts (**G,I**) and myotubes (**H,K**) (n=3-4, one-way ANOVA).
+Fig. 3
+
+A
+![Scheme of in vivo experiment. AAVs (CMV_GFP-Luciferase) were injected intravenously into 6wo C57BL6 mice (n=4) at the dose of 1E13 vg/kg.](page_120_120_1342_180.png)
+
+B
+![Representative images of the bioluminescence signal](page_120_320_1342_180.png)
+
+C
+![Bar graphs showing vector copy number in various tissues](page_120_520_1342_180.png)
+
+D
+![Bar graphs showing GFP mRNA level in various tissues](page_120_720_1342_180.png)
+
+E
+![Peptide insertion (VR8) and Loop replacement (VR4) diagrams for AAV9 and AAV9rh74](page_120_920_1342_180.png)
+
+F
+<table>
+  <tr>
+    <th></th>
+    <th>AAV9</th>
+    <th>AAV9rh74</th>
+    <th>AAV9rh74_4um9</th>
+  </tr>
+  <tr>
+    <td>583</td>
+    <td>NHQSAQ - - - - - - - AQ - AQT</td>
+    <td>NHQGSGRGLGLSAGAAQT</td>
+    <td>NHQGSGRGLGLSAGAAQT</td>
+  </tr>
+  <tr>
+    <td>593</td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+  <tr>
+    <td>MYOAAV1A</td>
+    <td>NHQSAQ - RGDLTTPAQ - AQT</td>
+    <td>NHQGP - GRGDQTTLAQ - AQT</td>
+    <td>NHQGP - GRGDQTTLAQ - AQT</td>
+  </tr>
+  <tr>
+    <td>MYOAAV2A</td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+  <tr>
+    <td>461</td>
+    <td>QS TGGTA - - - - GTGQL</td>
+    <td>QS TGGTA - - - - GTGQL</td>
+    <td>QS TGGTA - - - - GTGQL</td>
+  </tr>
+  <tr>
+    <td>462</td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+</table>
+
+G
+![Bar graphs showing vector copy number in various tissues](page_120_1120_1342_180.png)
+
+H
+![Bar graphs showing GFP mRNA level in various tissues](page_120_1320_1342_180.png)
+
+Figure 3
+
+Designed AAV_ITGs showed enhanced transduction in skeletal and cardiac muscles while strongly liver-detargeted in vivo.
+
+A. Scheme of in vivo experiment. AAVs (CMV_GFP-Luciferase) were injected intravenously into 6wo C57BL6 mice (n=4) at the dose of 1E13 vg/kg. B. Representative images of the bioluminescence signal
+20 days post-infection. **C-D.** VCN (**C**) and gene expression (**D**) (GFP mRNA level in the liver and luciferase activity in other organs) for different AAVs in liver, skeletal muscles, heart, lung, and kidney (n=4, one-way ANOVA). Both designed AAV_ITGs strongly detargeted from the liver compared to AAV9, while they significantly improved VCN and luciferase activities over AAV9rh74 (and AAV9 with AAV9rh74_4um9 variant) in skeletal and cardiac muscles, and were detected and expressed at low levels in lung and kidney. **E-H.** Comparison of the AAV9rh74_4um9 variant with other public myotropic AAVs (mAAVs) \(^{15,17}\). **E.** Illustration of the differences between mAAVs and AAV9rh74_4um9 at modification sites in capsid protein and modification methods. **F.** The VR8 loop sequences of mAAVs compared to VR8 of their backbone AAV9, and VR4 of AAV9rh74_4um9 compared to VR4 of AAV9rh74. **G-H.** VCN (**G**) and gene expression (**H**) (GFP mRNA level in liver and luciferase activity in other organs) of different AAVs in liver, skeletal muscles, heart, lung, kidney, and brain (n=4, one-way ANOVA). AAV9rh74_4um9 showed similar VCN and gene expression in skeletal muscle to other mAAVs, while being significantly more strongly detargeted from the liver.
+Fig. 4
+
+Figure 4
+
+Low-dose gene transfer by LICA1 was more effective and better at restoring dystrophic phenotypes than AAV9 in the DMD mouse model.
+
+A-B. Comparison of transduction efficacy between AAV9 and LICA1 in all three muscles that were tested, in terms of VCN (A), and μDys RNA level (B). C. Comparison of percentage of successfully transduced
+(dystrophin-positive) fibers in all three muscles that were tested. Note that TA, Qua, Dia muscles are presented in increasing order of severity. **D-E.** Comparison of restoration levels in dystrophic histological features between AAV9 and LICA1 in all three muscles that were tested, in terms of percentage of centro-nucleated fibers (**D**) and fibrosis level (**E**). Illustrated images in **C-E** are of quadriceps muscles (scale bar: 100μm). F. Serum MYOM3 level – indicator of muscle damage – 4 weeks post-injection (n=5, one-way ANOVA). **G-I.** Comparison of functional restoration between AAV9 and LICA1 by Escape test – global force measurement (**G**, n=6), tetanus force of TA muscle (**H**, n=10-12), and twitch force of TA muscle (**I**, n=9-12). **K-N.** Comparison of restoration in global transcriptomic changes in quadriceps muscle between AAV9 and LICA1 (n=4, adjusted p-values < 0.05). **K.** The heatmap presents the log2 fold change (log2FC) in comparison to WT muscle for all 8717 DEGs found in mdx muscle (compared to WT). The log2FC values are illustrated in row Z-scores, colored from blue to red, arranged from lowest to highest. **L-N.** Volcano plots of multiple comparisons illustrate transcriptomic changes before and after AAV treatment. As a reference, 4216 downregulated and 4501 upregulated DEGs found in mdx were colored blue and red, respectively, in all volcano plots. Among these DEGs, the number of genes found to be significantly different in each pair-wise comparison were labeled in the upper corners. **L.** Volcano plots comparing mdx/WT transcriptomes. **M.** Volcano plots comparing mdx to AAV-treated transcriptomes, in which significant DEGs are the genes correctly restored after AAV treatment. **N.** Volcano plots comparing AAV treatment to WT, in which significant DEGs are the genes that are not or incompletely restored after AAV treatment.
+Fig. 5
+
+A
+AAV9
+LICA1
+SGCA-/- 4-week-old
+Dose: 5E12 vg/kg
+(weeks)
+Sacrifice
+
+B
+Vector Copy Number
+TA   Qua   Dia
+
+C
+hSGCA mRNA level
+TA   Qua   Dia
+
+D
+DAPI SOCA
+WT   KO   AAV9   LICA1
+
+E
+HPS
+WT   KO   AAV9   LICA1
+
+F
+Red Sinus
+WT   KO   AAV9   LICA1
+
+G
+TA   Qua   Dia
+
+H
+Escape test
+
+I
+TA - Tetanus
+
+J
+Serum MYOM3
+
+K
+
+L
+
+M
+Dysregulated
+
+N
+Corrected by AAV
+
+O
+NOT corrected by AAV
+
+Figure 5
+
+Low-dose gene transfer by LICA1 was better at restoring dystrophic phenotypes and functionality than AAV9 in the LGMDR3 mouse model.
+
+A. Scheme of in vivo experiment: LICA1 (9rh74_4um9) or AAV9 were injected intravenously into a 4wo SGCA-KO mouse model at the dose of 5E12 vg/kg (expression cassette: hACTA1_hSGCA_HBB2-pA, n=3-
+5). Three skeletal muscles in increasing order of severity, TA, Qua, and Dia, were analysed 4 weeks post-injection. **B-D**. Comparison of transduction efficacy between AAV9 and LICA1 in all three muscles that were tested in terms of VCN (**B**), hSGCA mRNA level (**C**), and percentage of successfully transduced (SGCA-positive) fibers (**D**). **E-G**. Comparison of restoration levels in dystrophic histological features between AAV9 and LICA1 in all three muscles that were tested in terms of percentage of centro-nucleated fibers (**E**), fibrosis level (**F**), and fiber size distribution (**G**). Illustrated images in **D-F** are of quadriceps muscles (scale bar: 100 μm). **H-K**. Comparison of functional restoration between AAV9 and LICA1 using the escape test – global force measurement (**H**), tetanus force of TA muscle (**I**), and serum MYOM3 level – indicator of muscle damage (**K**). **L-O**. Comparison of restoration in global transcriptomic changes in quadriceps muscle between AAV9 and LICA1 (n=4, adjusted p values < 0.05). **L**. The heatmap presents the log2 fold change (log2FC) in comparison to WT muscle for all 8591 DEGs found in KO muscle (compared to WT). The log2FC values are illustrated by row Z-scores, colored from blue to red, arranged from lowest to highest. **M-O**. Volcano plots of multiple comparisons illustrate transcriptomic changes before and after AAV treatment. As a reference, 4035 downregulated and 4556 upregulated DEGs found in KO were colored blue and red, respectively, in all volcano plots. Among these DEGs, the number of genes found to be significantly different in each pair-wise comparison were labeled in the upper corners. **M**. Volcano plots comparing KO/WT transcriptomes. **N**. Volcano plots comparing KO to AAV-treated transcriptomes, in which significant DEGs are the genes correctly restored after AAV treatment. **O**. Volcano plots comparing AAV treatment to WT, in which significant DEGs are the genes that are not or incompletely restored after AAV treatment.
+
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• ALICA1supp.pdf
+• DatafileS2DEGSAGCA.xlsx
+• DatafileS1DEGDMD.xlsx

015d965c0cb0c74c387f9596e3f91c0197ff39a7333b1c0b5bef96993bcbbd18/preprint/preprint.md

@@ -0,0 +1,227 @@
+The emergence of three-dimensional chiral domain walls in polar vortices
+
+Sandhya Susarla ( sandhya.susarla@asu.edu )
+Arizona State University https://orcid.org/0000-0003-1773-0993
+Shang-Lin Hsu
+University of california, Berkeley
+Fernando Gómez-Ortiz
+Universidad de Cantabria https://orcid.org/0000-0002-7203-8476
+Pablo Garcia-Fernandez
+Universidad de Cantabria https://orcid.org/0000-0002-4901-0811
+Benjamin Savitzky
+Cornell University
+SUJIT DAS
+Indian Institute of Science, Bangalore https://orcid.org/0000-0001-9823-0207
+Piush Behera
+University of California, Berkeley
+Javier Junquera
+Universidad de Cantabria, Cantabria Campus Internacional https://orcid.org/0000-0002-9957-8982
+Peter Ercius
+Lawrence Berkeley National Laboratory https://orcid.org/0000-0002-6762-9976
+Ramamoorthy Ramesh
+Rice University
+Colin Ophus
+National Center for Electron Microscopy Facility, Molecular Foundry, Lawrence Berkeley National Laboratory
+
+Article
+
+Keywords:
+
+Posted Date: February 8th, 2023
+
+DOI: https://doi.org/10.21203/rs.3.rs-2551328/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on July 25th, 2023. See the published version at https://doi.org/10.1038/s41467-023-40009-2.
+The emergence of three-dimensional chiral domain walls in polar vortices
+
+Sandhya Susarla1,2,8**, Shanglin Hsu1,2#, Fernando Gómez-Ortiz4, Pablo García-Fernández4, Benjamin H. Savitzky1, Sujit Das2,6, Piush Behera3, Javier Junquera4, Peter Ercius1, Ramamoorthy Ramesh1,2,5,7,8*, Colin Ophus1*
+
+1: National Center for Electron Microscopy, Molecular Foundry, Lawrence Berkeley National Laboratory, Berkeley, CA, USA 94720
+2: Materials Sciences Division, Lawrence Berkeley Laboratory, Berkeley CA, USA 94720
+3: Department of Materials Science & Engineering, University of California, Berkeley, CA, USA 94720
+4: Departamento de Ciencias de la Tierra y Física de la Materia Condensada, Universidad de Cantabria, Cantabria Campus Internacional Santander, Spain, 39005
+5: Department of Physics, University of California, Berkeley Berkeley, CA, USA 94720
+6: Material Research Centre, Indian Institute of Science, Bangalore, 560012
+7: Department of Physics, Rice University, Houston, TX, USA, 77005
+8: Department of Materials Science and Nanoengineering, Houston, TX, USA, 77005
+#Equal contribution.
+$ Present address: School for Engineering of Matter, Transport, and Energy, Arizona State University
+
+Corresponding authors: sandhya.susarla@asu.edu, ramamoorthy.ramesh@rice.edu, and clophus@lbl.gov
+
+Abstract
+
+Chirality or handedness of a material can be used as an order parameter to uncover emergent electronic properties for quantum information science. Conventionally, chirality is found in naturally occurring biomolecules and magnetic materials. Chirality can be engineered in a topological polar vortex ferroelectric/dielectric system via atomic-scale symmetry-breaking operations. We use four-dimensional scanning transmission electron microscopy (4D-STEM) to map out topology-driven three-dimensional domain walls, where the handedness of two neighbor topological domains change or remain the same. The nature of the domain walls is governed by the interplay of local perpendicular (lateral) and parallel (axial) polarization with respect to the tubular vortex structures. Unique symmetry-breaking operations and the finite nature of domain walls results in a triple point at the junction of chiral and achiral domain walls. The unconventional nature of the domain walls with triple point pairs may result in unique electrostatic and magnetic properties potentially useful for quantum sensing applications.
+Introduction
+
+Chirality is a unique topological feature that drives many-body interactions in naturally occurring organic molecules and proteins \(^1\), subatomic particle physics \(^2\), and solid-state physics \(^3\). The symmetries in a chiral system are configured in such a way that its mirror image cannot be superimposed on itself, manifesting a handedness to the system as exemplified by screws and our own hands. Chirality also exists at the microscale/nanoscale level in inorganic and organic materials such as liquid crystals \(^4\), spin textures in ferromagnets \(^{5,6}\), amino acids, and D/L-glucose molecules \(^1\) with applications in spin selectivity-based quantum sensing \(^7\), non-linear optics \(^8\), and biosensing applications \(^9\). However, there are very few examples of chiral inorganic ferroelectric crystals which could have fundamentally different domain textures \(^{10-13}\). Over the past few years, novel polarization textures in ferroelectrics such as merons \(^{14}\), polar flux-closure domains \(^{15,16}\), vortices \(^{17}\), bubble domains \(^{18,19}\), super crystals \(^{20,21}\) and skyrmions \(^{22}\) have been engineered in oxide superlattices, emerging from the careful interplay of elastic, electrostatic and gradient energies of electric dipoles. The electric dipole arrangement and complex orbital hybridization in these systems have been probed by the x-ray scattering techniques \(^{23}\), scanning transmission electron microscopy (STEM) \(^{17}\)-electron energy loss spectroscopy (EELS) \(^{24}\), phase-field simulations\(^{12,14,15,17,20}\) and atomistic first- and second-principles calculations \(^{17,18,20,22,23}\). Surprisingly, the presence of chirality has been observed in one such topological texture i.e. polar vortices in PbTiO$_3$/SrTiO$_3$ superlattices from resonant soft x-ray scattering (RXS) \(^{23}\), second harmonic generation (SHG) and second principles calculations \(^{25}\). The presence of chirality in polar vortices is an emergent phenomenon because none of the parent compounds such as SrTiO$_3$ or PbTiO$_3$ are known to be chiral. It has been recently shown experimentally and theoretically that the presence of chirality in these systems might be due to different sources.
+First, the strongest one is the coexistence of vortices with an axial component of the polarization, perpendicular to the vortex plane \(^{26}\). The second factor is the buckling of the vortices (i.e., a staggered vortex configuration where the center of the clockwise and counterclockwise vortices are located at different heights) combined with different sizes of the up and down domains results in a chiral structure, although its strength is smaller than in the first scenario. This last source of chirality can be reversed by external electric fields. The first experimental demonstration was in \(^{25}\), where chirality switches in a reversible, deterministic, and non-destructive fashion over mesoscale regions \(^{25}\).
+
+Without any prior knowledge, one would expect the as-grown sample as a racemic mixture, i.e., equal amounts of left-handed and right-handed domain enantiomers, where chirality within each domain comes from a combination of the two sources described above. To have a non-destructive switchable chirality, it is essential to understand the role of the domain walls separating the enantiomers. In other words, what local physical parameters play a role when the handedness in neighboring domains changes? This includes the offset between the center of the cores, the axial component at the center of the clockwise/counterclockwise vortices, the sense of rotation of the vortices when they merge at the domain wall, the presence of dislocations, or the combined effect of all of them. A proper understanding of how the left/right-handed domains evolve at the nanoscale is crucial to design new electrically switchable chiral devices that can be measured without scientifically sophisticated techniques. Indeed, a proper answer to this question will pave the way for the use of these chiral textures in next-generation technologies.
+
+Four-dimensional (4D)-STEM can precisely measure strain, and thus spontaneous polarization in ferroelectrics due to the violation of Friedel’s Law \(^{27-29}\). This makes 4D-STEM a unique tool to probe polarization in three dimensions and understand emergent chirality in
+polarization vortices. In the current work, we have used 4D-STEM to probe three-dimensional domain walls in polar vortices in oxide superlattices and understand the nano-scale nature of chirality. We find that both achiral and chiral domain walls coexist in the same system. The chiral to achiral domain wall transition is driven by the change in local axial and lateral polarization direction across the domain wall. We have discovered new pair of triple points with the opposite/same sense of rotation at the junction of achiral and chiral domain walls. Finally, we unravel all the possible configurations of chiral and achiral domain walls in this system through different symmetry-breaking operations.
+
+Results
+
+Trilayer (SrTiO3)16/(PbTiO3)16/(SrTiO3)16 (STO/PTO/STO) were grown on orthorhombic DyScO3 (DSO) [110]₀ substrates with SrRuO3 as a buffer layer using reflection high-energy electron diffraction (RHEED)-assisted pulsed-laser deposition (PLD) (Figure 1A and Supplementary Information). The polar vortex phase in this system is stabilized as a consequence of the interplay between depolarization energy at the PTO/STO interfaces, elastic energy from the tensile strain imposed by the DSO substrate, and the gradient energy in the ferroelectric \(^{17}\).
+
+A direct way to measure the polar textures in vortex topologies is through electron microscopy using atomic resolution images. We have used different types of STEM and TEM techniques to characterize the exact positions of the vortex centers. Figure 1a shows a low-magnification bright-field STEM image of the superlattice trilayer with an SRO buffer layer along the [1\(\bar{1}0\)₀] zone axis. The diffraction contrast in BF-STEM allows us to directly locate the vortex center as dark contrast; marked using red circles in Figure 1a. To precisely understand the polarization or displacement texture around each vortex center, we obtained the A-site
+displacement vector maps at atomic resolution via gaussian fitting at A-sites (Methods, supplementary information). Figure 1b shows the High angle annular dark field (HAADF)-STEM image of trilayer STO/PTO/STO along [1\bar{1}0]_o zone axis where brighter regions are PTO and darker regions are STO. Overlaid yellow arrows show clockwise and counterclockwise rotating curls in the displacement of the A-cation in PTO/STO superlattices. The coexistence of the concomitant non-zero curl of polarization (red/blue contrast) with an alternating axial component of the polarization (perpendicular to the plane defined by the vortices) is the first symmetry-breaking operation that results in emergent chirality in an otherwise non-chiral system. The non-zero curl is larger in continuously rotating polarization textures such as polar vortices^{25,27}, merons^{30}, and skyrmions^{22} than in other polarization textures such as flux closure domains^{15,16} where the curl vanished in the central regions with 180^0 domain walls. Additionally, we observe that the cores of the polarization curls (indicated as blue/red contrast) are not located exactly at the center of the PTO layer, but follow a zig-zag type pattern, giving rise to a *net* in-plane polarization rotation along [001]_o (lateral component) indicated as P_x in Figure 1b. This buckling, combined with a small difference in the size of the up and down domains, is the second symmetry-breaking operation that results in net chirality; in agreement with previous observations^{25}.
+Figure 1. Chirality at different length scales. (a) Bright-Field STEM image showing the stack of 16 STO/16 PTO/16 STO with SRO buffer layer on DSO substrate. The red regions near the center of the PTO layer indicate the position of the vortex core approximately. (b) Vector mapping of the local displacements of A sites of highlighted region in (a) overlaid on HAADF-STEM image. The local red- and blue contrast at the center of the PTO layer indicates the local non-zero curl of displacement. The net lateral polarization resulting from vortex off-centering is indicated at the top. The dotted black line shows the center of the PTO layer. (c) Dark field TEM image along [110]₀ direction displaying long tubular vortex structure with domain walls shown as white dashed lines.
+
+The various permutations and combinations of atomic scale symmetry-breaking operations such as a non-zero curl of polarization together with the presence of an axial component, and the buckling of the vortices that yield a non-zero polarization component along [001]₀, result in different types of domains walls at the mesoscale. We can visualize the mesoscale domain walls in these topological structures by imaging the vortices along the [110]₀ zone axis using weak beam dark field TEM (Figure 1c). We can observe the tubular nature of vortex textures by long bright and dark stripes regions in the image. Additionally, we observe different domain wall features (indicated as white dashed lines) cutting across vortex tubes.
+
+Overall, if we combine our observations from DF-TEM and HR-STEM, we observe a three-
+dimensional structure in the PbTiO$_3$ layer sandwiched between SrTiO$_3$ layers where the polarization curls follow a tubular pattern (Figure 2a). Unfortunately, the images formed by weak beam dark field, HRSTEM, and BF-STEM are merely atomic projections and cannot give us an estimate of physical quantities such as polarization and chirality. On the other hand, 4D-STEM allows us to collect a diffraction pattern at each probe position, which can then be used to create precise maps of physical quantities. For the present experiment, we performed 4D-STEM imaging on a trilayer STO/PTO/STO along the [110]$_o$ zone axis (Figure 2a). We used a probe size of ~7 Å, larger than the STO/PTO unit cell dimensions (~4 Å) to remove the atomic-resolution signal and to estimate the polarization at unit cell resolution. The 4D-STEM analysis was carried out in open source py4DSTEM analysis package $^{31}$. We define the [1\bar{1}0]$_o$ direction as axial and [001]$_o$ direction as lateral. The rotation calibration was performed between the real space and diffraction space to determine the lateral and axial axis. Details are given in the supplementary information. For initial visualization of the polar textures, we created a virtual dark field image using the [2\bar{1}0]$_o$ disk (disk 3) as shown in Figure 2b-c. The polarization from the PTO layer can be determined qualitatively by subtracting opposite Friedel pair disks due to the violation of Friedel’s law $^{27-29}$. The polarization maps corresponding to regions delimited by rectangles in Figure 2b are shown in Figure 2d-e. We observe alternate longitudinal red and blue stripes representing positive and negative polarization in both the lateral ([001]$_o$) and axial [1\bar{1}0]$_o$ directions with domain walls as seen from the weak beam dark field images marked with green and magenta lines in Figure 2d and with a black line in Figure 2e. We observe that the axial polarization magnitude is relatively smaller than the lateral polarization in agreement with the predictions from previous second principles calculations $^{23,25}$.
+Figure 2. Identification of polarization using 4D-STEM mapping. (a) Schematic showing the e-beam scanned in 4D STEM mode across the vortex sample in in-plane geometry. (b) Virtual dark field image of the vortex region obtained via integrating the intensity of disk 3 from the mean diffraction pattern in (c). Zoomed-in images of the (d) dash-dot region (e) solid line region showing lateral (P [001]_o) and axial (P [1\bar{1}0]_o) polarization maps in vortices. \( \alpha \), \( \beta \), and \( \gamma \) domain walls can be identified. \( \alpha \) domain wall (magenta curve) has an anti-parallel lateral (P [001]_o) component, \( \beta \) (green curve) domain wall has an anti-parallel axial component (P [1\bar{1}0]_o), \( \gamma \) has both axial and lateral components antiparallel across the domain wall.
+
+We track the relative domain shift across the wall by the black dotted line as shown in Figure 2d-e. In Figure 2d, we observe a \( \alpha \)-domain wall (magenta line) where the lateral polarization shifts whereas the axial polarization remain the same, as shown in Figure 3. We also detect a \( \beta \)-domain wall (green line), where the axial polarization shifts, but the lateral polarization remains the same. Finally, in Figure 2e there is a third domain wall configuration i.e., a \( \gamma \)-domain wall (black line) as well where both the lateral and axial polarization shift. We verified this observation with average line profiles across the domain walls (Supplementary
+information, Figure S1). To detect whether a change of chirality occurs at these domain walls, a way to quantify the chirality is required. The order parameter that best captures the breakdown of chiral symmetry is the helicity H of the chiral field. In our case, the chiral field is the polarization, and for the helicity, we borrow the definition from fluid dynamics:
+
+\[
+\mathcal{H} = \int \vec{p} \cdot (\vec{\nabla} \times \vec{p})\ d^3r,
+\]
+
+where \( \vec{p} \) is the local value of polarization. Note that \( \mathcal{H} \) changes sign upon a mirror symmetry reflection \(^{32,33}\). A nonzero helicity means chirality or lack of mirror symmetry of the polarization texture: right (left) handedness can be associated with positive (negative) values of \( \mathcal{H} \). Assuming a vortex structure where the polarization lines in the plane defined by the vortices are closed, and that we can measure the axial and lateral components of the polarization at the topmost PbTiO$_3$ layer, then the previous equation can be estimated by
+
+\[
+\mathcal{H} = 2 \cdot < p_{lateral} > \cdot < p_{axial} >
+\]
+
+where \( p_{lateral} \) and \( p_{axial} \) are polarization along lateral and axial directions (Supplementary information , Figure S2). Using this equation we can understand the nature of domain walls found in Figure 2d-e. For the \( \alpha/\beta \) domain wall, only one of the lateral/axial polarization sign change across the domain wall. This causes a change in the overall sign of helicity, thus making them chiral domain walls. On the other hand, the \( \gamma \)-domain wall has both a lateral and axial polarization switch, which doesn’t change the overall helicity of the system, thus making it an achiral domain wall.
+Figure 3. Identification of triple point topologies and chiral domain walls. (a) The helicity map formed from Figure 2b shows left and right-handed domains separated by \( \alpha \) (magenta) and \( \beta \) (green). Additionally, achiral domain walls (black line) also coexist. The resultant triple point topologies formed due to the co-existence of chiral and achiral domain walls are shown in encircled regions. The sense of rotation of these triple point topologies is indicated in the encircled region. (b-c) Possible pairs of triple points. (b) represents triple points with opposite sense of rotation with the point of inversion along \( \beta \), \( \gamma \), and \( \alpha \) (c) represents triple points with the same sense of rotation. The green ticks on the side show what has been observed in experiments.
+
+We can use the qualitative lateral and axial polarization data from 4D-STEM and create a map of the helicity over a large scale using the helicity equation as shown in Figure 3. The red and blue regions in the chirality maps indicate different signs of helicity in the system, making them left-handed and right-handed chiral domains separated by the \( \alpha \) or \( \beta \) domain walls. We find that most of these chiral and achiral domain walls were not visible in the virtual dark field images. Further, we also observe a unique triple point topology at the mesoscale whenever the two types of chiral domain walls meet an achiral domain wall as seen from black-encircled areas, thus forming a quasi-1D defect in the network of chiral and achiral domain walls. These triple points tend to exist in pairs and exhibit a sense of rotation via the transition from \( \alpha \) to \( \beta \) to \( \gamma \) domain wall and vice-versa (Figure 3, S3-S4), similar to what has been observed previously in the trimerized domain walls in hexagonal manganates\(^{34}\), vortex-antivortex phases in intercalated Vander-Waal ferromagnets\(^{6}\), and ferroelectric vortex cores in BiFeO\(_3\)\(^{35}\). The sense of rotation in
+a triple point can be the same or opposite depending on the arrangement of \( \alpha, \beta \) or \( \gamma \) domain walls. Figure 3b,c illustrates this situation. Whenever a pair of domain walls (\( \beta, \gamma \) or \( \alpha, \gamma \) or \( \alpha, \beta \)) break inversion symmetry across \( \alpha, \beta \) or \( \gamma \) domains, we form triple point pairs with opposite sense of rotation. If the inversion symmetry across \( \alpha, \beta \) or \( \gamma \) domains is not broken, we form triple point pairs with the same sense of rotation. The origin of triple point pairs can be understood by the following hypothesis. Consider the example of the first triple point pair in Figure 3b. If this particular type of triple point has to be isolated, then \( \beta \) boundary would infinitely separate the positive and negative chirality regions. If \( \beta \) boundary is not infinite, then it has to meet somewhere an \( \alpha \) boundary to continue separating the positive and negative chirality regions. Now, if a \( \beta \) boundary (change in axial polarization) meets an \( \alpha \) boundary (change in lateral polarization), a \( \gamma \) boundary appears (change in both lateral and axial polarization). In such a scenario, we get another triple point in the vicinity of the first triple point thus explaining the origins of pairs for the majority of cases. \( \alpha/\beta \) boundary could be isolated only in very special circumstances when they are born/die at the surface or they are infinitely long.
+
+(a)
+
+![Diagram showing the arrangement of domain walls and their rotation sense](page_184_672_579_180.png)
+
+(b)
+
+![Diagram showing the arrangement of domain walls and their rotation sense](page_749_672_579_180.png)
+
+(c)
+
+![Diagram showing the arrangement of domain walls and their rotation sense](page_184_872_579_180.png)
+
+(d)
+
+![Diagram showing the arrangement of domain walls and their rotation sense](page_749_872_579_180.png)
+Figure 4. Types of chiral domain walls. (a) Four possible combinations of alternating clockwise/counterclockwise vortices are displaced, Top and bottom cartoons differ by the direction of the axial component of the polarization (red dot and blue cross). Left and right cartons differ by the curl of the polarization. The chirality for each type of domain is indicated by a sketched hand. The possible domain walls between these configurations are marked as type \( \alpha/\alpha' \), \( \beta/\beta' \), and \( \gamma/\gamma' \). \( \alpha/\alpha' \) and \( \beta/\beta' \) domain walls change the chirality at the domain wall. \( \gamma/\gamma' \) domains preserve the chirality. (b-d) Three-dimensional representation of three types of chiral domain walls observed by 4D-STEM measurements.
+
+From a theoretical perspective, there are three symmetry-breaking operations for the formation of domain walls, 1) Change in the lateral polarization direction, 2) Change in axial polarization direction, and 3) Change in the net lateral polarization due to the vortex core shifting away from the center. If we consider all three factors, then we expect to have six types of domain walls as shown in Figure 4. The first pair of chiral domain walls, \( \alpha/\alpha' \), results from a combination of factors 1 and 3. The second chiral domain wall pair, \( \beta/\beta' \), results from a combination of factors 2 and 3. The third achiral domain wall pair results from all three factors. Unfortunately, it is challenging to measure the quantitative net lateral component in the vortices due to the very low sensitivity of electron scattering to sample changes along the beam direction. Due to this, \( \alpha/\alpha' \), \( \beta/\beta' \), and \( \gamma/\gamma' \) are degenerate in this 4D-STEM and thus we observe only three types of domain walls. This has been consistent in multiple such 4D STEM as observed in Figure S3-S4. Future studies, such as depth sectioning or sample tilting experiments, may be able to probe this variation 36.
+
+Discussion
+
+We have unraveled the nanoscale three-dimensional domain wall network in topological polar vortices using quantitative 4D-STEM techniques. The polar vortex oxide superlattice has emergent chirality through different symmetry-breaking operations in the non-zero curl of polarization along with the alternate axial polarization component and vortex buckling-induced net-in plane rotation or lateral polarization component. The interplay of these symmetry-breaking
+operations results in the formation of two types of chiral and an achiral domain wall within tubular vortex topologies. Topology-driven domain wall existence in our work is unusual in comparison to other electrostatic wall conditions-driven domain walls in the ferroic materials \(^{16,34}\). The finite nature of chiral and achiral domain walls results in the formation of unique triple points whenever these domain walls intersect. The most probable existence of these points is in pairs with the same/different handedness, similar to multiferroic materials such as barium hexaferrite \(^{37}\) and BiFeO\(_3\) \(^{35}\). To our best understanding, such an unconventional scenario has not been seen yet in improper ferroics literature. We hope that our studies could inspire future experiments to understand the electronic and magnetic transport at these triple points within the network of chiral and achiral domain walls in polar vortices oxide superlattices.
+
+Materials and Methods
+
+Synthesis: [(PbTiO\(_3\))\(_{16}\)/(SrTiO\(_3\))\(_{16}\)] trilayer with SrRuO\(_3\) buffer layer was synthesized on single-crystalline DyScO\(_3\) (011) substrates via reflection high-energy electron diffraction (RHEED)-assisted pulsed-laser deposition (KrF laser). The PbTiO\(_3\) and the SrTiO\(_3\) layers were grown at 610 °C in 100 mTorr oxygen pressure.
+
+HR-STEM vector mapping: The vector mapping was performed via Gaussian fitting of A site atomic positions on the drift-corrected HR-STEM images \(^{38}\). First, all the A-sites in the drift-corrected images were identified using “Atomap” atom finding tool \(^{39}\). Once the atoms were identified, the atomic planes were divided into different zone axis such as along [001]\(_0\) and [001]\(_0\). The deviation in local A-displacement was found by taking the difference between the local A site displacement and the corresponding average displacement in the local zone axis plane. The displacement vectors were further interpolated into a grid Cartesian grid and then
+differentiated to obtain strain tensor maps. The infinitesimal rotation or the curl of the displacement of vortices was calculated using the following equation:
+
+\[
+\theta = \frac{1}{2} \left( \frac{\partial u}{\partial y} - \frac{\partial v}{\partial x} \right)
+\]
+
+The color bar in the curl of displacement plot is plotted with respect to the mean intensities in the PbTiO$_3$ layer.
+
+**4D STEM analysis:** All 4D STEM experiments were carried out on TEAM I microscope (aberration-corrected Thermo Fisher Scientific Titan 80-300) using a Gatan K3 direct detection camera located at the end of a Gatan Continuum imaging filter. The microscope was operated at 300 kV with a probe current of 100 pA. The probe semi-angle used for the measurement was 2 mrad. Diffraction patterns were collected using a step size of 1 nm with 514 by 399 scan positions. The K3 camera was used in full-frame electron counting mode with a binning of 4 pixels by 4 pixels and a camera length of 1.05 mm. The exposure time for each diffraction pattern was 47 ms. The 4D STEM analysis was carried out using the py4DSTEM modules. Briefly, rotation calibration was performed between the diffraction and image plane to identify the right orientation of the zone axis. For that process, the defocused image in the Ronchigram was compared to the focused scan image and the relative orientation of the two images was compared. Once the zone axis was identified, all the disks in the diffraction pattern at each probe position were fitted using the disk fitting function. The so-called polarization maps were generated by taking the normalized intensity difference between the opposite Friedel pair disks. Subsequently, the signal-to-noise in these polarization maps was improved by using a combination of low-pass and high-pass Gaussian filters. By using a high-pass Gaussian filter, we also minimized the dominating thickness contrast.
+References:
+
+1. Sowa, W. Synthesis of L-glucurone. Conversion of D-glucose into L-glucose. Can. J. Chem. **47**, 3931–3934 (1969).
+2. Nambu, Y. Spontaneous symmetry breaking in particle physics: A case of cross fertilization. *Int. J. Mod. Phys. A* **24**, 2371–2377 (2009).
+3. Spaldin, N. A., Fiebig, M. & Mostovoy, M. The toroidal moment in condensed-matter physics and its relation to the magnetoelectric effect. *J. Phys. Condens. Matter* **20**, 434203 (2008).
+4. Nakata, M., Shao, R.-F., Maclennan, J. E., Weissflog, W. & Clark, N. A. Electric-field-induced chirality flipping in smectic liquid crystals: the role of anisotropic viscosity. *Phys. Rev. Lett.* **96**, 067802 (2006).
+5. Chen, G. *et al*. Unlocking Bloch-type chirality in ultrathin magnets through uniaxial strain. *Nat. Commun.* **6**, 6598 (2015).
+6. Horibe, Y. *et al*. Color theorems, chiral domain topology, and magnetic properties of Fe(x)TaS2. *J. Am. Chem. Soc.* **136**, 8368–8373 (2014).
+7. Qian, Q. *et al*. Chiral molecular intercalation superlattices. *Nature* **606**, 902–908 (2022).
+8. Fischer, P. & Hache, F. Nonlinear optical spectroscopy of chiral molecules. *Chirality* **17**, 421–437 (2005).
+9. Ma, W., Xu, L., Wang, L., Xu, C. & Kuang, H. Chirality-based biosensors. *Adv. Funct. Mater.* **29**, 1805512 (2019).
+10. Cherifi-Hertel, S. *et al*. Non-Ising and chiral ferroelectric domain walls revealed by nonlinear optical microscopy. *Nat. Commun.* **8**, 15768 (2017).
+11. Lee, S. *et al*. Single ferroelectric and chiral magnetic domain of single-crystallineBiFeO3in
+an electric field. *Phys. Rev. B Condens. Matter Mater. Phys.* **78**, (2008).
+
+12. Liu, D. *et al*. Phase-field simulations of vortex chirality manipulation in ferroelectric thin films. *npj quantum mater.* **7**, (2022).
+
+13. Hu, Z.-B. *et al*. An effective strategy of introducing chirality to achieve multifunctionality in rare-earth double perovskite ferroelectrics. *Small Methods* **6**, e2200421 (2022).
+
+14. Wang, Y. J. *et al*. Polar meron lattice in strained oxide ferroelectrics. *Nat. Mater.* **19**, 881–886 (2020).
+
+15. Li, X. *et al*. Atomic-Scale Observations of Electrical and Mechanical Manipulation of Topological Polar Flux Closure. *Proc. Natl. Acad. Sci.* **117**, 18954 (2020).
+
+16. Tang, Y. L. *et al*. Ferroelectrics. Observation of a periodic array of flux-closure quadrants in strained ferroelectric PbTiO$_3$ films. *Science* **348**, 547–551 (2015).
+
+17. Yadav, A. K. *et al*. Observation of polar vortices in oxide superlattices. *Nature* **530**, 198–201 (2016).
+
+18. Zhang, Q. *et al*. Nanoscale Bubble Domains and Topological Transitions in Ultrathin Ferroelectric Films. *Adv. Mater.* **29**, 1702375 (2017).
+
+19. Zhang, Q. *et al*. Deterministic Switching of Ferroelectric Bubble Nanodomains. *Adv. Funct. Mater* **29**, 1808573 (2019).
+
+20. Hadjimichael, M. *et al*. Metal-ferroelectric supercrystals with periodically curved metallic layers. *Nat. Mater.* **20**, 495–502 (2021).
+
+21. Stoica, V. A. *et al*. Optical creation of a supercrystal with three-dimensional nanoscale periodicity. *Nat. Mater.* **18**, 377–383 (2019).
+
+22. Das, S. *et al*. Observation of room-temperature polar skyrmions. *Nature* **568**, 368–372 (2019).
+23. Padraic Shafer, Pablo García-Fernández, Pablo Aguado-Puente, Anoop R. Damodaran, Ajay K. Yadav, Christopher T. Nelson, Shang-Lin Hsu, Jacek C. Wojdeł, Jorge Íñiguez, Lane W. Martín, Elke Arenholz, Javier Junquera, Ramamoorthy Ramesh. Emergent Chirality in the Electric Polarization Texture of Titanate Superlattices. Proc. Natl. Acad. Sci. U. S. A **115**, 915–920 (2018).
+
+24. Susarla, S. *et al.* Atomic scale crystal field mapping of polar vortices in oxide superlattices. *Nat. Commun.* **12**, 6273 (2021).
+
+25. Behera, P. *et al.* Electric field control of chirality. *Sci. Adv.* **8**, eabj8030 (2022).
+
+26. Louis, L., Kornev, I., Geneste, G., Dkhil, B. & Bellaiche, L. Novel complex phenomena in ferroelectric nanocomposites. *J. Phys. Condens. Matter* **24**, 402201 (2012).
+
+27. Nguyen, K. X. *et al.* Transferring orbital angular momentum to an electron beam reveals toroidal and chiral order. *arXiv [cond-mat.mtrl-sci]* (2020).
+
+28. Ophus, C. Four-dimensional scanning transmission electron microscopy (4D-STEM): From scanning nanodiffraction to ptychography and beyond. *Microsc. Microanal.* **25**, 563–582 (2019).
+
+29. Deb, P. *et al.* Imaging polarity in two dimensional materials by breaking Friedel’s law. *Ultramicroscopy* **215**, 113019 (2020).
+
+30. Shao, Y.-T. *et al.* Emergent chirality in a polar meron to skyrmion phase transition. *arXiv [cond-mat.mes-hall]* (2021).
+
+31. Savitzky, B. H. *et al.* Py4DSTEM: Open source software for 4D-STEM data analysis. *Microsc. Microanal.* **25**, 124–125 (2019).
+
+32. Moffatt, H. K. & Ricca, R. L. Helicity and the Călugăreanu invariant. *Proc., Math. phys. sci.* **439**, 411–429 (1992).
+33. Moffatt, H. K. Helicity and singular structures in fluid dynamics. Proc. Natl. Acad. Sci. U. S. A. **111**, 3663–3670 (2014).
+
+34. Holtz, M. E. *et al.* Topological defects in hexagonal manganites: Inner structure and emergent electrostatics. Nano Lett. **17**, 5883–5890 (2017).
+
+35. Balke, N. *et al.* Enhanced electric conductivity at ferroelectric vortex cores in BiFeO3. Nat. Phys. **8**, 81–88 (2012).
+
+36. Zeltmann, S. E. *et al.* Uncovering polar vortex structures by inversion of multiple scattering with a stacked Bloch wave model. *arXiv [cond-mat.mtrl-sci]* (2022).
+
+37. Karpov, D. *et al.* Nanoscale topological defects and improper ferroelectric domains in multiferroic barium hexaferrite nanocrystals. *Phys. Rev. B*. **100**, (2019).
+
+38. Ophus, C., Ciston, J. & Nelson, C. T. Correcting nonlinear drift distortion of scanning probe and scanning transmission electron microscopies from image pairs with orthogonal scan directions. *Ultramicroscopy* **162**, 1–9 (2016).
+
+39. Nord, M., Erik Vullum, P., MacLaren, I., Tybell, T. & Holmestad, R. Atomap - automated analysis of atomic resolution STEM images. *Microsc. Microanal*. **23**, 426–427 (2017).
+
+**Acknowledgments**
+
+All the electron microscopy experiments were carried out at the National Center for Electron Microscopy (NCEM) located in the Molecular Foundry user facility at Lawrence Berkeley National Laboratory. Work at the Molecular Foundry was supported by the Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231. S.S. and R.R. are supported by the DOE Quantum Materials Program. CO acknowledges support from a DOE Early Career Research Award. F.G.-O., P.G.-F., and J.J. acknowledge financial support from Grant No. PGC2018-096955-B-C41 funded by
+MCIN/AEI/10.13039/501100011033 and by ERDF “A way of making Europe,” by the European Union. F.G.-O. acknowledges financial support from Grant No. FPU18/04661 funded by MCIN/AEI/10.13039/501100011033. BHS was supported by the Toyota Research Institute. National Institute of Health Researh UK.
+
+Author contributions: S.S., R.R., and C.O. conceived the idea, and designed the experiments. S.S. analyzed the 4D STEM datasets, made the figures and wrote the initial draft of the manuscript. S.L. performed the 4D STEM experiments. B.H. provided inputs for the scripts of the 4D STEM analysis. F.G.O, P.G. F, and J.J helped in providing intellectual inputs regarding the origin of chiral domain walls and triple points. P.B. and P.E. participated in revising the manuscript. R.R. and C.O. supervised the project.
+
+Competing interests: Authors declare that they have no competing interests.
+
+Data and materials availability: All data are available in the main text or the supplementary materials.
+
+Supplementary Materials.
+
+All the supplementary materials is available in supplemental information.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SI.pdf

02624773f6e6ce56158b53f0fcddcdef4027c464f514126d8dae690687431871/peer_review/peer_review.md

@@ -0,0 +1,386 @@
+Peer Review File
+
+Sphingosine d18:1 Promotes Nonalcoholic Steatohepatitis by Inhibiting Macrophage HIF-2α
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+This first section of this manuscript describes the identification of a specific lipid (sphingosine [So]18:1) whose plasma levels are associated with NASH progression (in humans and mouse models). The second section is a series of mechanistic studies aimed at understanding how So18:1 promotes NASH progression. The conclusion is that So18:1 promotes NLRP3 activation in macrophages through affecting Hif2a levels/activity. There are many aspects of this data set that are interesting and that have the potential to be an important contribution to the field, but I think key additional control experiments are required to support the conclusions.
+
+The data presented in Figure 1 and Supp Figure 1 on the identification of So18:1 is fairly compelling and sets up the subsequent mechanistic work well. However, the subsequent mechanistic studies fall short of proving the thesis of the manuscript. The major concern (across many of the studies) is the nature of the vehicle used when administering So18:1. It is not explicitly stated what this is, but in my opinion the authors should be using the other lipids they ID from the metabolomics, but that are not associated with NASH progression. Only by doing this can they be sure that the effects they see across many of the experiments (Figure 2, Figure 3, Figure 6) are due to specific effects of So18:1. Such evidence would provide a far more compelling argument that So18:1 is a specific driver of NASH progression.
+
+The HIF2a KO and overexpression studies in Figure 3 and 4 are also quite convincing, but how this phenotype is related to So18:1 is not clear. For example, the authors do not provide evidence that the expression of Hif2a is decreased in macrophages from So18:1 treated mice or NASH mouse models. Finally, the suggestion that the So18:1 and Hif2a effects are driven through NLRP3 inflammasome activation are overstated, particularly in vivo.
+
+My specific comments are also outlined below.
+
+Line 32, sphingomyelins are the predominant sphingolipid, please add to your statement.
+
+For the heat maps in Figure 1 it should be stated what the values are, are these z-scores, concentrations, something else, it is not known from the figure or legend. I would personally move the data in Sup Figure 1D, 1E and 1J into the main text, this is important and strong evidence.
+Line 70. I think this is a one-sided interpretation. One could equally postulate that because the liver So18:1 levels decrease over the course of the CDAA-HFD, and that liver enzymes also increase over this time, perhaps the increased plasma levels are the result of hepatocyte damage and release into the circulation. I’m not saying this is the case, but a more balanced interpretation may be warranted.
+
+There is a lack of description of how the sub-class of disease progression (e.g. the data in Supp 1B) data is obtained. Is this a re-classification of the NASH group, I assume so, but please be more descriptive in the text. Also, I would like to see the concentration data (i.e. that shown in Supp Figure 1D) for the other Sa and So species that you show in the heat map. They all look to change to a very similar extent. Also, I would re-format the data in Supp Figure 1B/C to look like Supp Figure 1E. Collectively, these changes will make it easier for the reader to see that while a number of sphingolipids changes in NASH, only So18:1 actually increases during disease progression.
+
+Line 83. The conclusion is not convincing based on the data. Indeed, from a statistical point of view, they are not different. The conclusion should be tempered.
+
+Line 88. ‘accepted’ is rather an odd phrasing. Perhaps just say mice ‘were injected with’.
+
+The data in Figure 2 seem reasonably robust, but I have a number of issues that I think should be addressed. (i) what is the nature of the vehicle? It does not say in the methods what this is; indeed, the methods only state that the So treated group were injected with So everyday, while there is no mention of the vehicle treated group. So, are these mice not being injected at all? If they are, what are they receiving? (ii) Relatedly, I’d very much like to see other lipids that the authors identified in the data in Figure 1 being used as controls – i.e. use lipids that are not associated with NASH progression. This would present a far stronger case that So18:1 has specific effects. (iii) You need to perform the metabolomics to show that your injections are increasing the plasma concentrations of So18:1, and also to what extent, they may be increased to levels far beyond what you see endogenously. Ideally, you want your So18:1 injections to increase plasma So18:1 within pathological levels. (iii) What is happening to lipid levels in the liver following these injections? Collectively, I think several additional experiments are required to convincingly show that So18:1 specifically has the claimed effects.
+
+Figure 3A,B. n numbers are very low, this is obviously from a single experiment. I would like to see a greater sample number used. Additionally, my comments above about the vehicle control apply here also. So, I would like to see the authors using the other So and Sa species they identify as being increased in NASH, but not increasing during disease progression, to show that the specific pro-inflammatory/immune recruitment effects they see are specifically due to So18:1.
+
+Figure 3E-I. Same comment as above, repeat with other lipids to show the effect is specific to So18:1.
+Line 149 seems to be the first mention of the GAN diet. Please define.
+
+Line 148-149. You are not testing this. You are only determining if Hif2a contributes to NASH. You don’t present any evidence that the deletion of Hif2a exacerbates NASH due to inflammasome activation. To support such a claim you should at least show that various inflammasome components, within macrophages, are increased in the Hif2a KO model – yes, il-1b is up a bit but so are lots of other inflammatory markers and this measure is just in a whole liver homogenate, not isolated cells. While inflammatory pathways are activated in Hif2a KO cells, figure 3E shows a large reduction in several markers of M2 macrophages, this is also likely to be a major reason for the potentiated inflammation in the Hif2a KO model – i.e. you are altering the balance of pro/anti-inflammatory macrophages. Furthermore, while you show that So18:1 decreases Hif2a in vitro, you do not show that either So18:1 administration in vivo or the CDAA-HFD diet reduced Hif2a in macrophages in vivo. Nor do you show that blocking NLRP3 inflammasome activity can reverse the effects described in Figure 4.
+
+Line 221, I’m again concerned that a solvent is not the most appropriate control. Your whole thesis is that So18:1, and not other closely related lipids, promote NASH progression. Therefore, the best control would be the other related lipids. If you could show that So16:1 and some of the Sn lipids do not have the same effects as So18:1, this would present a far more compelling argument that So18:1 has specific properties not shared by other related lipid species.
+
+Your central argument is that So18:1 decrease Hif2a protein levels (e.g. figure 2F). Yet looking at the input controls for your IP experiment in Figure 6C, it does not appear to do so. What is the reason for this apparent discrepancy?
+
+Line 229. You somewhat establish this in vitro (see my previous comments about appropriate lipid controls), you do not provide evidence that (a) hif2a is reduced in vivo, either in response to So18:1 or in the NASH model, (b) you don’t demonstrate that the effects of manipulating Hif2a expression (KO/over-expression) on NASH are via effects on NLRP3.
+
+Reviewer #2 (Remarks to the Author):
+
+In the current study, Xia et al. reported the association between plasma sphingosine d18:1 [So(d18:1)] levels and the severity of NASH. They also found that So(d18:1) aggravated the NASH phenotype in mice. Regarding the mechanism, they studied the effects of So(d18:1)-promoted activation of the NLRP3 inflammasome by inhibiting HIF-2a activity in macrophages. They found that So(d18:1) directly bound
+with HIF-2a and inhibited the interaction of HIF-2a and ARNT. This study integrates clinical patient data and animal experiments. In general, this study provides a novel serological indicator for NASH diagnosis and novel targets for NASH treatment. However, there are several critical issues that should be addressed.
+
+1. So(d18:1) increased ALT and AST levels in CADD-HFD-fed mice. Did So(d18:1) directly promote hepatocyte injury during NASH?
+
+2. The possibility that So(d18:1) directly acts on other liver cell types should be discussed.
+
+3. Did So(d18:1) treatment change ceramide and S1P levels in macrophage?
+
+4. In Line 116, the author mentioned So(d18:1) promoted macrophage activation. To support this statement, did So(d18:1) upregulate inflammatory gene expression from the RNA-seq data?
+
+5. Could macrophage-specific HIF-2a overexpression confront the effects of So(d18:1) on NASH in vivo? This is important because So(d18:1) may also act on other cell types, and this experiment could demonstrate the central role of macrophages in this process.
+
+6. Regarding to “HIF-2α plasmid with two proven missense mutations in the pocket which disabled other molecules to bind with HIF-2α”, the mutations need to be described in detail and the related literature should be cited.
+
+7. The direct binding of So(d18:1) with HIF-2a is an important mechanism of this study. However, the authors only provide docking model to predict the direct binding, which is not enough. The evidence of direct binding of So(d18:1) with HIF-2a should be provided (such as SPR or ITC). It will be helpful if the disturbed binding of HIF-2a with mutated HIF-2a is also proven by experiments.
+
+8. From the input of Figure 6C, So(d18:1) did not influence the protein level of HIF-2a. Was HIF-2a exogenous overexpressed in this experiment? If so, this should be clearly described in Figure legends or in method section.
+
+9. Regarding Figure 6F, the figure shows that when HIF-2a was overexpressed, So(d18:1) increased luciferase activity. However, the authors describe the data as “similar to the fluorescence values of the empty plasmid” in line 251. In addition, in the same paragraph, there seems to be 4 groups of treatment in Figure 6F, however, the authors only showed 2 groups. Pleased check this result.
+
+Minor
+
+1. Figure legend of Fig. 2, “Il1b in J, statistical analysis was performed using two tailed Mann-Whitney U-tests”. Il1b is in panel K not J.
+
+2. For the western blot of HIF-2a, why there are two bands in Figure 3F and one band in Figure 6C?
+Reviewer #3 (Remarks to the Author):
+
+In this study, Jialin Xia et al. found that sphingosine d18:1 (So(d18:1)) is elevated in the plasma of NASH patients as well as in the plasma of mice subjected to a NASH mouse model. Authors claim that this So(d18:1) elevation results in liver inflammation and fibrosis and that the pro-fibrotic and pro-inflammatory potential of sphingosine d18:1 rely on its ability to block HIF2a and HIFbeta heterodimerization and therefore HIF2 activity. Along this line authors clearly show that macrophage HIF2 inactivation leads to liver inflammation and fibrosis. These data are interesting but the following comments should be addressed.
+
+1.- In Figure 3F, authors should run in parallel samples of BMDMs obtained from LysM-HIF2aLSL/LSL and HIF2aLysM mice in order to confirm that the two bands shown in the western blot correspond to HIF2a.
+
+2.- Authors should show whether gene expression of Arginase 1, VEGF-a, Spint, Depdc7 and IL-10 are elevated in LysMHIF2aLSL/LSL BMDMs or reduced in LysMHIF2a deficient BMDMs.
+
+3.- Authors show that So(d18:1) blunted HIF2a activity in BMDMs. But is this also in So(d18:1)-treated mice? Authors should try to perform liver immunostaining using antibodies against HIF2 (or HIF2 target genes) in costaining with macrophage markers in order to evaluate whether macrophage HIF2 expression/activity is really reduced in macrophages of So(d18:1)-treated mice in vivo. Or alternatively assess this point in macrophages isolated from the liver of control and So(d18:1)-treated mice. If these experiments reveals a partial decline of HIF2a, authors might assess whether macrophage-HIF2 heterozygous mice also results in increased liver inflammation and fibrosis.
+
+4.- Authors should further assess whether So(d18:1) inhibits specifically HIF2 activity but not HIF1a. Along this line - in Figure 6C - authors should assess whether HIF1a and ARNT heterodimerization is affected by So(d18:1)?. Moreover, in Figure 6B, authors should assess the effect of So(d18:1) on the luciferase activity driven by a pBIND-HIF1a construct.
+
+5.- In page 8, authors mentioned ‘We administered So(d18:1) intraperitoneally to mice for 1 week, results showed that So(d18:1) increased the proportion of liver macrophages among all immune cells (Figure S3C, 3A-C)”. However it is not clear whether Figure S3C refers to HFD-fed mice or So(d18:1)-treated mice. Authors should clarify this point.
+6.- Authors use a GAN diet to assess liver inflammation and fibrosis in macrophages HIF2a-deficient mice. Why authors not use CDAA-HFD diet as in other experiments presented in this study?. Moreover in figure S3A, authors used HFHFD diet. Authors should clarify why they used these different diets.
+
+7.- In Figure 3A, control group correspond to CDAA-HFD-fed mice?
+Response to Reviewers’ Comments
+
+Reviewer #1 (Remarks to the Author):
+
+This first section of this manuscript describes the identification of a specific lipid (sphingosine [So]18:1) whose plasma levels are associated with NASH progression (in humans and mouse models). The second section is a series of mechanistic studies aimed at understanding how So18:1 promotes NASH progression. The conclusion is that So18:1 promotes NLRP3 activation in macrophages through affecting Hif2a levels/activity. There are many aspects of this data set that are interesting and that have the potential to be an important contribution to the field, but I think key additional control experiments are required to support the conclusions.
+
+Response: We appreciate the comments from reviewer #1 on our manuscript. We believe that all concerns raised by the reviewer can be addressed by additional rigorous experimentation and more detailed discussions. We have added the following experiments:
+1) regarding the control for So18:1 administration, we added So(d16:1) as control to the animal experiment in Figures 2, and So(d16:1), So(d20:1) and So(d18:1) as control to the cell experiments in Figure 3 and Figure 6 to illustrate the specific effects of So(d18:1).
+2) We used flow cytometry to detect the decrease of HIF-2a in liver macrophage in vivo under So(d18:1) treatment and in the NASH model. We have increased the sample size for flow cytometry analysis of liver macrophages in Figures 3A and 3B.
+3) We conducted metabolomics studies to detect the plasma concentration of So(d18:1) in mice treated with vehicle and So(d18:1). Besides, we conducted lipidomics testing to determine what changes have occurred in the lipid levels in the liver after the treatment.
+
+The data presented in Figure 1 and Supp Figure 1 on the identification of So18:1 is fairly compelling and sets up the subsequent mechanistic work well. However, the subsequent mechanistic studies fall short of proving the thesis of the manuscript. The major concern (across many of the studies) is the nature of the vehicle used when administering So18:1. It is not explicitly stated what this is, but in my opinion the authors should be using the other lipids they ID from the metabolomics, but that are not associated with NASH progression. Only by doing this can they be sure that the effects they see across many of the experiments (Figure 2, Figure 3, Figure 6) are due to specific effects of So18:1. Such evidence would provide a far more compelling argument that So18:1 is a specific driver of NASH progression.
+
+Response:
+We thank the reviewer believe that our data have the potential to be an important contribution to the field. In this original revision, we mainly used 0.5% CMC-Na
+solution as solvent to make a suspension of So(d18:1) for the administration. We agree with the reviewer that using another lipid from the metabolomics as a control can better highlight the unique role of So(d18:1). We have found that So(d16:1), So(d20:1) and Sa(18:1) changed in the NASH patients’ serum, but their concentrations were not increased as NASH progression (Fig1D, SupFig1C). So we used So(d16:1), So(d20:1) and Sa(d18:1) as control to the cell experiments and used So (d16:1) as control to the animal expriments in Figure 2, 3, and 6 to illustrate the specific effects of So(d18:1). We found only So18:1 promotes macrophages secreting inflammation factors through affecting HIF-2a levels/activity and promotes the progression of NASH, which reinforced the specific role of So(d18:1) in NASH progression.
+
+The HIF2a KO and overexpression studies in Figure 3 and 4 are also quite convincing, but how this phenotype is related to So18:1 is not clear. For example, the authors do not provide evidence that the expression of Hif2a is decreased in macrophages from So18:1 treated mice or NASH mouse models. Finally, the suggestion that the So18:1 and Hif2a effects are driven through NLRP3 inflammasome activation are overstated, particularly in vivo.
+
+Response:
+We used flow cytometry to detect the expression of HIF-2a in macrophages in So(d18:1) treated mice or NASH mouse models. Data shows that the expression of Hif2a decreased in macrophages after So(d18:1) administration or NASH mouse models. The data and figures were added in the article now (Figure 3G and Supplementary Figure 3E). And as to NLRP3 activation, we agree that there will be other machnisms that could affect NASH progression too. Thus, we revised the statement that the So(d18:1) and Hif2a effects are driven only by NLRP3 inflammasome activation in the article.
+
+My specific comments are also outlined below.
+
+1. Line 32, sphingomyelins are the predominant sphingolipid, please add to your statement.
+For the heat maps in Figure 1 it should be stated what the values are, are these z-scores, concentrations, something else, it is not known from the figure or legend. I would personally move the data in Sup Figure 1D, 1E and 1J into the main text, this is important and strong evidence.
+
+Response:
+We’ve added Sphingomyelin as the main sphingolipids, and explained what the mean value of the heat map is in the figure legend of Figure 1. And we have moved the data Sup Figures 1D, 1E, and 1J into the main text in Figure 1E, 1F, 1G.
+
+2. Line 70. I think this is a one-sided interpretation. One could equally postulate that because the liver So18:1 levels decrease over the course of the CDAA-HFD, and that
+liver enzymes also increase over this time, perhaps the increased plasma levels are the result of hepatocyte damage and release into the circulation. I’m not saying this is the case, but a more balanced interpretation may be warranted.
+
+Response:
+    We agree with the reviewer that the statement in line 70 is a one-sided interpretation. We changed the statement (now Line 121) and discussed the reason for the increase of sphingosine in plasma in discussion. In Supplementary figures, we tested the whole liver So(d18:1) content during the NASH model processing (SupFig 1C), and it showed a downward trend within the first two weeks, but no further change in the coming 6 weeks. We agreed that the increased plasma levels may be the result of hepatocyte damage and release into the circulation. Besides, according to existing articles, So(d18:1) is also present in feces (PMID: 32610095), so we infer that the increased So(d18:1) may originate from the intestine, too. These origins of So(d18:1) may exit together.
+
+3. There is a lack of description of how the sub-class of disease progression (e.g. the data in Supp 1B) data is obtained. Is this a re-classification of the NASH group, I assume so, but please be more descriptive in the text. Also, I would like to see the concentration data (i.e. that shown in Supp Figure 1D) for the other Sa and So species that you show in the heat map. They all look to change to a very similar extent. Also, I would re-format the data in Supp Figure 1B/C to look like Supp Figure 1E. Collectively, these changes will make it easier for the reader to see that while a number of sphingolipids changes in NASH, only So18:1 actually increases during disease progression.
+
+Response:
+    Yes, we used the NAS scores of the liver pathological slices to classify the disease progression. We added the description of the subgroup in the method (NAS scoring section) and the figure legend of Supplementary Figure 1. As for the heatmap, we used the intensity of response of each sphingolipids to make that figure. Although they all increased in NASH patients, So(d18:1) is the most significant one. And as for Supp Figure 1B/C (now supplementary Figure 1E-1I), we already redrown them to match the format of Supp Figure 1E (now Figure 1F), indicating that only So(d18:1) truly increases during disease progression.
+
+4. Line 83. The conclusion is not convincing based on the data. Indeed, from a statistical point of view, they are not different. The conclusion should be tempered.
+
+Response:
+    We corrected and tempered the statement on line 83 (now line 132).
+
+5. Line 88. ‘accepted’ is rather an odd phrasing. Perhaps just say mice ‘were injected with’.
+Response:
+    We have changed 'accepted' to 'injected'.
+
+6. The data in Figure 2 seem reasonably robust, but I have a number of issues that I think should be addressed. (i) what is the nature of the vehicle? It does not say in the methods what this is; indeed, the methods only state that the So treated group were injected with So everyday, while there is no mention of the vehicle treated group. So, are these mice not being injected at all? If they are, what are they receiving? (ii) Relatedly, I’d very much like to see other lipids that the authors identified in the data in Figure 1 being used as controls – i.e. use lipids that are not associated with NASH progression. This would present a far stronger case that So18:1 has specific effects. (iii) You need to perform the metabolomics to show that your injections are increasing the plasma concentrations of So18:1, and also to what extent, they may be increased to levels far beyond what you see endogenously. Ideally, you want your So18:1 injections to increase plasma So18:1 within pathological levels. (iii) What is happening to lipid levels in the liver following these injections? Collectively, I think several additional experiments are required to convincingly show that So18:1 specifically has the claimed effects.
+
+Response:
+    (i) In this paper we mainly used 0.5% CMC-Na as the solvent. Control group was injected with the solvent every day. We have added this description in the method.
+    (ii) In Fig1D, we found that So(d16:1) increased in NASH patients serum, but they didn’t show any future accumulation with the disease progression (SupFig 1H-I). In order to verify the unique function of So(d18:1), we chose So(d16:1) as a control lipid to administrate the NASH mice. As the figures showed in Fig2 and SupFig2, So(d16:1) injection didn’t exacerbate NASH in mice.
+    (iii) We conducted metabolomics studies to detect whether So18:1 injection increases plasma concentration within the pathological range as expected. Data is showed in SupFig 2A.
+    (iii) After injection of vehicles and So(d18:1), we conducted lipidomics testing to determine what changes have occurred in the lipid levels in the liver (Response Figure 1).
+Response Figure 1: A, PLS-DA analysis of sphingolipids in liver of CDAA-HFD mice treated with vehicle or So(d18:1). B, VIP score plot of the difference sphingolipids between the two groups.
+
+7. Figure 3A,B. n numbers are very low, this is obviously from a single experiment. I would like to see a greater sample number used. Additionally, my comments above about the vehicle control apply here also. So, I would like to see the authors using the other So and Sa species they identify as being increased in NASH, but not increasing during disease progression, to show that the specific pro-inflammatory/immune recruitment effects they see are specifically due to So18:1.
+
+Response:
+    We have increased the sample size for flow cytometry analysis of liver macrophages in Figures 3A and B. And we supplied So(d16:1) as control in this experiment.
+
+8. Figure 3E-I. Same comment as above, repeat with other lipids to show the effect is specific to So18:1.
+
+Response:
+    We have supplemented other So and Sa species (So(d16:1), So(d20:1), Sa(d18:1)) in Figure 3E-I as controls (now figure 3E, supplementary figure 3G, 3H). As these So and Sa species (So(d16:1), So(d20:1), Sa(d18:1)) didn’t influence the HIF-2a downstream genes and the protein level of IL-1β and IL-18, we didn’t detect their influence on HIF-2a and caspase-1 protein further.
+
+9. Line 149 seems to be the first mention of the GAN diet. Please define.
+
+Response:
+    GAN diet is Gubra Amylin NASH (GAN) diet, which is a high fat, high cholesterol, and high fructose feed (40% fat powered, 20% fructose, and 2% cholesterol), is fed to mice for more than 24 weeks (generally, steatosis begins at 16 weeks, NASH forms at 24 weeks, and fibrosis is induced at 26-32 weeks). We have added the definition of
+GAN diet on Line 149 (now it is Line 213).
+
+10. Line 148-149. You are not testing this. You are only determining if Hif2a contributes to NASH. You don’t present any evidence that the deletion of Hif2a exacerbates NASH due to inflammasome activation. To support such a claim you should at least show that various inflammasome components, within macrophages, are increased in the Hif2a KO model – yes, il-1b is up a bit but so are lots of other inflammatory markers and this measure is just in a whole liver homogenate, not isolated cells. While inflammatory pathways are activated in Hif2a KO cells, figure 3E shows a large reduction in several markers of M2 macrophages, this is also likely to be a major reason for the potentiated inflammation in the Hif2a KO model – i.e. you are altering the balance of pro/anti-inflammatory macrophages. Furthermore, while you show that So18:1 decreases Hif2a in vitro, you do not show that either So18:1 administration in vivo or the CDAA-HFD diet reduced Hif2a in macrophages in vivo. Nor do you show that blocking NLRP3 inflammasome activity can reverse the effects described in Figure 4.
+
+Response:
+We changed the description of lines 148 to 149 (now line 212) as “To investigate whether HIF-2α in macrophages can influence NASH disease progression”. Our previous work has demonstrated that the absence of HIF-2a leads to the activation of macrophage inflammasomes (PMID: 34433035), and many other reports have also found that the activation of NLRP3 inflammasomes is an important mechanism for the occurrence and development of NASH. We detected the levels of IL-1β and IL-18 in macrophages of HIF2aΔLysm, which is consistent with our previous report (PMID: 34433035), indicating an increase in IL-1β levels in macrophages of HIF2a KO. However, we also agree that NLRP3 inflammasome activation induced by reduced HIF-2a may be only one of the mechanisms by which HIF-2a deficiency leads to NASH. We would therefore modify the state of the conclusions and add discussions.
+As to whether So(d18:1) also reduces HIF-2a in vivo, we added flow cytometry analysis of liver macrophages to verify (Fig 3G, SupFig 3E). Data shows that HIF-2a also reduced in liver macrophages.
+
+![Bar graphs showing quantification of IL-1β and IL-18 secretion from Hif2αfl/fl and Hif2αΔLysm BMDMs transfected with two independent Nlrp3-](page_1016_1342_377_246.png)
+
+Response Figure 2: A-B, Quantification of IL-1β (A) and IL-18 (B) secretion from Hif2αfl/fl and Hif2αΔLysm BMDMs that were transfected with two independent Nlrp3-
+targeting siRNAs or a nontargeting siRNA after treatment with LPS and nigericin (n = 6). C, Representative immunoblot analysis of caspase-1 and IL-1β from Hif2αnull and Hif2αΔLysm BMDMs that were stimulated with LPS and nigericin (n = 3).
+
+11. Line 221, I’m again concerned that a solvent is not the most appropriate control. Your whole thesis is that So18:1, and not other closely related lipids, promote NASH progression. Therefore, the best control would be the other related lipids. If you could show that So16:1 and some of the Sn lipids do not have the same effects as So18:1, this would present a far more compelling argument that So18:1 has specific properties not shared by other related lipid species.
+
+Response:
+    We have added other sphingolipids (So(d16:1), So(d20:1), Sa(d18:1)) as controls to repeat the HIF-2α Luciferase assay for transcriptional activity test.
+
+12. Your central argument is that So18:1 decrease Hif2a protein levels (e.g. figure 2F). Yet looking at the input controls for your IP experiment in Figure 6C, it does not appear to do so. What is the reason for this apparent discrepancy?
+
+Response:
+    Because the Hif2a protein in Input (now Figure 6D) is exogenously overexpression. We transfected the HIF2a overexpression plasmid ( oxygen stable HIF-2α triple mutant (HIF-2αTM) plasmid) into the cells, so that there is stable and large amount of HIF2a present in the cell, which avoids the influence of HIF2a degradation on the experiment, and also amplifies the effect and makes the inhibition of HIF-2α binding with ARNT more obvious. However, in Figure 3F, HIF2a is endogenous and unstable. So(d18:1) inhibit the binding of HIF-2α to ARNT, which impedes HIF-2α entry into the nucleus for transcriptional regulation. HIF-2α that remains in the cytoplasm is very easily hydrolysed and therefore protein levels are reduced.
+
+13. Line 229. You somewhat establish this in vitro (see my previous comments about appropriate lipid controls), you do not provide evidence that (a) hif2a is reduced in vivo, either in response to So18:1 or in the NASH model, (b) you don’t demonstrate that the effects of manipulating Hif2a expression (KO/over-expression) on NASH are via effects on NLRP3.
+
+Response:
+    (a) We used flow cytometry to detect the decrease of hif2a in vivo, whether for So18:1 or in the NASH model (Fig3G, S3E). (b) Yes, we agree that NLRP3 may not the only pathway which lead Hif2a KO to NASH. We have therefore modified the state about NLRP3 inflammation in the conclusion and discussion. We mainly focused on the impact of macrophage hif2a KO/overexpression on the progression on NASH in this paper.
+Reviewer #2 (Remarks to the Author):
+
+In the current study, Xia et al. reported the association between plasma sphingosine d18:1 [So(d18:1)] levels and the severity of NASH. They also found that So(d18:1) aggravated the NASH phenotype in mice. Regarding the mechanism, they studied the effects of So(d18:1)-promoted activation of the NLRP3 inflammasome by inhibiting HIF-2a activity in macrophages. They found that So(d18:1) directly bound with HIF-2a and inhibited the interaction of HIF-2a and ARNT. This study integrates clinical patient data and animal experiments. In general, this study provides a novel serological indicator for NASH diagnosis and novel targets for NASH treatment. However, there are several critical issues that should be addressed.
+
+Response: We appreciate the comments from reviewer #2 on our manuscript. We believe that all concerns raised by the reviewer can be addressed by additional rigorous experimentation and more detailed discussions. We have added the following experiments: We used LysM-HIF-2aLSL/LSL mouse experiments to test whether macrophage specific HIF-2a overexpression can counteract the effect of So(d18:1) on NASH in vivo. And we treated primary hepatocytes and LX-2 cell with So(d18:1) in vitro. Moreover, we conducted SPR experiments to verify the evidence of direct binding of So(d18:1) with HIF-2a. Besides, we have detected the level of ceramide and S1P levels after treatment of So(d18:1) on macrophages and analyzed the inflammatory gene expression from the RNA-seq data.
+
+1. So(d18:1) increased ALT and AST levels in CADD-HFD-fed mice. Did So(d18:1) directly promote hepatocyte injury during NASH?
+
+Response:
+    We used 20 \( \mu \)M So(d18:1) to stimulate the primary hepatocytes directly and found that So(d18:1) didn’t have a brutal ability to damage hepatocytes (SupFig 3I).
+
+2. The possibility that So(d18:1) directly acts on other liver cell types should be discussed.
+
+Response:
+    We used 20 \( \mu \)M So(d18:1) to stimulate the primary hepatocytes and LX-2 cells, which are derived from human hepatic stellate cells, directly and found that So(d18:1) didn’t have a brutal ability to damage hepatocytes (SupFig 3I), neither induced fibrosis in LX-2 cells (SupFig 3J). So(d18:1) may still have possibility directly on other liver cell types, such as endothelial cells, bile duct cells, T cells, B cells, dendritic cells, and so on. However, macrophage-specific HIF-2a overexpression confronted the effects of So(d18:1) largely (Figure 5), which could demonstrate the central role of macrophages in this process. We have added these into discussion.
+3. Did So(d18:1) treatment change ceramide and S1P levels in macrophage?
+
+Response:
+    So(d18:1) treatment didn’t change the level of ceramide, while increased the concentration of S1P significantly in macrophages (Response Figure 3A-B). Besides, So(d18:1) treatment increased the concentration of So(d18:1) in macrophages as expected (Response Figure 3C). Furthermore, when we used a S1P synthesis inhibitor SKI178 to avoid the change of S1P totally, So(d18:1) treatment still inhibits transcriptional regulation function of HIF-2a significantly (Response Figure 3D). These results suggest that the function of treatment of So(d18:1) on macrophages mainly comes from So(d18:1) itself rather than the S1P or ceramide.
+
+![Bar graphs showing ceramide, S1P, and So(d18:1) concentrations in macrophages at different time points, and HIF-2α luciferase activity assays](page_346_563_1092_496.png)
+
+Response Figure 3: A-C, The concentration of total ceramide, S1P, So(d18:1) in macrophages after treatment of So(d18:1) at different time point. D, HRE-based luciferase assay in HEK293T transfected with HIF-2a, followed by treatment of PT-2385, So(d18:1) and So(d18:1) with S1P synthesis inhibitor SKI178.
+
+4. In Line 116, the author mentioned So(d18:1) promoted macrophage activation. To support this statement, did So(d18:1) upregulate inflammatory gene expression from the RNA-seq data?
+
+Response:
+We analyzed the expression of inflammatory genes in RNA seq data and found that So(d18:1) didn’t upregulate inflammatory gene expression of macrophages from the RNA-seq data. The result suggests that so (d18:1) didn’t directly affect transcription of inflammation factors. BP pathway enrichment revealed So(d18:1) inhibits HIF-2α-regulated signalling pathway (Figure 3C). So (d18:1) inhibits HIF2a downstream genes M2 macrophages marker genes, Arg1, Vegf, and so on, which may induce macrophages transition to an inflammatory state. On the other hand, So(d18:1) promotes synthesis and secretion of inflammatory factors but not transcription of inflammatory factors through HIF2a which promotes macrophage fatty acid oxidation, thereby promoting the second signal of inflammasomes (PMID: 34433035).
+
+![RNA-seq data scatter plot showing transcripts per million (Tpm) of inflammatory genes from the RNA-seq data.](page_370_670_698_312.png)
+
+Response Figure 4: transcripts per million (Tpm) of inflammatory genes from the RNA-seq data.
+
+5. Could macrophage-specific HIF-2a overexpression confront the effects of So(d18:1) on NASH in vivo? This is important because So(d18:1) may also act on other cell types, and this experiment could demonstrate the central role of macrophages in this process.
+
+Response:
+We used LysMHif2aLSL/LSL mouse experiments to test whether macrophage specific HIF-2a overexpression can counteract the effect of So(d18:1) on NASH in vivo in Fig5 and SupFig5. Specifically, the mice were divided into four groups: Hif2a+/+, Hif2a+/+ + So(d18:1), LysMHif2aLSL/LSL and LysMHif2aLSL/LSL+So(d18:1), all fed with CDAA-HFD. The level of ALT, AST, the lobular inflammation score, the fibrosis area of HE and sirus staning and the inflammation and fibrosis genes, all showed that macrophage-specific HIF-2a overexpression confronted the effects of So(d18:1) on NASH in vivo.
+
+6. Regarding to “HIF-2α plasmid with two proven missense mutations in the pocket which disabled other molecules to bind with HIF-2α”, the mutations need to be described in detail and the related literature should be cited.
+
+Response:
+We cited relevant literature in these part. The Missense mutations are S304M and G323E, which disabled the binding of other molecules, such as 1,3-diaminopropane and PT2385, with HIF-2a (PMID: 31708445; PMID: 27595394; PMID: 27595393).
+7. The direct binding of So(d18:1) with HIF-2a is an important mechanism of this study. However, the authors only provide docking model to predict the direct binding, which is not enough. The evidence of direct binding of So(d18:1) with HIF-2a should be provided (such as SPR or ITC). It will be helpful if the disturbed binding of HIF-2a with mutated HIF-2a is also proven by experiments.
+
+Response:
+We conducted SPR experiments to verify the binding of So (d18:1) with HIF-2a. The result showed that So(d18:1) had a direct binding with HIF-2a and \( K_D \) (binding affinity) values were determined as 19.51 \( \mu \)M (Response Figure 5A). We also conduct SPR experimen to detect the binding of So(d18:1) with mutated HIF-2a （Response Figure 5B). The result showed that So(d18:1) had no binding with mutated HIF-2a which indicated that the S304M and G323E mutation of HIF-2a disturbed the binding of So(d18:1) with HIF-2a.
+
+![Surface plasmon resonance (SPR) indicated binding of So(d18:1) to the HIF2a. Surface plasmon resonance (SPR) indicated no binding of So(d18:1) to the mutant HIF2a.](page_324_670_1097_324.png)
+
+Response Figure 5: A, Surface plasmon resonance (SPR) indicated binding of So(d18:1) to the HIF2a. B, Surface plasmon resonance (SPR) indicated no binding of So(d18:1) to the mutant HIF2a.
+
+8. From the input of Figure 6C, So(d18:1) did not influence the protein level of HIF-2a. Was HIF-2a exogenous overexpressed in this experiment? If so, this should be clearly described in Figure legends or in method section.
+
+Response:
+We clearly described that HIF-2a is exogenous overexpression in the Figure legends of Figure 6C (now Figure 6D).
+
+9. Regarding Figure 6F, the figure shows that when HIF-2a was overexpressed, So(d18:1) increased luciferase activity. However, the authors describe the data as “similar to the fluorescence values of the empty plasmid” in line 251. In addition, in the same paragraph, there seems to be 4 groups of treatment in Figure 6F, however, the authors only showed 2 groups. Pleased check this result.
+
+Response: In Figure 6F (now Figure 6G), The Y axis of this figure represents the ratio
+of fluorescence values between overexpressing plasmids to empty plasmids. In the control group, due to the inhibition of cpt1a by overexpressed HIF-2a, the fluorescence value of HIF-2a overexpression plasmid group was lower than that of the empty plasmid group, so their ratio was close to 0.5; When administered with So(d18:1), the effect of overexpressed HIF-2a was inhibited by So(d18:1). Therefore, the fluorescence value of the overexpressed plasmid group is approximately equal to the fluorescence of the empty plasmid group, so their ratio is close to 1. As the Y axis of this figure represents the ratio of two groups, so the figure only showed 2 groups rather than 4 groups.
+
+Minor
+1. Figure legend of Fig. 2, “II1b in J, statistical analysis was performed using two tailed Mann-Whitney U-tests”. II1b is in panel K not J.
+
+Response:
+    We have changed J to K.
+
+2. For the western blot of HIF-2a, why there are two bands in Figure 3F and one band in Figure 6C?
+
+Response:
+    Figure 3F showes endogenous HIF-2α protein, which is more delicate and can be influenced by post-translational modifications, post-translational cleavage and relative charge. Figure 6C (now Figure 6D) shows exogenous HIF-2α protein derived from the overexpression plasmid, oxygen-stable HIF-2α triple mutant (HIF-2αTM) plasmid, which is more stable and hard to be degrade.
+
+Reviewer #3 (Remarks to the Author):
+
+In this study, Jialin Xia et al. found that sphingosine d18:1 (So(d18:1)) is elevated in the plasma of NASH patients as well as in the plasma of mice subjected to a NASH mouse model. Authors claim that this So(d18:1) elevation results in liver inflammation and fibrosis and that the pro-fibrotic and pro-inflammatory potential of sphingosine d18:1 rely on its ability to block HIF2a and HIFbeta heterodimerization and therefore HIF2 activity. Along this line authors clearly show that macrophage HIF2 inactivation leads to liver inflammation and fibrosis. These data are interesting but the following comments should be addressed.
+
+Response: We appreciate the comments from reviewer #3 on our manuscript. We believe that all concerns raised by the reviewer can be addressed by additional rigorous experimentation and more detailed discussions. We have added the following experiments: We run in parallel samples of BMDMs obtained from LysMHIF2aLSL/LSL and HIF2aΔLysM mice to confirm that the two bands shown in the western blot correspond to HIF2a. And we detected gene expression of Arginase 1, VEGF-a, Spint,
+Depdc7 and IL-10 in LysMHif2aLSL/LSL BMDMs. Moreover, we performance flow cytometry to detect the HIF2a levels in liver macrophages. Besides, we used luciferase assay to illustrate that So(d18:1) inhibits specifically HIF2a activity but not HIF1a. We evaluated whether HIF1a and ARNT heterodimerization are affected by So(d18:1) by co-IP and pBIND-HIF1a construct.
+
+1.- In Figure 3F, authors should run in parallel samples of BMDMs obtained from LysM-HIF2aLSL/LSL and HIF2aLysM mice in order to confirm that the two bands shown in the western blot correspond to HIF2a.
+
+Response: We run in parallel samples of BMDMs obtained from LysMHIF2aLSL/LSL and HIF2aΔLysM mice to confirm that the two bands shown in the western blot correspond to HIF2a.
+
+![Representative immunoblot analysis of HIF-2α of BMDMs from stimulated with vehicle and So(d18:1) isolated from wild-type mice and BMDMs isolated from LysMHif2aLSL/LSL and Hif2aΔLysM BMDMs.](page_495_613_495_120.png)
+
+Response Figure 6: representative immunoblot analysis of HIF-2α of BMDMs from stimulated with vehicle and So(d18:1) isolated from wild-type mice and BMDMs isolated from LysMHif2aLSL/LSL and Hif2aΔLysM BMDMs.
+
+2.- Authors should show whether gene expression of Arginase 1, VEGF-a, Spint, Depdc7 and IL-10 are elevated in LysMHIF2aLSL/LSL BMDMs or reduced in LysMHIF2a deficient BMDMs.
+
+Response:
+The gene expression of Arginase 1, VEGF-a, Spint, Depdc7 and IL-10 were elevated in LysMHif2aLSL/LSL BMDMs. They all increased because of Hif2a overexpression.
+
+![Relative mRNA levels of Hif2a and its downstream target genes in Hif2a+/+ or LysMHif2aLSL/LSL BMDMs.](page_495_1097_495_246.png)
+
+Response Figure 7: Relative mRNA levels of Hif2a and its downstream target genes in Hif2a+/+ or LysMHif2aLSL/LSL BMDMs.
+
+3.- Authors show that So(d18:1) blunted HIF2a activity in BMDMs. But is this also in
+So(d18:1)-treated mice? Authors should try to perform liver immunostaining using antibodies against HIF2 (or HIF2 target genes) in costaining with macrophage markers in order to evaluate whether macrophage HIF2 expression/activity is really reduced in macrophages of So(d18:1)-treated mice in vivo. Or alternatively assess this point in macrophages isolated from the liver of control and So(d18:1)-treated mice. If these experiments reveals a partial decline of HIF2a, authors might assess whether macrophage-HIF2 heterozygous mice also results in increased liver inflammation and fibrosis.
+
+Response:
+We performance flow cytometry to detect the HIF2a levels in liver macrophages (Fig 3G and SupFig 3E). These suggested that macrophage HIF2 expression/activity is really reduced in macrophages of So(d18:1)-treated or CDAA-HFD fed mice in vivo. As HIF2a^{LysM} mice have a partial decline of HIF2a in BMDMs (Response Figure 6), so we used the HIF2a^{LysM} mice to detected the liver inflammation and fibrosis (Figure 4).
+
+4.- Authors should further assess whether So(d18:1) inhibits specifically HIF2 activity but not HIF1a. Along this line - in Figure 6C - authors should assess whether HIF1a and ARNT heterodimerization is affected by So(d18:1)?. Moreover, in Figure 6B, authors should assess the effect of So(d18:1) on the luciferase activity driven by a pBIND-HIF1a construct.
+
+Response:
+We used luciferase assay to illustrate that So(d18:1) inhibits specifically HIF2a activity but not HIF1a (Fig 6B). We evaluated whether HIF1a and ARNT heterodimerization are affected by So(d18:1) by co-IP (Fig 6E) and pBIND-HIF1a construct (Response Figure 8).
+
+![Schematic and experimental representation of TAY-cyclo or So(d18:1) mediated HIF-1a-ARNT interaction by pG5GAL4 luciferase assay followed by pBINDHIF-1a and pACT-ARNT transfection of HEK293T cells (n = 6).](page_1012_1347_377_181.png)
+
+Response Figure 8: Schematic and experimental representation of TAY-cyclo or So(d18:1) mediated HIF-1a-ARNT interaction by pG5GAL4 luciferase assay followed by pBINDHIF-1a and pACT-ARNT transfection of HEK293T cells (n = 6).
+
+5.- In page 8, authors mentioned ‘We administered So(d18:1) intraperitoneally to mice for 1 week, results showed that So(d18:1) increased the proportion of liver macrophages among all immune cells (Figure S3C, 3A-C)”. However it is not clear whether Figure S3C refers to HFD-fed mice or So(d18:1)-treated mice. Authors should clarify this
+point.
+
+Response:
+    Figure S3C is the gating strategy of liver macrophages from chow diet mice treated with vehicle. Figure3A-B showed the flow cytometric and statistical analysis of chow-diet fed mice treated with control, So(d16:1) and So(d18:1). We have clarified this point in the Figure legend of Figure S3C and Figure3A-B, as well as in the text of Line 169-172 in page 9.
+
+6.- Authors use a GAN diet to assess liver inflammation and fibrosis in macrophages HIF2a-deficient mice. Why authors not use CDAA-HFD diet as in other experiments presented in this study?. Moreover in figure S3A, authors used HFHFD diet. Authors should clarify why they used these different diets.
+
+Response:
+    CDAA-HFD food induced NASH inflammation and fibrosis are more severe. Since we found in Fig1 that So (d18:1) may be more closely related to inflammation and fibrosis compared to fat accumulation, we chose CDAA-HFD to highlight the phenotypic changes of inflammation and fibrosis. GAN diet has a longer processing time and is more in line with the disease development process of NASH in humans. Therefore, we chose this model to observe whether the impact of HIF-2a on NASH disease progression only affects inflammation and fibrosis, without affecting fat accumulation. But now, to further emphasise that macrophage-specific HIF-2a overexpression can resist NASH, we switched to a more intense CDAA-HFD.
+
+7.- In Figure 3A, control group correspond to CDAA-HFD-fed mice?
+
+Response:
+    In Figure 3A, the control group correspond to chow-diet fed mice. We have added the descriptions in the figure legend of Figure 3A.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The authors have done an excellent job of addressing my comments. They have added in the essential controls that I felt were necessary to more fully understand the specificity of So(18:1). The new data clearly show that indeed So18:1 has specific effects not shared by related lipids, thus supporting the main conclusion of the manuscript. They have also added essentially all the new pieces of data/experiments I suggested and made various textual changes, all of which strengthens the manuscript.
+
+There remain a few odd phrases in the manuscript, for example line 407 describing the potential of So18:1 to causes cell death directly – “brutal ability”. The authors may wish to carefully check through the manuscript for other textual/grammatical issues. In the main the language is fine, just a few minor things could be corrected.
+
+Reviewer #2 (Remarks to the Author):
+
+The authors have addressed my concerns.
+
+Reviewer #3 (Remarks to the Author):
+
+Authors have addressed most of my comments. However, regarding my original comment #3, authors should demonstrate more convincingly that So(d18:1) reduces HIF2a expression in macrophages in vivo. Moreover, I am including a minor comment related to my original comments #1.
+
+Regarding my original comment #1.
+
+In the figure legend of the figure shown in the response letter, BMDMs isolated from LysMHif2αLSL/LSL and Hif2αΔLysM were treated with vehicle?, is this correct? If so, this should be included in the figure legend. Moreover, this figure might be included in the Supplementary information section.
+Regarding my original comment #3.
+
+In Figure 3G and S3E, authors have assessed HIF2a expression by flow cytometry in macrophages isolated from So(d18:1)-treated or CDAA-HFD fed mice in vivo.
+
+However, authors should include - as positive and negative control - BMDMs isolated from LysMHif2αLSL/LSL and Hif2αΔLysM mice to confirm that signal detected by flow cytometry really corresponds to mouse HIF2a.
+
+Alternatively, authors could use western blot instead of flow cytometry.
+
+On the other hand, Figure legend 3G should be corrected to ‘Flow cytometric and statistical analysis of HIF-2a in liver macrophages…..’ instead of ‘Flow cytometric and statistical analysis of HIF-1a in liver macrophages…..’
+
+Finally, in the Figure legend S3E, 'chow dier' should be corrected to 'chow diet'.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The authors have done an excellent job of addressing my comments. They have added in the essential controls that I felt were necessary to more fully understand the specificity of So(18:1). The new data clearly show that indeed So18:1 has specific effects not shared by related lipids, thus supporting the main conclusion of the manuscript. They have also added essentially all the new pieces of data/experiments I suggested and made various textual changes, all of which strengthens the manuscript.
+
+There remain a few odd phrases in the manuscript, for example line 407 describing the potential of So18:1 to causes cell death directly – “brutal ability”. The authors may wish to carefully check through the manuscript for other textual/grammatical issues. In the main the language is fine, just a few minor things could be corrected.
+Response:
+    We have changed the “brutal ability” to “didn’t directly promote hepatocyte death”. We have carefully check through the manuscript for other textual/grammatical issues and corrected a few minor things.
+
+Reviewer #2 (Remarks to the Author):
+
+The authors have addressed my concerns.
+
+Reviewer #3 (Remarks to the Author):
+
+Authors have addressed most of my comments. However, regarding my original comment #3, authors should demonstrate more convincingly that So(d18:1) reduces HIF2a expression in macrophages in vivo. Moreover, I am including a minor comment related to my original comments #1.
+
+Regarding my original comment #1.
+
+In the figure legend of the figure shown in the response letter, BMDMs isolated from LysMHif2αLSL/LSL and Hif2αΔLysM were treated with vehicle?, is this correct? If so, this should be included in the figure legend. Moreover, this figure might be included in the Supplementary information section.
+Response:
+    Yes, BMDMs isolated from LysMHif2α^{LSL/LSL} and Hif2α^{ΔLysM} were treated with vehicle. We have included this in the figure legend of . And we have included the figure in the Supplementary information section.
+
+Regarding my original comment #3.
+In Figure 3G and S3E, authors have assessed HIF2a expression by flow cytometry in macrophages isolated from So(d18:1)-treated or CDAA-HFD fed mice in vivo. However, authors should include - as positive and negative control - BMDMs isolated from LysMHif2αLSL/LSL and Hif2αΔLysM mice to confirm that signal detected by flow cytometry really corresponds to mouse HIF2a.
+Alternatively, authors could use western blot instead of flow cytometry.
+
+Response:
+    We have included BMDMs isolated from LysMHif2αLSL/LSL and Hif2αΔLysM mice as positive and negative control to confirm that signal detected by flow cytometry really corresponds to mouse HIF2a.
+
+On the other hand, Figure legend 3G should be corrected to ‘Flow cytometric and statistical analysis of HIF-2a in liver macrophages…….’ instead of ‘Flow cytometric and statistical analysis of HIF-1a in liver macrophages…….’
+
+Response:
+    We have corrected the Figure legend 3G as “Flow cytometric and statistical analysis of HIF-2a in liver macrophages”.
+
+Finally, in the Figure legend S3E, 'chow dier' should be corrected to 'chow diet'.
+
+Response:
+    We have corrected the Figure legend S3E 'chow dier' to 'chow diet'.
+REVIEWERS' COMMENTS
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have satisfactorily addressed my final concerns.

027b827c3b80c86306a8c299d162b6cb2bb24a38c0ffa3c834d579a1e6c6338c/peer_review/peer_review.md

@@ -0,0 +1,282 @@
+Peer Review File
+
+Toll-like receptor mediated inflammation directs B cells towards protective antiviral extrafollicular responses
+
+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/.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+This manuscript by Lam and Baumgarth investigates the signals that control the generation of the EF versus GC B cell response following infection or vaccination. This is a question of high import given that early control of infection can determine the extent of pathology and severity of illness. The authors reveal a regulatory role of TLR mediated signals in determining the fate decision of activated B cells, with a sustained innate signal promoting differentiation along the EF pathway resulting in early production of antibody capable of promoting clearance. The study includes a careful timecourse of EF versus GC cell and the source of virus-specific antibody. This is an important study that significantly extends our understanding of B cell differentiation. Overall the conclusions are well supported although at times conclusions are based on results that do not appear to be significant (detailed below).
+
+Major comments:
+1. One would expect an increase in HA-specific B cells in Fig 3e given the substantial increase in GC cells in the vaccinated animals, but this seems not to be the case. Could the authors comment on the specificity of these new GC cells?
+2. What percentage of cells fall into the positive quadrant in non vaccinated mice (Fig. 3d)? As there is no increase in HA-sp B cells from d3 to d7, without a control, can one be certain these are in fact HA specific cells being analyzed?
+3. The data in SF2 appear to be total and not HA-specific. Without that, it is difficult to draw conclusions about the HA response.
+4. In SF3, there is a trend towards increased EF in TLR7KO mice. The low number of animals and variability in the TLR7 KO mice make achieving significance challenging. If the two types of EF cells are pooled for analysis, does the picture become clearer?
+5. In Fig 4b, did the DKO serum provide any protection? Was a control serum transfer experiment performed?
+6. The data in SF5 are challenging for comparison between conditions as there are limited times when the same dose is used in combination versus individual stimulations. Nonetheless, one can compare the 1ug/ml anti-IgM and 1ug/ml LPS in this way. When doing so it is hard to see the how the combination strongly supported viability (SF5a) or proliferation (SF5b) compared to the LPS alone as is stated. They appear similar.
+7. Are the reductions in Ki67+ cells in SF7 significant? If not this limits the conclusion that can be drawn in lines 247-49.
+8. In lines 264-65 authors should state trend as it does not appear to significantly increase. For many of the animals there is a very modest increase and one TKO has a major decrease. Thus conclusions should be drawn with care.
+9. Additional discussion of the TLR ligand independent requirement for TLR pathway in B cell activation would be welcome.
+
+Minor comments
+1. Line 81: believe meant vesicular not vaccinia
+2. Line 127: add virus after influenza
+3. Line 197: EFR-derived serum is a bit unclear, suggest “serum from d10 animals wherein antibody is predominantly EFR derived” if this is what is meant
+4. Line 211: the chimera to which the authors are referring should be noted prior to BMC
+5. Line 370: induce for indue
+6. The authors are asked to increase axis labels and percentage values in flow plot font size where possible as many are quite hard to read.
+7. Consider keeping nomenclature similar- “fold difference” axis label Fig 6e
+Reviewer #2 (Remarks to the Author):
+
+The paper by Lam et al explores the mechanisms underlying the formation of extrafollicular plasmablasts (EFRs) and short term humoral memory toward influenza. The authors provide support for a correlation between the formation of EFRs and protective properties of serum at early time points after antigen challenge. Using a series of KO models the authors determine that a key component for the formation of EFRs is Toll like receptor signaling that they find stimulate IRF4 transcription in the activated B-cell. They conclude that inflammatory signals impact the fate of activated B-cells to generate efficient short term response to antigen.
+
+This is in many regards an interesting report representing an impressive amount of work in different model systems. The large number of models and experimental setups explored does, however, become a liability as parts of the data are of limited quality. The authors base many of their conclusions on data collected from few animals creating a challenge in interpretation of the results. This problem is aggravated by that substantial portions of the data are presented as relative controls, in some cases wt, and other cases as compared to non-stimulated cells. It is difficult to fully delineate how this normalization was made. Furthermore, the authors does at times ignore significant differences in their data when they and put forward differences that are not indicated as significant. This creates a challenge to read and validate the data in a proper manner making it difficult to fully appreciate the reported findings and to understand the relevance of the findings.
+
+Specific comments.
+1: The authors should go over their data and decide what parts that are conclusive and where there is a need to repeat the experiments, alternatively take some data out of the paper or rephrase their conclusions. This is a general problem throughout the report and my list below in mainly to exemplify the problem rather than to provide a complete list of concerns.
+- Figure 1, no indication of # of animals or experiments.
+- Figure 1c-e, no significances indicated, hence not possible to support the authors conclusion on row 148 that the response peak at day 9.
+- Figure 2 c-e, no stats indicated. There would appear to be as much HA+GC as HA+EF cells at day 6.
+- The only stats indicated in figure 3b are infection day 10 and 3 days postimmunization. This is not an optimal or relevant comparison. Why have the authors used day 7 data to analyze EF and 10 for GC?
+- In S3a, significant increase in GC is not mentioned. Many data points so spread out that phenotypes well can be lost.
+- In s3b, is there really not any significant difference in EFs in the TLR7 KO?
+- In figure 4b, the TKO display a significant reduction in CD138+EFs that is ignored.
+- In row 204 the authors claim that there is no significant difference between wt and Tko serum in figure 4c. However, the graph shows clear differences day 7 and 8.
+- Row 246 the authors claim that fewer DK and TK cells express Ki67, however, no statistical analysis has been performed in figure S7.
+- Row 261, the authors claim that pMtor is increased in TLR signaling deficient cells. However, stat analysis only shown for 0 and 100 Ig concentrations within each one of the genotypes. Hence, I cannot see proper validation of KO vs wt.
+- Row 264, the authors claim that TLR signaling led to enhanced surface IgD expression, however, figure 9a does not show any statistical significances.
+- Row 331 the authors claim that Ag+LPS boosted mice had the highest anti influenza serum levels but in panel 8f, there is no significant difference between Ag boosted and Ag LPS boosted animals.
+
+Minor comments:
+- Figure legend for fig 2 contains ref to panels I, j, k not in the figure.
+- Row 162-165 talks about smaller GCs. Should this be fewer cells as the size of the GC hardly is investigated?
+- 3b GC cells compared to infection day 10 and EF cells day 7.
+- Row 214 talks about larger ERFs, correct?
+- Are the significance values in figure S8c really correct?
+Reviewer #3 (Remarks to the Author):
+
+In this manuscript the authors show that Influenza infection induces rapid early antibody response that is driven by signals from the toll-like receptors (TLRs). This response appears to be protective as seen by serum transfer experiments and inclusion of TLR signals also increases early antibody response during immunization with Influenza antigens. This study is interesting and important in terms of understanding signals regulating early antibody response against viruses and will be important in thinking about vaccine design for Influenza virus. However, the manuscript is presented in manner that makes it difficult understand the relevance and contribution of extrafollicular response. It is known that TLR signals can affect both early B cell responses and germinal center (GC) B cell responses. The main interesting point of this study is the specific effects of TLR signals on extra follicular (EF) response during infection and immunization. But the lack of clarity on what is considered EF response, whether EF response is protective or pathogenic makes it difficult to understand the specific role of EF response. Similarly, it is also not clear which effects of TLR signals are specific to EF B cells and not GC B cells. Details on EF response as well as additional clarification of the data and figures is necessary to appreciate the importance of EF response during Influenza infection and for publication of the manuscript.
+
+Main comments:
+
+It is not described anywhere in the manuscript what the authors consider to be extra follicular response and whether it is the timing or response, location or the markers on the cells that define this type of response. Similarly, many of the figures are based on using CD24 to delineate both GC and EF B cells and again it is not stated anywhere why this strategy was chosen as CD24 is not a marker normally used to delineate GC cells. More details are needed with regards to these.
+
+Strategy for gating of extrafollicular B cells, extrafollicular plasma blasts and germinal center B cells should be explained at least in the first figure on the figure legends and the results section. Similarly in Figure 2 gating strategy for gating on Influenza HA – specific B cells is unclear and there is very little details given in the text or figure legends. Figure2 legend has multiple errors including mislabeling of the figure legends.
+
+GC B cell data on Figure 1 and 2 or at least just one of the figures should be confirmed with other GC markers apart from CD24 such as PNA or FAS or GL7.
+
+Influenza infection induces early antibody responses in the lungs and recent studies have shown the importance of antigen localization and B cell responses in the lungs (Allie SR et. al 2019 Nature Immunology, Oh EJ Science Immunology 2021) and part of this could also be driven by EF response. It is not clear why the authors decided to investigate only EF response in the mediastinal lymph nodes and not in the lungs which could be more relevant for intranasal infection. Authors could include findings from the lungs or discuss the rationale for only looking at the lymph nodes.
+
+In Figure 4a, the data show that loss of TLR signals lead to specific changes in EF cells but the bone marrow chimera data in Figure 5 b shows that the DKO and TKO chimeras show reduction in both EF and GC responses indicating B cell specific TLR signals have effects on both populations. Similarly, the BCR are responses data are from total B cells showing that changes in TLR signals alters the response of all B cells. Could the authors clarify based on these data how they are concluding that TLR signals specifically effect EF cells? How does the changes in BCR dynamics in all B cells in the knockout condition only lead to effects only on EF cells?
+
+The authors conclude that repeated LPS stimulation polarizes the cells to EF fate, but the data are from time points where you see mostly EF cells and not GC cells, therefore effect is only seen on EF cells. Can the authors look at later time points when GC response are optimal and show that repeated LPS does not have any effect on GC cells?
+
+Could the authors discuss how repeated LPS immunization leads to protective response, is it due to changes in amount of antibody, affinity of antibody or cytokines that might be present in the serum?
+Overall, the exact contribution of the EF response during infection is not clear. If EF responses are protective, they should induce viral clearance at early time points? Therefore, in supplementary Fig4a should the viral titre not be higher in the chimeric knockout mice also since they are not able to induce optimal EFR response? If EFR response does not affect viral clearance could the authors discuss how it could be leading to protection ? Similarly, In page 9 it is stated “Thus B cell-intrinsic TLR signaling supports early EFR formation, while additional B cell extrinsic signal further drives EFR generation in a manner that correlates with pathogen burden. How is early EFR response protective and later EFR response pathogenic? Could the authors clarify this and discuss this further as this would be important in thinking of vaccine design?
+
+The data on repeated LPS immunization inducing EFR response are very interesting. Do the authors think this is specific to LPS or inclusion of other TLR ligands such as TLR9 or TLR7 ligands also lead to this feature? The Influenza virus incorporates ssRNA therefore could this be due to stimulation of both TLR4 and TLR7? Additional discussion on this would be useful.
+Point-by-point response to reviewer’s comments
+
+We thank the reviewers and editors for the thoughtful and constructive comments provided to our manuscript. As we outline below, we have considered and addressed each of the comments provided. This included removal of some data from the supplemental data that lacked statistical power, the strengthening of key findings with additional experimental data and the provision of additional data that demonstrate the induction of protective extrafollicular-derived antibodies against lethal influenza challenge in wild type mice but not those deficient in TLR signaling. Furthermore, we have altered a number of figures and the accompanying text to enhance clarity as requested by the reviewers. We feel that these changes have significantly strengthened the manuscript on both the technical and conceptual aspects of the data. Responses to comments are outlined in blue.
+
+Reviewer #1 (Remarks to the Author):
+
+This manuscript by Lam and Baumgarth investigates the signals that control the generation of the EF versus GC B cell response following infection or vaccination. This is a question of high import given that early control of infection can determine the extent of pathology and severity of illness. The authors reveal a regulatory role of TLR mediated signals in determining the fate decision of activated B cells, with a sustained innate signal promoting differentiation along the EF pathway resulting in early production of antibody capable of promoting clearance. The study includes a careful timecourse of EF versus GC cell and the source of virus-specific antibody. This is an important study that significantly extends our understanding of B cell differentiation. Overall the conclusions are well supported although at times conclusions are based on results that do not appear to be significant (detailed below).
+
+Major comments:
+1. One would expect an increase in HA-specific B cells in Fig 3e given the substantial increase in GC cells in the vaccinated animals, but this seems not to be the case. Could the authors comment on the specificity of these new GC cells?
+We thank the reviewer for this question. In response we have reanalyzed the data that now show a significant increase in HA-specific B cells between day 0 and day 3, with further increases on day 10 post infection (Fig. 3c-d). Furthermore, we now also demonstrate that HA - binding B cells are enriched for a GC phenotype over an EF phenotype (Fig 3e). The number of HA-specific cells captured with our HA-baits is likely smaller than the total number of HA-specific cells present in the population, as not all HA-specific cells can retain binding to the HA-protein. This is a well known limitation of antigen-specific B cell staining for non-transgenic B cells. In addition, as HA is one of 10 influenza proteins encoded by the virus, other GC cells are directed especially against the nuclear protein and the neuraminidase.
+
+2. What percentage of cells fall into the positive quadrant in non vaccinated mice (Fig. 3d)? As there is no increase in HA-sp B cells from d3 to d7, without a control, can one be certain these are in fact HA specific cells being analyzed?
+We thank the reviewer for this question. In response we have made changes to the HA data in Fig. 3c-e, as also outlined above. We are now showing representative flow plots of HA staining on B cells (c), quantifying total HA B cells per timepoint (d), and showing the phenotype (EF vs
+GC) of HA B cells per time point (e) post-immunization. We hope this clarifies the identity of the cells in question.
+
+3. The data in SF2 appear to be total and not HA-specific. Without that, it is difficult to draw conclusions about the HA response.
+We thank the reviewer for this comment. SF2 does not attempt to specifically address antigen-specific responses. Rather, it supplements Fig. 3’s immunization data by showing how EFRs do not respond to greater antigen levels alone, while overall GC sizes do increase upon increased antigen exposure. To clarify and simplify this message, we have incorporated SF2 into subfigures of Fig. 3 as Fig. 3b, which immediate follows the quantification of GC and EF responses in the immunization of original dosage. Although not identifying antigen-specific cells the data very clearly demonstrate the antigen-dose dependent nature of the GC but not the EF response, further supporting our claims that antigen alone does not drive EF responses.
+
+4. In SF3, there is a trend towards increased EF in TLR7KO mice. The low number of animals and variability in the TLR7 KO mice make achieving significance challenging. If the two types of EF cells are pooled for analysis, does the picture become clearer?
+This question by the reviewer is much appreciated. We did find statistical significance for TLR7 KO EFRs, but the displayed p-value was lost during editing, we apologize for the oversight. The CD138+ EFs (which are NOT significant) are a subset of the total EF PBs, not a separate population, and therefore are already included in the EF PB population. Text has been added to improve clarification for former SF3 (now SF2) on these populations at Line 188: However, infection of mice lacking TLR3, TLR4, or TLR7 did not result in significant decreases in total EF PBs, nor EF PBs that were CD138+, compared to their WT controls (Suppl. Fig. 3b). In fact, there was a slight but significant increase in EFRs of TLR7 KOs. Thus, individual cytokines or innate signaling receptors appeared either not necessary or redundant for EFR development and, in the case of TLR7, may even contribute towards negative regulation.
+
+5. In Fig 4b, did the DKO serum provide any protection? Was a control serum transfer experiment performed?
+We thank the reviewer for this question. Control experiments from naïve animal serum transfer of all strains (WT, DKO, TKO) have been added to the revised Fig. 4b as per the reviewer’s suggestion. The data demonstrate that serum from naïve mice, independent of their genotype, could not provide immune protection against a lethal influenza irus challenge.
+
+6. The data in SF5 are challenging for comparison between conditions as there are limited times when the same dose is used in combination versus individual stimulations. Nonetheless, one can compare the 1ug/ml anti-IgM and 1ug/ml LPS in this way. When doing so it is hard to see the how the combination strongly supported viability (SF5a) or proliferation (SF5b) compared to the LPS alone as is stated. They appear similar.
+We thank the reviewer for this comment. In response we have removed some of the data in SF5 and now only compare the most relevant groups, as suggested by the reviewer. The data now clearly demonstrate that 1) TLR signaling has a significantly stronger effect on B cell viability than BCR stimulation at the given doses. However, there is still a small and significant additive effect using both and 2) both TLR and BCR signaling have sizable, additive effects on proliferation and IRF4 upregulation. Clarification has also been added to the relevant text.
+7. Are the reductions in Ki67+ cells in SF7 significant? If not this limits the conclusion that can be drawn in lines 247-49.
+We thank the reviewer for this comment. After consideration of the importance of demonstrating a difference in Ki67+ cell frequencies, we have decided to remove this particular supplemental data and the associated text. While we are confident that additional sample size would yield significant differences, due to time constraints we opted to omit their repeat. Furthermore, we like to point out that we already demonstrated a relation between TLR signaling and increased proliferation in the in vitro B cell activation data and LPS-boosted immunization data, where we measured Ki67 among HA-specific B cells, providing models that are B cell intrinsic and both intrinsic/extrinsic.
+
+8. In lines 264-65 authors should state trend as it does not appear to significantly increase. For many of the animals there is a very modest increase and one TKO has a major decrease. Thus conclusions should be drawn with care.
+We thank the reviewer for this comment have decided to remove the supplemental figure in question. While we are confident that additional sample size would yield significant differences in IgD expression in vivo and support the significant differences found in vitro, it is an observation that is not central to understanding the mechanisms of EF fate decisions explored in this manuscript.
+9. Additional discussion of the TLR ligand independent requirement for TLR pathway in B cell activation would be welcome.
+We thank the reviewer for the suggestion. In response we have provided additional discussion on the topic of TLR-ligand-independent requirement of optimal BCR activation, stating on line 370 of the amended text: “Evidence of enhanced TLR9-MyD88-BCR complexing in activated B cell-like lymphoma cells suggests that TLRs may provide a platform for downstream TLR targets to become activated through the BCR and its effector pathway (Phelan et al., 2018)”. Consequentially, this would make activation of the integrated TLR pathway as the most crucial source of BCR-mediated IRF4.”
+
+Minor comments
+1. Line 81: believe meant vesicular not vaccinia
+This error has been corrected.
+
+2. Line 127: add virus after influenza
+This error has been corrected.
+
+3. Line 197: EFR-derived serum is a bit unclear, suggest “serum from d10 animals wherein antibody is predominantly EFR derived” if this is what is meant
+This edit has been made, we thank the reviewer for this suggestion.
+
+4. Line 211: the chimera to which the authors are referring should be noted prior to BMC
+This error has been corrected.
+
+5. Line 370: induce for indue
+The error has been corrected.
+6. The authors are asked to increase axis labels and percentage values in flow plot font size where possible as many are quite hard to read.
+Axes and labels have been increased where needed, we thank the reviewer for this suggestion.
+
+7. Consider keeping nomenclature similar- “fold difference” axis label Fig 6e
+The axis label has been changed, we thank the reviewer for this suggestion.
+
+Reviewer #2 (Remarks to the Author):
+
+The paper by Lam et al explores the mechanisms underlying the formation of extrafollicular plasmablasts (EFRs) and short term humoral memory toward influenza. The authors provide support for a correlation between the formation of EFRs and protective properties of serum at early time points after antigen challenge. Using a series of KO models the authors determine that a key component for the formation of EFRs is Toll like receptor signaling that they find stimulate IRF4 transcription in the activated B-cell. They conclude that inflammatory signals impact the fate of activated B-cells to generate efficient short term response to antigen.
+
+This is in many regards an interesting report representing an impressive amount of work in different model systems. The large number of models and experimental setups explored does, however, become a liability as parts of the data are of limited quality. The authors base many of their conclusions on data collected from few animals creating a challenge in interpretation of the results. This problem is aggravated by that substantial portions of the data are presented as relative controls, in some cases wt, and other cases as compared to non-stimulated cells. It is difficult to fully delineate how this normalization was made. Furthermore, the authors does at times ignore significant differences in their data when they and put forward differences that are not indicated as significant. This creates a challenge to read and validate the data in a proper manner making it difficult to fully appreciate the reported findings and to understand the relevance of the findings.
+
+Specific comments.
+1. The authors should go over their data and decide what parts that are conclusive and where there is a need to repeat the experiments, alternatively take some data out of the paper or rephrase their conclusions. This is a general problem throughout the report and my list below in mainly to exemplify the problem rather than to provide a complete list of concerns.
+We thank the reviewer for these suggestions and have taken specific steps to clarify data representation and conclusions drawn, some in response also to comments from reviewer #1. We now explicitely describe fold-differences to WT controls in figure legends and text for appropriate data. Data lacking significance have been removed (chimera HA-specific Ki67 and IgD expression) or repeated (immunization with LPS boosting experiments in Figs. 7 and 8). The results have strengthened and further confirmed our initial conclusions.
+
+2. Figure 1, no indication of # of animals or experiments.
+We apologize for this oversight. The number of animals and experiments conducted have now been added to the legend for Figure 1.
+3. Figure 1c-e, no significances indicated, hence not possible to support the authors conclusion on row 148 that the response peak at day 9.
+We thank the reviewer for this comment and have conducted statistical analysis using one-way ANOVAs on each time course measurement (Fig. 1c-e). Significance has now been indicated at the amended Fig. 1 and the corresponding text has been clarified.
+
+4. Figure 2 c-e, no stats indicated. There would appear to be as much HA+GC as HA+EF cells at day 6.
+One-way ANOVAs were conducted on each time course measurement (Fig. 2c-e) and significance has now been indicated in the amended Figure. The observation by the reviewer that HA+GC and HA+EF cell frequencies are similar was likely based on the fact that we had used y-axes of different scales to indicate small changes in the GC populations. As the most important comparision here is between GC and EF responses and to avoid that potential confusion, we have now adjusted both y-axes for Figs. 2d and 2e to be the same.
+
+5. The only stats indicated in figure 3b are infection day 10 and 3 days postimmunization. This is not an optimal or relevant comparison. Why have the authors used day 7 data to analyze EF and 10 for GC?
+We thank the reviewer for this comment. We had included the data simply to indicate the large differences between these responses, but agree with the reviewer that this is hard to justify. Therefore, we have made several changes to Figure 3 to better articulate that GC responses are the dominant B cell fate during immunization and that EF responses are minimal. As the reviewer has suggested, we have removed the infection timepoint comparisons and focused solely on characterizing the immunization response. Fig. 3a now quantifies EF and GC responses, while data from SF2 has been incorporated as Fig. 3b to show that increasing antigen dose does not increase or rescue total EF responses, while it does increase overall GC responses in an antigen dose-dependent manner. Furthermore, Figs. 3c-e have been reconfigured to better show EF vs GC fate of HA-specific B cells, showing their expansion after immunization in Fig. 3c - d, and characterizing their phenotype (EF vs GC) in Fig. 3e. We believe this amended figure more clearly demonstrates that immunization s.c. favors GC responses and provides contrast to primary infection’s early bias towards EF responses.
+
+6. In S3a, significant increase in GC is not mentioned. Many data points so spread out that phenotypes well can be lost.
+We thank the reviewer for this comment. We have now added text to the manuscript to mention the significant increase in GC responses in mice lacking TNFa signaling at line 183.
+We acknowledge the reviewer’s concern that some data points are spread out in some of the many groups of gene-targeted mice we screened (we only show the most pertinent ones we tested in the paper), we are confident that each of the genotypes tested failed to show measurable reductions in EFRs at the chosen timepoint, thus that these genes/signaling pathways are no critical for early EFR formation after influenza virus infection. The data were always obtained by simultaneously testing the various KO mice against age and sex-matched congenic strains of C57BL/6 mice and for numerous of these comparisons we performed ELISA for virus-specific serum antibodies. Because these ELISA data failed to show any significant differences, and because day 7 is a rather early timepoint to measure antibody levels, we are not showing this additional data in the manuscript.
+
+7. In s3b, is there really not any significant difference in EFs in the TLR7 KO?
+We apologize for the omission, also remarked by reviewer #1. Indeed, we did find statistical significance for TLR7 KO EFRs, but the displayed p-value was lost during editing, we apologize for the oversight. The CD138+ EFs (which are NOT significant) are a subset of the total EF PBs, not a separate population, and therefore are already included in the EF PB population. Text has been added to improve clarification for former SF3 (now SF2) on these populations at Line 188: However, infection of mice lacking TLR3, TLR4, or TLR7 did not result in significant decreases in total EF PBs, nor EF PBs that were CD138+, compared to their WT controls (Suppl. Fig. 3b). In fact, there was a slight but significant increase in EFRs of TLR7 KOs. Thus, individual cytokines or innate signaling receptors appeared either not necessary or redundant for EFR development and, in the case of TLR7, may even contribute towards negative regulation.
+
+8. In figure 4b, the TKO display a significant reduction in CD138+EFs that is ignored.
+We thank the reviewer for this comment. In response we now added to the amened manuscript that there was a significant reduction in TKO CD138+ EFs (Fig. 4a) at line 206: Surprisingly, infection of another TLR-null model, through deletion of genes for TLR2^{37}, TLR4^{38} and a missense mutation of Unc93b^{39} (TKO), showed EFRs similar to WT controls (**Fig. 4a**) along with nominal passive protective capacity (**Fig. 4c**), despite slight reductions in CD138+ EF PBs at 7 dpi (**Fig. 4a**).
+
+9. In row 204 the authors claim that there is no significant difference between wt and Tko serum in figure 4c. However, the graph shows clear differences day 7 and 8.
+We thank the reviewer for this comment and have now changed the wording of the results section for Fig. 4c to specifically address the outcome of TKO serum transfer in these protection experiments on Line 205: “Surprisingly, infection of another TLR-null model, through deletion of genes for TLR2^{37}, TLR4^{38} and a missense mutation of Unc93b^{39} (TKO), showed EFRs similar to WT controls (**Fig. 4a**) along with nominal passive protective capacity (**Fig. 4c**), despite slight reductions in CD138+ EF PBs at 7 dpi (**Fig. 4a**).”
+
+10. Row 246 the authors claim that fewer DK and TK cells express Ki67, however, no statistical analysis has been performed in figure S7.
+We thank the reviewer for this comment also made by reviewer #1. As outlined above, we decided to remove the supplemental figure and associated text, as we have shown this to be the case in other parts of the manuscript.
+
+11. Row 261, the authors claim that pMtor is increased in TLR signaling deficient cells. However, stat analysis only shown for 0 and 100 lg concentrations within each one of the genotypes. Hence, I cannot see proper validation of KO vs wt.
+We apologize for this oversight. In the revised manuscript we have now added the statistical analysis comparing each treatment from each KO to their respective WT treatment control, indicated by stars above each individual treatment condition. The associated text has been clarified/changed as well to better reflect the data.
+
+12. Row 264, the authors claim that TLR signaling led to enhanced surface IgD expression, however, figure 9a does not show any statistical significances.
+We thank the reviewer for this comment. As also remarked in response to reviewer #1, we have decided to remove the supplemental figure in question. While we are confident that additional sample size would yield significant differences in IgD expression *in vivo* and support the
+significant differences found in vitro, it is a supplemental observation that does not specifically address the question of the mechanisms of EF fate decisions explored in this manuscript.
+
+13. Row 331 the authors claim that Ag+LPS boosted mice had the highest anti influenza serum levels but in panel 8f, there is no significant difference between Ag boosted and Ag LPS boosted animals.
+We thank the reviewer for this comment. In response we have conducted additional experiments and added the data to the figure. The data now clearly demonstrate significant differences between Ag+LPS and Antigen only (Fig. 8f) both when expressed as relative units total flu-specific IgG (left panel) and when showing normalized data to the average “Antigen Only” IgG levels from each experiment, then expressed as fold-change. The data demonstrate the significant increases in antigen-specific, serum IgG when LPS boosting occurs in both absolute and relative terms, explaining the enhanced passive protective capacity of the sera from these mice, shown in Fig. 8g-h.
+
+Minor comments:
+1. Figure legend for fig 2 contains ref to panels l, j, k not in the figure.
+This error has been corrected, we apologize for the oversight.
+
+2. Row 162-165 talks about smaller GCs. Should this be fewer cells as the size of the GC hardly is investigated?
+This error has been corrected.
+
+3. 3b GC cells compared to infection day 10 and EF cells day 7.
+We believe we have addressed this above (Major Comment #5) and made the appropriate changes to the figure in question (Fig. 3).
+
+4. Row 214 talks about larger ERFs, correct?
+The text has been clarified to address the reviewers comment.
+
+5. Are the significance values in figure S8c really correct?
+We thank the reviewer for this question. The significances shown in (former) SF8c are from one-way ANOVAs of each strain to demonstrate that different anti-IgM treatment concentrations were actually causing changes in the measured protein’s phosphorylation signature. With the changes made from the reviewer’s previous comment on the figure, the effect on p38 in TLR-null B cells has been clarified.
+
+Reviewer #3 (Remarks to the Author):
+
+In this manuscript the authors show that Influenza infection induces rapid early antibody response that is driven by signals from the toll-like receptors (TLRs). This response appears to be protective as seen by serum transfer experiments and inclusion of TLR signals also increases early antibody response during immunization with Influenza antigens. This study is interesting and important in terms of understanding signals regulating early antibody response against viruses and will be important in thinking about vaccine design for Influenza virus. However, the manuscript is presented in manner that makes it difficult understand the relevance and contribution of extrafollicular response. It is known that TLR signals can affect both early B
+cell responses and germinal center (GC) B cell responses. The main interesting point of this study is the specific effects of TLR signals on extra follicular (EF) response during infection and immunization. But the lack of clarity on what is considered EF response, whether EF response is protective or pathogenic makes it difficult to understand the specific role of EF response. Similarly, it is also not clear which effects of TLR signals are specific to EF B cells and not GC B cells. Details on EF response as well as additional clarification of the data and figures is necessary to appreciate the importance of EF response during Influenza infection and for publication of the manuscript.
+
+Main comments:
+
+1. It is not described anywhere in the manuscript what the authors consider to be extra follicular response and whether it is the timing or response, location or the markers on the cells that define this type of response. Similarly, many of the figures are based on using CD24 to delineate both GC and EF B cells and again it is not stated anywhere why this strategy was chosen as CD24 is not a marker normally used to delineate GC cells. More details are needed with regards to these.
+We thank the reviewer for this comment. In response we have added additional details to the description of EFRs in the introduction of the amended manuscript at Line 59: “Instead, early antibodies are produced from short-lived plasmablasts of the extrafollicular response (EFR), which develop and localize within the medulla and interfollicular regions2 of the respiratory tract-draining mediastinal lymph nodes (medLN) shortly after infection and before GC formation3.” Furthermore, on Line 67: “EFRs thus appear physiologically distinct from GCs and can generate protective, germline-encoded, antigen-specific ASCs from the restrictive repertoire of inbred mice.”
+We and others previously demonstrated that CD24 staining can delineate B cell subsets. Specifically, that CD45R+/PNA+ cells identified as GC B cells had high CD24 expression and additionally co-stained as GL7+ and Fas+ (Shinall et al., 2000, JI; Baumgarth, 2004, Methods Cell Biol.; Elsner et al. 2015 PloSPathog). Nonetheless, to more clearly delineate our gating strategy we have updated Fig. 1 with added labels and we show the expression of GL7 and relatively high IRF8 on CD24hi gated GC and expression of CD138 and high IRF4, in EF PB, respectively.
+
+2. Strategy for gating of extrafollicular B cells, extrafollicular plasma blasts and germinal center B cells should be explained at least in the first figure on the figure legends and the results section. Similarly in Figure 2 gating strategy for gating on Influenza HA – specific B cells is unclear and there is very little details given in the text or figure legends. Figure 2 legend has multiple errors including mislabeling of the figure legends.
+We thank the reviewer for this comment. In response we have made multiple changes to the relevant text in the results section and figure legends to Figs 1 and 2 to clarify gating strategy. Furthemore, labels were added to Fig. 1 to aid the identification of target cell populations. We hope that these changes are clarifying identification of each cell subset.
+
+3. GC B cell data on Figure 1 and 2 or at least just one of the figures should be confirmed with other GC markers apart from CD24 such as PNA or FAS or GL7.
+We thank the reviewer for this suggestion. CD24/CD38 gating for GCs is now confirmed by GL7 staining in Figure 1 along with IRF8 expression level as outlined in response to the previous comment. In addition, we refer to our previous publications that identify this gating strategy as
+appropriate as these cells are PNA+ FAS+ and GL7+ (Elsner et al, 2015 PloSPathog), the latter now added to the Figure.
+
+4. Influenza infection induces early antibody responses in the lungs and recent studies have shown the importance of antigen localization and B cell responses in the lungs (Allie SR et. al 2019 Nature Immunology, Oh EJ Science Immunology 2021) and part of this could also be driven by EF response. It is not clear why the authors decided to investigate only EF response in the mediastinal lymph nodes and not in the lungs which could be more relevant for intranasal infection. Authors could include findings from the lungs or discuss the rationale for only looking at the lymph nodes.
+We thank the reviewer for this comment. In response, we have now added to the text at the beginning of the results section that “Influenza-specific ASCs can be found predominantly in the medLN within 7 days after primary infection but are not found in the lungs until 14 dpi3, well after virus clearance. This indicated that medLN EFRs are the main source of the early antigen-specific antibody response and were thus investigated.”
+
+5. In Figure 4a, the data show that loss of TLR signals lead to specific changes in EF cells but the bone marrow chimera data in Figure 5 b shows that the DKO and TKO chimeras show reduction in both EF and GC responses indicating B cell specific TLR signals have effects on both populations. Similarly, the BCR are responses data are from total B cells showing that changes in TLR signals alters the response of all B cells. Could the authors clarify based on these data how they are concluding that TLR signals specifically effect EF cells? How does the changes in BCR dynamics in all B cells in the knockout condition only lead to effects only on EF cells?
+We thank the reviewer for this comment and acknowledge the observations made about differences in phenotypes between the knockout models, with the TLR-null bone-marrow chimera data and in vitro data showing a universal effect on B cell activation, suggesting an effect on both EF and GC, while the global TLR knockout in vivo data demonstrated a more specific effect on EF responses. Firstly, the global knockout and chimera infection data demonstrate that there is a bona-fide B cell-intrinsic effect through TLRs on the EF response, not that it is exclusive to the EF response. We have ensured the text reflects this conclusion carefully as well. However, this does indicate that the GC response can be rescued if non-B cells also lack TLR signaling. As the focus of this paper is on the EF response, we did not explore the mechanisms by which GC responses are rescued in a global TLR-null context, but have provided a possible explanation in the discussion at Line 361: “As TLR signaling in DCs leads to increased Th1 polarization50, perhaps a total ablation of TLR signaling may polarize more CD4 T cells towards a Tfh phenotype, compensating for the GC B cell-intrinsic defect in TLR-mediated activation”.
+
+6. The authors conclude that repeated LPS stimulation polarizes the cells to EF fate, but the data are from time points where you see mostly EF cells and not GC cells, therefore effect is only seen on EF cells. Can the authors look at later time points when GC response are optimal and show that repeated LPS does not have any effect on GC cells?
+We thank the reviewer for this question. Additional data has been added (Suppl. Fig. 8) that shows GCs were not affected by the LPS-boosting regimen. Thus, demonstrating no detrimental effect on the development of GC responses. Text addressing these data have been added as well.
+7. Could the authors discuss how repeated LPS immunization leads to protective response, is it due to changes in amount of antibody, affinity of antibody or cytokines that might be present in the serum?
+Additional discussion on how repeated LPS boosting leads to more protection has been included as per the reviewer’s request at Line 398: “As the data suggest, this may be due to an overall increase in anti-influenza antibodies with functional protective capacity under continuous TLR-activating conditions (Fig. 7f) but could also be due to differences in antibody quality, i.e differences in repertoire that target unique or a higher number of epitopes. As TLR activation provides B cells with increased IRF4 expression, this allows for clones with relatively weaker BCR interactions to partake in antibody secretion by reaching the required IRF4 threshold.”
+
+8. Overall, the exact contribution of the EF response during infection is not clear. If EF responses are protective, they should induce viral clearance at early time points? Therefore, in supplementary Fig4a should the viral titre not be higher in the chimeric knockout mice also since they are not able to induce optimal EFR response? If EFR response does not affect viral clearance could the authors discuss how it could be leading to protection? Similarly, in page 9 it is stated “Thus B cell-intrinsic TLR signaling supports early EFR formation, while additional B cell extrinsic signal further drives EFR generation in a manner that correlates with pathogen burden. How is early EFR response protective and later EFR response pathogenic? Could the authors clarify this and discuss this further as this would be important in thinking of vaccine design?
+We thank the reviewer for these pertinent questions. The follow text has been added to address the reviewer’s questions on the contributions of EFRs towards virus clearance at Line 70:
+“Addressing how these distinct B cell activation outcomes contribute to humoral immunity against acute respiratory tract virus infections, where rapid induction of immunity is a key determinant of survival, is pertinent for our understanding of the pathogenesis of these infections and the role of B cell immunity. While CD8 T cells are credited the most in clearance of influenza during primary infection, they alone cannot prevent mortality5 and may collaterally target non-infected antigen-presenting cells6. Additionally, lack of B cells lead to a ~50 fold increase in virus titers by day 10 post-infection7 demonstrating the potential importance of early antibody generation against influenza. This has important implications for vaccine design, as vaccines are generally considered only successful if inducing GC-derived, long-lived plasma cells and memory B cells. However, vaccinations during the ongoing COVID-19 pandemic or during seasonal influenza virus infections are likely more effective if they can induce immune protection more quickly, i. e. through EFRs.”
+
+As for addressing if EFRs are protective vs pathogenic, we demonstrate here that EFRs during influenza are protective when MyD88/TRIF signaling is replete and that this protective effect is lost when MyD88/TRIF is absent, but upstream TLR ablation does not lead to this defect. Differences in virus clearance between global TLR-null mice and B cell-specific, TLR-null chimeras demonstrate the contribution of TLR signaling in both manners, with TLR-null chimeras having replete TLR signaling in non-B cells that allows for optimal innate and T cell responses leading to nominal virus clearance (Suppl. Fig. 3a), despite disruptions in both B cell responses (Fig. 5b).
+
+9. The data on repeated LPS immunization inducing EFR response are very interesting. Do the authors think this is specific to LPS or inclusion of other TLR ligands such as TLR9 or TLR7
+ligands also lead to this feature? The Influenza virus incorporates ssRNA therefore could this be due to stimulation of both TLR4 and TLR7? Additional discussion on this would be useful.
+We thank the reviewer for this comment and have addressed questions regarding specific TLR activation with the following text at Line 408: “Whether the type of TLR agonist, i.e. one which activates MyD88 or TRIF exclusively, would differentially affect EFR dynamics is unclear. As stated previously, using TLR4 (MyD88 and TRIF) and TLR7 (MyD88 only) adjuvant made no difference in early antibody responses after immunization compared to either alone15.
+Additionally, TLR9 activation by CpG enhanced titers but worsened the quality of antigen-specific antibody responses due to a lack of GC-mediated affinity maturation to the hallmark antigen hapten19. Affinity for hapten increases over time as GCs mature and affinity maturation takes place54, indicating that clones with low avidity interactions with hapten may be ‘pulled’ into differentiation through activation of TLRs, thus lowering the overall avidity of the response. Given the PAMPs present in influenza virus, characterization of TLR/BCR synergy upon virus recognition and uptake by B cells, and how this may contribute towards plasmablast formation, would be of interest. Yet increases in serum antibody affinities over time were not observed following infection with vesicular stomatitis virus8,9 and high affinity, germline-encoded antibodies to hemagglutinin were induced early after influenza inoculation4.”
+REVIEWERS' COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The authors have addressed the concerns raised.
+
+I note a few edits that are needed.
+1. Line 333 and "=" mark
+2. Line 399 "in in"
+3. Line 407 Was the or meant to be an and?
+
+Reviewer #2 (Remarks to the Author):
+
+The paper is now of acceptable quality and the general story holds up. I do, however, believe that the authors should be more careful in future submissions as it is not reasonable that reviewers should act as proofreaders of submitted reports.
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have addressed all of my concerns from the previous review , including adding new data related to markers for GC cells and clarification of phenotype and function of extra follicular cells. Additional comments in the discussion about the role of TLR signaling are also adequate.
+Point-by-point response to reviewer’s comments:
+We’d like to thank all the reviewers for their comments that improved this manuscript’s message and data robustness/fidelity. It is much appreciated.
+
+Reviewer #1 (Remarks to the Author):
+
+The authors have addressed the concerns raised.
+We thank the reviewer for their helpful feedback in making this a stronger manuscript.
+
+I note a few edits that are needed.
+1. Line 333 and "=" mark
+The error has been corrected, apologies for the oversight.
+
+2. Line 399 "in in"
+This error has been corrected, apologies for the oversight.
+
+3. Line 407 Was the or meant to be an and?
+This error has been corrected, apologies for the oversight.
+
+Reviewer #2 (Remarks to the Author):
+
+The paper is now of acceptable quality and the general story holds up. I do, however, believe that the authors should be more careful in future submissions as it is not reasonable that reviewers should act as proofreaders of submitted reports.
+We thank the reviewer for their helpful feedback, apologize for the original manuscript appeared insufficiently edited/proofread.
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have addressed all of my concerns from the previous review, including adding new data related to markers for GC cells and clarification of phenotype and function of extra follicular cells. Additional comments in the discussion about the role of TLR signaling are also adequate.
+We thank the reviewer for their helpful feedback in making this a stronger manuscript.

027b827c3b80c86306a8c299d162b6cb2bb24a38c0ffa3c834d579a1e6c6338c/preprint/preprint.md

@@ -0,0 +1,265 @@
+Toll-like receptor mediated inflammation directs B cells towards protective antiviral extrafollicular responses
+
+Jonathan Lam
+University of California, Davis
+Nicole Baumgarth (nbaumga3@jhmi.edu)
+School of Veterinary Medicine, University of California, Davis
+
+Article
+
+Keywords:
+
+Posted Date: November 22nd, 2022
+
+DOI: https://doi.org/10.21203/rs.3.rs-2226474/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on July 5th, 2023.
+See the published version at https://doi.org/10.1038/s41467-023-39734-5.
+Toll-like receptor mediated inflammation directs B cells towards protective antiviral extrafollicular responses
+
+Jonathan H. Lam1,2 and Nicole Baumgarth1,2,3,4
+1Graduate Group in Immunology, 2Center for Immunology and Infectious Diseases, 3Dept. Pathology, Microbiology and Immunology, University of California Davis, Davis, USA
+
+Correspondence
+Nicole Baumgarth DVM PhD
+W. Harry Feinstone Dept. Molecular Microbiology and Immunology
+Johns Hopkins Bloomberg School of Public Health
+615 N Wolfe Street, E4135
+Baltimore, MD 21205
+(p) 410 614 2718
+nbaumga3@jhmi.edu
+Abstract
+
+Extrafollicular plasmablast responses (EFRs) are considered to generate antibodies of low affinity that offer little protection from infections. Paradoxically, high avidity antigen-B cell receptor engagement is thought to be the main driver of B cell differentiation, whether in EFRs or the slower-developing germinal centers (GCs). This study demonstrates that influenza infection rapidly induced EFRs generating protective antibodies in a B cell intrinsic and extrinsic Toll-like receptor (TLR)-dependent manner. B cell-intrinsic TLR signals supported antigen-stimulated B cell survival, clonal expansion, and the differentiation of B cells via induction of IRF4, the master regulator of B cell differentiation, through activation of NF-kB c-Rel. Provision of sustained TLR4 stimulation after immunization altered the fate of virus-specific B cells towards EFRs instead of GCs, accelerating rapid antibody production and improving their protective capacity over antigen/alum administration alone. Thus, inflammatory signals act as B cell fate-determinants for the rapid generation of protective, antiviral EF responses.
+Introduction
+
+Acute respiratory tract infections induce neutralizing antibody responses that are critical for long lasting protection. Germinal center (GC) responses are considered the most effective in generating protective antibodies, as antigen-specific GC B cells undergo extensive somatic hypermutation, resulting in long-lived antibody-secreting plasma cells (ASCs) that generate high-affinity, strongly neutralizing antibodies. However, after primary influenza virus infection, GCs appear relatively late, usually after viral contraction, and thus are unlikely to contribute towards virus clearance (Lam and Baumgarth, 2019). Instead, early antibodies are produced from extrafollicular plasmablast responses (EFRs) that develop in the respiratory tract-draining mediastinal lymph nodes (medLN) shortly after infection and before GC formation (Rothaeusler and Baumgarth, 2010). Early studies by Gerhard and colleagues demonstrated that influenza inoculations of BALB/c mice resulted in rapid production of early hemagglutinin (HA)-specific, neutralizing IgG antibodies that were protective and in repertoire distinct from those induced later in the response (Kavaler et al., 1990). This included unmutated IgG from B cells of the prototypic HA-specific C12 idiotype, which were excluded from GCs after intra-nasal (i.n.) influenza infection (Rothaeusler and Baumgarth, 2010). This indicates that protective, germline-encoded, antigen-specific ASCs can be generated from the restrictive repertoire of inbred mice via EFRs that are temporally distinct from GCs.
+
+Addressing how these distinct B cell activation outcomes contribute to humoral immunity against acute respiratory tract virus infections, where rapid induction of immunity is a key determinant of survival, is pertinent for our understanding of the pathogenesis of these infections and the role of B cell immunity. Additionally, it is important for vaccine design. Vaccines are generally considered only successful if inducing GC-derived, long-lived plasma cells and memory B cells. However, vaccinations during the ongoing COVID-19 pandemic or during seasonal influenza virus infections, are likely more effective if they can induce immune
+protection more quickly, i. e. through EFRs. The signals required for EFR induction, however, have not been resolved. Indeed, EFR induction has been considered of little consequence, as these responses are thought of as only short-lived and of low protective capacity.
+
+Yet, groundbreaking studies by the Hengartner’s group over two decades ago, demonstrated that antibody responses to vaccinia stomatitis virus showed a surprising lack of changes in virus-specific serum antibody affinities over the course of infection. Instead, they demonstrated that antibodies of relatively high affinity for their cognate antigen were generated both early and late after infection (Kalinke et al., 1996; Roost et al., 1995), suggesting that following a viral infection both EFR and GC-derived antibodies might generate antibody responses of overall high affinity. These data are consistent also with reports by the Brink lab, who demonstrated using the BCR-transgenic swHEL model, that strong BCR-affinity for antigen drove the rapid proliferation and differentiation of hen egg lysozyme (HEL)-specific B cells in EFRs, while lower affinity interactions induced stronger GC responses instead (Paus et al., 2006). Generation of EFRs from high-affinity B cells is consistent also with findings that strong BCR-signaling drives upregulation of interferon regulatory factor 4 (IRF4), a critical transcriptional regulator of plasma cell development (Ochiai et al., 2013). Such a model of affinity-based induction of proliferation and differentiation would be consistent with EFRs’ potential to generate high affinity antibodies.
+
+However, whether BCR-antigen interactions alone drive B cell fate decisions towards EFRs remains unknown. Furthermore, in contrast to studies indicating that highly functional antibodies emerge from EFRs, other work has shown that EFRs developing in the spleen following Salmonella thyphimurium and Ehrlichia infections generate large quantities of predominantly non-specific antibodies (Di Niro et al., 2015; Trivedi et al., 2019), in support of the idea that EFR are of little protective consequence. Together, these data seem to indicate that
+additional infection-induced signals shape EFRs. What these signals are, and how they might affect the functionality and protective capacity of EFR-derived antibodies, is unresolved.
+
+Work interrogating pattern recognition receptors (PPR) signaling after immunization has identified numerous effects on B cells and it is well appreciated that certain PAMPS can work as adjuvants to support vaccine responses. For example, RNA of sheep red blood cells stimulated RNA-sensing PPR mitochondrial antiviral signaling protein (MAVS) and TLR3 (Loetsch et al., 2017), that supported more robust B cell responses. Also, mice immunized with nanoparticles containing the TLR4 ligand 4'-monophosphoryl lipid A induced a more robust, antigen-specific ASC response compared to mice given antigen alone, while the combination of TLR4 and TLR7 agonists was reported to fate B cells towards early memory and germinal center responses, resulting in persistent antibody responses from bone marrow long-lived plasma cells, rather than rapid EFRs (Kasturi et al., 2011). B cell-intrinsic MyD88 signaling was shown also to increase proliferation and differentiation of plasma cells and induced expansion of Bcl6+ germinal center B cells to virus-like particles (Tian et al., 2018). And in Friend virus infection and after infection with influenza virus, B cell-intrinsic expression of TLR7 (but not TLR3) was shown to be required for germinal center formation (Browne, 2011; Heer et al., 2007). In contrast, stimulation with the TLR9 ligand CpG antagonized B cell antigen uptake and processing resulting in disruption of affinity maturation and a reduction in early-formed, antigen-specific plasma cells in the spleen, along with a reduction in long-term, antigen-specific serum IgG avidity (Akkaya et al., 2018).
+
+Observations of TLR integration with canonically distinct B cell activation pathways may play a role in the reported effects of TLR agonists on antibody responses, as TLR4 was shown to integrate with BCR signaling via the phosphorylation of syk (Schweighoffer et al., 2017), while the TLR adaptor MyD88 was shown to be critical for signaling via the B cell survival receptor TACI (He et al., 2010). Collectively, existing evidence suggests that TLR and/or MyD88-
+mediated signaling affects B cell responses, but how these signals integrate to regulate B cell responses remains incompletely resolved.
+
+Here we demonstrate that inflammatory signals induced by influenza infection, but not immunization with virus particles in alum, triggers the rapid generation of protective antibody responses via formation of EFRs in a TLR signaling-dependent manner. TLR- signaling fated B cells towards the EFR/plasma cell state after infection through the strong induction of IRF4 via activation of NFkB c-Rel. Similarly, sustained co-administration of LPS with virus/alum immunization rescued EFR induction after vaccination and improved antibody-mediated protection against lethal influenza challenge.
+
+Results
+
+The extrafollicular B cell response generates antigen-specific antibodies after intranasal influenza infection but not after peripheral immunization
+
+Intranasal infection of C57BL/6 mice resulted in the appearance of pre-GC/GC-like (GC) B cells (CD45Rhi/CD19hi/CD38lo/CD24hi) at 7 days post infection (dpi) that were also interferon regulatory factor 8 (IRF8) high, a transcription factor associated with GC polarization(Xu et al., 2015) (Fig. 1a, top). Early formed plasmablasts of the EFR (EF PBs) were identified as CD45Rlo/CD19+/CD38lo/CD24+ as well as IRF4-high, the latter associated with an ASC fate (Ochiai et al., 2013; Xu et al., 2015), with many also expressing CD138 (Fig. 1a, bottom), a canonical marker of ASCs. EF PBs and GC B cells both had lost surface IgD and most had lost IgM by 7 dpi (Fig. 1b), indicating a high level of class-switching. While B cell frequencies in the medLN remained relatively constant throughout the time course (Fig. 1c), drastic changes in EF and GC compartments took place. Relatively few GC B cells were found
+until after 9 dpi (Fig. 1d), while EF PBs were seen as early as 5 dpi, peaking at 9 dpi and contracting by 14 dpi (Fig. 1e).
+
+Only EF PBs, purified by flow cytometry, secreted pathogen-specific antibodies at 7 dpi, detected as influenza-bound total Ig and IgG2c by ELISPOT on cells (Fig. 2a), demonstrating that EF PBs contain the only functional, influenza-specific ASCs in the medLN at this timepoint. Additionally, use of two distinct fluorophore-labeled, recombinant hemagglutinin (HA) of A/PR8 identified HA-specific (HA) B cells (Fig. 2b) and their preferred participation in EF over GC B cell responses (Fig. 2c-e), with HA-bound B cells comprising as much as 15% of the EFR compartment at the early time points. The independence of EFR formation from GCs during influenza infection, suggested previously (Miyauchi et al., 2016), was confirmed with the presence of EF B cells in infected Mb-1-Cre Bcl6 f/f mice that are unable to form GCs (Suppl. Fig. 1). Thus, EFRs are responsible for the earliest antigen-specific antibody response to influenza infection and are independent of GCs.
+
+A different B cell response quality was seen after subcutaneous (s.c.) immunization with influenza virions in alum adjuvant. Compared to infection, immunizations yielded smaller GCs and little to no EFRs in draining LN at 3, 7 and 10 dpi (Fig. 3a, b). At the antigen-dose used, GCs were only 3-fold smaller but EFRs were at least 40-fold smaller and barely detectable after immunization (Fig. 3c). Consequently, fewer HA B cells were detected over the course of immunization compared to infection (Fig. 3d, e). The HA-specific B cells that were present expressed no CD138 and only a few expressed Ki67 (Fig. 3d, e), a marker of cell cycling. Increasing the virus antigen-dose for immunization increased GC B cell numbers but had no effect on the size of the EFR (Suppl. Fig. 2). We conclude that infection-induced signals are required for EFRs to influenza.
+To identify the influenza infection-induced signals that support EFRs, we first considered inflammatory cytokines that were previously identified as contributing towards B cell differentiation and ASC maintenance, as well as S100A9, a damage-associated molecular pattern protein produced by stressed and dying cells and released during influenza infection(Tsai et al., 2014). Among the cytokines tested, IL-1, Type I interferons (IFN), IL-6, and TNF\( \alpha \) are induced early after influenza infection (Coro et al., 2006; Hayden et al., 1998; Sanders et al., 2011) and support ASCs (Aversa et al., 1993; Chatziandreou et al., 2017; Jego et al., 2003). IL-12 and the effector cytokine it supports, IFN\( \gamma \), which is produced by T cells, NK cells, and ILC1, are known to support ASC maintenance (Dubois et al., 1998; Miyauchi et al., 2016). However, mice deficient in each of these soluble cytokines or their receptors showed EFRs similar to their wild type (WT) controls at 7 dpi (**Suppl. Fig 3a**). B cells are also importantly affected through innate signals received via Toll-like receptors (TLRs). Influenza pathogen-associated molecular patterns (PAMPs) activate endosomal TLR3(Le Goffic et al., 2007) and TLR7 (Diebold et al., 2004), while TLR4 has a role in infection-mediated pathology (Nhu et al., 2010). However, infection of mice lacking TLR3, TLR4, or TLR7 had no significant effects on the number of total EF PBs and CD138+ EF PBs compared to their WT controls (**Suppl. Fig. 3b**). Thus, individual cytokines or innate signaling receptors appeared either not necessary or redundant for EFR development.
+
+The potential for redundancy of inflammatory signals contributing to the regulation of EFRs was addressed with mice double- deficient for both TLR adaptors, TRIF (Yamamoto et al., 2003) and MyD88 (DKO), which also transduces IL-1 and IL-18 signaling (Adachi et al., 1998). Indeed, the DKO mice showed strongly reduced EFRs at 7 dpi (**Fig. 4a**). TRIF single knockouts had EFRs comparable to that of wild type. While MyD88 single knockouts EFRs were variably reduced on average, these differences did not reach statistical significance (**Fig. 4a**).
+Importantly, while EFR-derived serum from WT mice provided robust protection against a lethal influenza virus challenge after adoptive transfer, DKO serum provided substantially reduced passive protection (Fig. 4b). Surprisingly, infection of another TLR-null model, through deletion of genes for TLR2 (Takeuchi et al., 1999), TLR4 (Hoshino et al., 1999) and a missense mutation of Unc93b (Tabeta et al., 2006) (TKO), showed EFRs similar to WT controls (Fig. 4a) and had no significant reduction in serum passive protective capacity compared to controls (Fig. 4c), despite slight reductions in CD138+ EF PBs at 7 dpi (Fig. 4a). This indicated a divergence in EFR dynamics mediated by the method of TLR abrogation.
+
+To distinguish potential B cell extrinsic from intrinsic effects of TLR signaling on EFR induction, mixed bone marrow irradiation chimeras, in which only B cells lacked either MyD88 plus TRIF (DKO BMC) or all TLRs (TKO BMC), were infected with influenza and analyzed at 7 dpi (Fig. 5a). Both the DKO and the TKO BMCs showed reduced EF and GC responses compared to WT chimera controls (Fig. 5b). The data indicate the importance of B cell intrinsic MyD88/TRIF and TLR signaling in regulating B cell responses overall. The differences in EF responses between the BMC and the total TLR-null chimeras is likely due to their differences in virus clearance. While virus titers at 10 dpi were no different between control and TLR-null BMC (Suppl. Fig. 4a), global DKO and TKO mice showed higher viral loads compared to wild type (Suppl. Fig 4b). These higher virus titers correlated with significantly larger EFRs in TLR-null mice (Suppl. Fig. 4c). Thus, B cell-intrinsic TLR signaling support early EFR formation, while additional B cell-extrinsic inflammatory signals, further drive EFR generation in a manner that correlates with pathogen burden,
+
+B cell intrinsic TLRs support B cell proliferation and survival
+To assess the direct effects of TLR signaling on B cell dynamics, negatively enriched naïve, follicular B cells were cultured with graded doses of anti-IgM (Fab)₂ and LPS, BCR and TLR agonists, respectively. Anti-IgM plus LPS co-treatment modestly enhanced viability compared to LPS alone and strongly supported B cell proliferation, as indicated by increased Ki67 expression compared to either treatment alone (**Suppl. Fig. 5a, b**). Co-stimulation also strongly induced IRF4 and IRF8, critical transcriptional regulators of the B cell fate (**Suppl. Fig. 5c, d**), while anti-IgM enhanced (and LPS inhibited) induction of IL21R expression, a cytokine receptor required for the generation of ASCs (Ozaki et al., 2002) (**Suppl. Fig. 5e**). Taken together, intrinsic TLR stimulation enhanced BCR-induced B cell entry into the cell cycle, promoted cell survival, and along with BCR signaling, maintained expression of IL21R.
+
+Canonical TLR signaling is known to integrate with the BCR (Pone et al., 2012; Schweighoffer et al., 2017) and with TNF superfamily receptors (He et al., 2010), suggesting that TLR signaling-deficient B cells are altered not only in their response to TLR agonists, but also to signals induced via the BCR, or via co-stimulation through CD40 or BAFFR. Indeed, stimulation of naïve, follicular DKO and TKO B cells pulsed with anti-IgM(Fab)₂ for three hours, followed by incubation with CD40L and BAFF for 48 hours (**Fig. 5c**) showed reduced viability (**Fig. 5d top**) and a near inability to enter the cell cycle, as measured by Ki67 staining (**Fig. 5d, bottom**), compared to WT controls. MyD88 and TRIF single KO B cells showed reductions in survival (**Suppl. Fig. 6a**) and proliferation (**Suppl. Fig. 6b**) similar to each other with frequencies approximately half between those of WT and DKO B cells, indicating that TRIF, along with MyD88, support BCR-mediated activation signals in a non-redundant, additive manner. Similar results were obtained with BCR-stimulation alone (**Suppl. Fig 6c, d**), demonstrating participation of the TLR signaling axis in antigen-mediated activation. Consistent with these data, analysis of non-EF/GC B cells from influenza infected DKO and TKO B cell chimeras revealed significantly reduced expression of Ki67 ex vivo, compared to controls at 5
+dpi, when IRF4 is highest within this uncommitted population during infection (**Fig. 5e** and not shown). Among the few HA-specific B cells present in DKO and TKO chimeras at this timepoint fewer expressed Ki67 (**Suppl. Fig. 7**). Thus, lack of antigen mediated integrated TLR signaling significantly reduced B cell survival and cell cycle entry consistent with earlier reports (Schweighoffer et al., 2017).
+
+*Lack of functional TLR signaling leads to abnormal BCR complex dynamics and transcriptional control*
+
+BCR-mediated calcium flux, an immediate readout of BCR crosslinking, was comparable between WT, DKO or TKO B cells (**Suppl. Fig. 8a**), indicating that TLR-null B cells were not merely defective overall. Effector protein phosphorylation downstream of the BCR showed minor pre-treatment differences (**Suppl. Fig. 8b**), but activation of mitogenic pathway MAPK p38 (Khiem et al., 2008) and the pro-inflammatory NFκB1 (Liu et al., 1991) also were roughly similar following BCR stimulation for 30 min (**Suppl. Fig. 8c, d**). Somewhat surprisingly, Syk phosphorylation, a major activation node for several BCR-mediated signaling pathways (Kurosaki et al., 1994), and phosphorylation of the pro-growth regulator mTOR (Donahue and Fruman, 2007) were increased in TLR-signaling deficient B cells compared to WT (**Supp. Fig. 8e, f**). Together, these data indicate that TLR signaling defects have little effects on early induction of IgM-BCR mediated signal transduction pathways.
+
+In contrast, the absence of TLR signaling led to enhanced surface IgD expression in HA-specific B cells *in vivo* compared to controls (**Suppl. Fig. 9a**). Indeed, loss of surface IgD after anti-IgM plus BAFF/CD40L stimulation was strongly attenuated in DKO and TKO B cells relative to WT (**Suppl. Fig. 9b**). Along with the enhanced induction of cell death and the inability to proliferate to antigen-mediated activation, despite apparently normal early downstream BCR
+signal transduction, TLR-signaling deficient B cells resembled anergic B cells (Goodnow et al., 1988).
+
+IRF4 is upregulated proportionately to BCR signaling strength (Ochiai et al., 2013). Consistent with that, ex vivo analysis of EF plasmablasts showed their distinct higher expression of IRF4 and intermediate expression of IRF8 compared to non-EFR B cells (**Fig. 6a, left**). *While ex vivo* baseline levels of IRF4 in naïve B cells were similar between the strains (**Fig. 6b, left**), non-differentiated B cells from DKO and TKO chimeras expressed significantly less IRF4 and IRF8 than WT at 5 dpi (**Fig. 6a, right**), indicating defects in IRF4 upregulation just before nascent EFRs form. In vitro IgM-BCR stimulation increased IRF4 and IRF8 expression in B cells from WT mice in an anti-IgM dose-dependent manner in the presence of CD40L and BAFF (**Fig. 6b right, Fig. 6c**). Strikingly, B cells from DKO and TKO mice failed to upregulate IRF4 under these conditions (**Fig. 6b right, Fig. 6c**), while IRF8 expression remained similar in all strains (**Fig. 6b right, Fig. 6d**). The data thus indicate defective BCR-mediated IRF4 induction in the absence of TLRs.
+
+NF-kB c-Rel is known to promote IRF4 expression upon nuclear re-localization and is downstream of both BCR and TLR4 (Grumont and Gerondakis, 2000). Strong, BCR dose-dependent stimulation-induced reductions in cytoplasmic c-Rel, inferring translocation of c-Rel to the nucleus, were seen in WT but much less so in TLR-signaling deficient B cells by flow cytometry as early as 30 min post stimulation (**Fig. 6e, Suppl. Fig. 10a**). Consistent with that result, WT but not DKO nor TKO B cells showed significant accumulation of c-Rel in the nucleus 1h but not 2h after anti-IgM and LPS stimulation, as assessed by ELISA on isolated nuclear-fractions (**Suppl. Fig. 10b, c**, concomitant with significant increases in total c-Rel expression (**Suppl. Fig 10d**). However, this delayed normalization in BCR-induced c-Rel expression was short-lived, as sustained c-Rel expression, which is associated with initialization of the B cell differentiation program (Roy et al., 2019), remained drastically lower in B cells lacking TLR-
+signaling than in WT B cells 48h after anti-IgM pulse (**Fig. 6f**). Thus, even in the absence of deliberate addition of a TLR agonist, B cells require the presence of TLRs for proper activation of the c-Rel circuitry and for the long-term maintenance of c-Rel expression in response to antigen-mediated stimulation.
+
+*Reconstitution of EFRs during influenza immunization through LPS adjuvant*
+
+Since both B cell-intrinsic and -extrinsic TLR signals influenced EFR magnitude and kinetics, we tested whether LPS, a TLR4 agonist that initiates both MyD88 and TRIF signaling, could overcome the lack of EFRs induction after s.c. immunization with influenza virions in alum (**Fig. 3**). Indeed, C57BL/6 mice inoculated with influenza in alum plus LPS and provided with repeated LPS boosts thereafter (Ag+LPS; **Fig. 7a**), showed increased total B cells, GC B cells, and EF PBs compared to mice receiving influenza in alum alone (Ag Only) (**Fig. 7b**). Importantly, the number of HA-binding B cells were twice as high than in mice receiving antigen/alum alone (**Fig. 7c**), with several-fold increases of HA B cells in the EFR but not GC compartment (**Fig. 7c**). Thus, indicating that TLR activation not only increased expansion of antigen-specific B cells but preferentially shunted them towards an EFR fate. HA B cells from Ag+LPS mice were mostly positive for Ki67 and CD138, IRF4hi IRF8int., similar to EF PBs from influenza-infected mice (**Fig. 7d**, e). This level of EFR polarization was not seen in Ag Only mice (**Fig. 7d**, e). HA B cells from Ag+LPS immunized mice also showed improved survival compared to Ag Only mice (**Suppl. Fig. 11a**), consistent with results seen after infection, where HA-specific B cells from DKO and TKO mice showed much lower ratios of live/dead cells compared to those of WT mice (**Suppl. Fig. 11b**). Thus, sustained TLR-mediated inflammation in the presence of antigen leads to greater expansion of antigen-specific B cells and polarizes them towards the EFR fate.
+Recent reports suggest that increased antigen valency (Kato et al., 2020) and antigen availability (Glaros et al., 2021) bias B cells towards a plasmablast fate. Given the above results, we asked how B cell fate dynamics and EFR-derived antibody functionality is affected by repeated antigen exposure with or without TLR agonist provision. For that, all mice were primed with influenza and LPS to ensure equivalent initiation of LN activation (Denton et al., 2022), followed by two additional boosts with antigen alone (Ag Boosted), or antigen plus LPS (Ag+LPS Boosted) or LPS alone (LPS Boosted) as a control (**Fig. 8a**). Both Ag Boosted and Ag+LPS Boosted mice had similar frequencies of HA B cells in the draining LN (**Fig. 8b**), and similar frequencies Ki67+ cells (**Fig. 8c**). However, HA B cells from Ag Boosted mice significantly polarized towards a GC fate (**Fig. 8d**), while HA B cells from Ag+LPS Boosted mice polarized significantly towards EFRs (**Fig. 8e**), indicating that despite repeated antigen inoculations, continued TLR stimulation was required for B cell development towards an EFR fate.
+
+Ag+LPS Boosted mice had the highest levels of serum anti-influenza antibodies (**Fig. 8f**), demonstrating that increased EFRs correlated with enhanced antigen-specific antibody responses compared to a GC-biased response at 10 days post-prime. To determine whether the increased in IgG levels correlated with increased serum passive protective capacity, pooled serum from each boosted group was transferred to naïve animals, who were subsequently challenged with a lethal dose of influenza. Mice receiving Ag+LPS Boosted serum showed no mortality, in contrast to mice receiving Ag Boosted or LPS Boosted serum (**Fig. 8g**). Moreover, mice that received serum from Ag+LPS Boosted mice lost significantly less weight overall than mice receiving serum from Ag Boosted animals (**Fig. 8h**). Together, these data demonstrate that sustained TLR-mediated inflammation polarizes antigen-specific B cells towards the EFR, leading to faster and stronger increases in protective, antigen-specific serum antibodies.
+Discussion
+
+These studies demonstrate that TLR-mediated inflammatory signals direct antigen-specific B cells towards the formation of ASCs through EFRs and that EFR-derived antibodies induced after both influenza infection and following LPS-boosted immunization are functionally protective. Thus, EFRs triggered and supported by inflammatory stimuli can provide a high quality antibody response at a fraction of the time relative to GCs by taking a more direct route to becoming ASCs, forming actively secreting, hemagglutinin-specific plasmablasts during the first 7-14 days of influenza infection prior to formation of GCs.
+
+EFR development seems to be driven by specificities already present in the repertoire at the time of infection, including in a naïve repertoire (Kalinke et al., 1996; Paus et al., 2006; Roost et al., 1995). In support, high affinity interactions between the BCR and its cognate antigen can drive a B cell effector fate, while lower affinity interactions confers a predispositon for the GC (Paus et al., 2006). However, the presence of high avidity B cells alone unlikely explains B cell fate decisions, as we show here that GC formation dominated early B cell responses to influenza immunization, while EFR dominated responses after influenza infection in the same inbred mice. If antigen-BCR affinity alone drives polarization towards an ASC fate, then the presence of antigen alone, assuming optimal delivery, stability, etc., should have resulted in an appreciable expansion of the same high affinity clones into the EFR than we saw after infection. Together, the data presented here demonstrate the need for infection-induced inflammation as a critical addition that supports EFR development. Inflammation affected EFR induction in an intrinsic manner, as functional Toll-like receptor (TLR) signaling axes either through MyD88/TRIF or TLR2/4/Unc93b induced optimal activation of the NF-kB c-Rel:IRF4 pathway (Suppl. Fig 12, top), as well as in an extrinsic manner, where TLR-mediated inflammation drove expansion of antigen-specific B cells into the EFR over the GC (Suppl. Fig 12, bottom), perhaps through alterations of the LN stromal compartment (Denton et al., 2022).
+TLR stimulation leads to the activation of multiple gene programs, but a defect in NF-kB c-Rel nuclear localization and upregulation after BCR stimulation was specifically observed in DKO and TKO B cells, along with suboptimal survival and the inability to proliferate or induce IRF4 expression. Additionally, the TLR adaptor TRIF, which has not been shown previously to influence BCR-mediated activation, was demonstrated here to contribute equally and non-redundantly with MyD88 towards B cell survival and proliferation after anti-IgM treatment. The observed defect in IRF4 upregulation in TLR-null B cells is consistent with previous studies demonstrating the dependence of IRF4 induction on c-Rel nuclear translocation after both, TLR4 and BCR activation (Grumont and Gerondakis, 2000). Delayed normalization of BCR-mediated c-Rel localization in TLR-null B cells did occur two hours after initial stimulation. Given that c-Rel has multiple c-terminal phosphorylation sites (Harris et al., 2006), perhaps TLR components are required for an optimal phosphorylation signature in addition to release of c-Rel from IkBs. Indeed, it was observed that the regulatory activity of c-Rel carrying a truncated c-terminus was severely altered, despite functional dimerization, nuclear localization, and DNA binding (Carrasco et al., 1998). Therefore, ablation of a functional TLR axis may dictate the nuclear activity of c-Rel, while maintaining localization potential. Further work is needed to determine how TLRs affect phosphorylation of the c-terminal trans-activation domain of c-Rel and how specific gene regulation is altered in their absence. Additionally, while total c-Rel levels did increase after 48 hours in TLR-null B cells, they were still significantly below levels observed in respective WT controls at every concentration of anti-IgM treatment measured. Therefore, IRF4 and c-Rel expression correlate and reaching a certain threshold of c-Rel seems required for the optimal induction of IRF4 in B cells. Indeed, c-Rel dominates the NF-kB program of B cells after antigen-mediated activation (Roy et al., 2019), potentiating an activated clone for several rounds of proliferation and enabling access to genes associated with terminal differentiation into plasma cells.
+Vaccination with antigen in alum, whether used as a prime or a boost, led to an expansion of antigen-specific clones primarily within the GC compartment, generating protracted serum antibody responses that were less protective at early times after immunization compared to the EFR dominated responses generated via antigen plus TLR agonist boosting. This suggests that increasing antigen valency (Kato et al., 2020) and/or amounts (Glaros et al., 2021) alone have a limited capacity to direct B cells towards early plasmablast responses following vaccinations, in contrast to vaccines adjuvanted with TLR agonists. The question that remains to be resolved is whether a drive towards EFR comes at the cost of effective GC-induced humoral immunity. Indeed, a recent study noted that TLR activation can worsen the quality of antigen-specific antibody responses due to a lack of GC-mediated affinity maturation (Akkaya et al., 2018), measuring a hallmark anti-hapten antibody response, where antibody affinity for the hapten increases over time as GCs mature and affinity maturation takes place (Foote and Milstein, 1991). However, increases in serum antibody affinities over time were not observed following infection with vesicular stomatitis virus (Kalinke et al., 1996; Roost et al., 1995) and high affinity, germline-encoded antibodies to hemagglutinin were induced early after influenza inoculation (Kavaler et al., 1990). Thus, the level of EFR-derived antibody avidity is contextual and relies on the inherent specificities of the host’s pre-infection repertoire, while the initiation, kinetics, and magnitude of the EFR rely on TLR-mediated inflammatory signals. The data are consistent with findings that memory B cells upon reactivation preferentially form EFR rather than enter GCs, even during heterotypic responses (Wong et al., 2020). Given the predominance of inflammatory signals during acute infection, this allows for antigen-specific B cells to be shunted into EFR for rapid production of protective antibodies to infections. The data also provide a mechanistic explanation for the association of EFRs with severe COVID-19 infection (Woodruff et al., 2020), and increased EFR-derived auto-antibody production with chronic inflammation, where a positive feed-forward loop may induce antibody-mediated pathology, driving enhanced inflammation, and thus further supporting ongoing EFRs. Even
+when the host may carry a highly restricted BCR repertoire, TLR activation may allow for EFR-derived antibodies of low affinity to contribute towards protection, without which these antibodies’ respective B cell clones would not reach the threshold of differentiation, nor activation. We conclude that B cell response fates are critically regulated by the innate, inflammatory milieu during antigen encounter.
+
+METHODS
+
+Mice. Male and female 8- to 12-wk-old C57BL/6 (WT; CD45.2), B6.SJL-Ptprrca Pepcb/BoyJ (CD45.1), B cell–deficient (μMT) mice as well asTNFAR1/2 KO, IFN-gamma KO, IL-12 KO, CD19-Cre IFNAR KO, IL-1R KO, TLR3 KO, TLR4 KO, TLR7 KO were commercially obtained (The Jackson Laboratories). Breeding pairs of MyD88/TRIF DKO and TLR2/4/unc93b TKO mouse strains were gifts from Dr. Barton (UC Berkeley). Breeding pairs of S100A9 KO mice were a kind gift of Dr. Rafatellu (UC San Diego).
+
+Mixed bone marrow (BM) chimeras were generated by adoptively transferring 5x \(10^6\) total mixed BM cells from slgM-deficient (CD45.2, 75%) and either C57BL/6 (WT; CD45.2), MyD88/TRIF double knockout (CD45.2), or TLR4/TLR2/Unc93b triple knockout (CD45.2) BM (25%) into 5-6 week-old B6.SJL-Ptprrca Pepcb/BoyJ (CD45.1) mice, lethally irradiated by exposure to a gamma irradiation source 24 h prior to transfer. Chimeras were rested for at least 6 weeks before infection and analysis.
+
+Infections, and immunizations. Mice were anesthetized with isoflurane and infected intranasally with a sublethal dose (10 PFU/ml) of influenza A/Puerto Rico/8/34 (A/PR8) in 40 μl volumes in PBS. Virus was grown in hen eggs as previously outlined(Doucett et al., 2005) and each virus batch was titrated for its effect on mice prior to use. Specifically, sublethal infection doses were chosen that incurred no more than 20% weight loss. For immunizations, mice were
+inoculated subcutaneously with 1x10^7 PFU A/PR8 in a 50:50 alum to PBS mixture. For some experiments immunizations were supplemented with 3μg LPS, or mice were in addition boosted repeatedly with 1x10^6 PFU A/PR8 and 3μg LPS in PBS or PBS alone as indicated.
+
+Adoptive serum transfer for passive protection. Indicated strains of mice were infected with 10 PFU A/PR8. Blood from terminally anesthetized mice at 10 dpi was collected via cardiac puncture and spun down for serum separation. Serum from each strain was pooled and naïve C57BL/6 mice were subsequently injected i.v. with a mixture of 50μl pooled serum and 150μl 1x PBS. These mice were then inoculated i.n. with 100 PFU A/PR8 one day later and measured for weight loss.
+
+Magnetic B cell enrichment. Splenic B cells were treated with Fc Block (anti-mouse CD16/32, clone 2.4.G2) and were then enriched using a mixture of biotinylated Abs (anti-CD90.2 (30-H12), anti-CD4 (GK1.5), anti-CD8a (53-6.7), anti-Gr-1 (RB6-8C5), anti-CD11b (M1/70), anti-NK1.1 (PK136), anti-F4/80 (BM8), anti-CD5 (53-7.3), anti-CD9 (MZ3), anti-CD138 (281-2) and anti-biotin MicroBeads (Miltenyi Biotec). Nylon-filtered stained splenocytes were separated using autoMACS (Miltenyi Biotec). Purities of enriched mouse B cells were >98% as determined by subsequent FACS analysis.
+
+Flow cytometry and phospho-flow. Single-cell suspensions from mediastinal lymph nodes (medLN) were made and labeled for phenotyping as previously outlined(Doucett et al., 2005). Briefly, after Fc receptor block with anti-CD16/32 (5 mg/ml for 20 min on ice), cells were stained with the following antibody-fluorophore conjugates at temperatures and times according to manufacturer/provider: HA-PE and HA-APC oligomers (kindly provided by Dr. Frances Lund, UAB), BV786 anti-CD19 (1D3) (BD Bioscience), APC-eFluor780 anti-CD45R (RA3-6B2), PE-Dazzle 594 anti-CD38 (90) (both Thermo Fisher), BV711 anti-CD24 (M1/69), BV605 anti-CD138 (281-2) (both Biolegend), eFluor450 anti-GL-7 (GL7), PE or PE/Cy7 anti-IRF4 (3E4), PerCP-
+eFluor710 anti-IRF8 (V3GYWCH), eFluor450 anti-Ki67 (SolA15) (all Thermo Fisher), FITC anti-IgM (331, in-house), and BV650 anti-IgD (11-26c.2a) (Biolegend). For a non-B cell “dump”, the following antibodies on AlexaFluor 700 were used: anti-CD90.2, anti-CD4, anti-CD8a, anti-Gr-1, anti-CD11b, anti-NK1.1, anti-F4/80 (all Thermo Fisher). The Foxp3 Staining Buffer Set (Thermo Fisher) was used for fixation and permeabilization of cells for staining of transcription factors according to manufacturer’s protocol. For cytoplasmic only staining, Cytofix/cytoperm buffer set (BD Biosciences) was used according to manufacturer’s protocol. For phospho-flow, APC anti-p-Syk (moch1ct), PerCP-eFluor710 anti-p-p38 (4NIT4KK), PE/Cy7 anti-p-mTOR (MRRBY), and PE anti-p-p65 (B33B4WP) were stained according to manufacturer’s protocol (Thermo Fisher). B cells from 7 dpi medLN were sorted by flow cytometry for ELISPOT using pooled antibodies for dump channel, anti-CD19, anti-CD45R, anti-CD24, and anti-CD38. Purity of sorted cells was assessed immediately afterwards (>96%).
+
+In vitro B cell cultures. Magnetically enriched B cells were cultured at 5 x 10^6 cells/ml at 37 °C. Cells were incubated with anti-IgM (Fab)_2 and/or LPS in culture media at the indicated concentrations for 30 minutes, one, two, and three hours. Three-hour anti-IgM-pulsed B cells were washed twice with PBS, and then cultured in culture media containing 200 ng/mlCD40L (Peprotech) and 5 ng/ml BAFF (R&D Systems) in 96-well round-bottom plates for 48 hours at 5% CO_2. Subsequent flow cytometric analysis was done using Fixable Aqua, PE anti-c-Rel (1RELAH5) (both Thermo Fisher), BV786 anti-CD19, eFluor450 anti-Ki67, PE/Cy7 anti-IRF4, PerCP-eFluor710 anti-IRF8 and APC anti-IL-21R (4A9) (all eBioscience).
+
+ELISPOT. A/PR8-specific Ig-secreting cells were measured. Briefly, ELISPOT plates were coated with 500 HAU of purified A/PR8 overnight, then blocked for non-specific binding for 1 hour. Serial dilutions of FACS-sorted EF PBs and pooled non-EF B cells were incubated overnight at 37 °C. Ab-secreting cells (ASC) were revealed with goat anti-mouse IgM, IgG-biotin
+(Southern Biotech) followed by SA-HRP (Vector Laboratories) and 3-amino-9-ethylcarbazole (Sigma-Aldrich).
+
+Nuclear fraction ELISA. c-Rel nuclear localization was measured. Briefly, nuclear and cytoplasmic protein fractions were extracted from cultured, purified B cells using NE-PER Nuclear and Cytoplasmic Extraction (Thermo Fisher) according to manufacturer’s protocol. ELISA plates were coated at 4 \( \mu \)g/ml dilution of polyclonal anti-c-Rel (Thermo Fisher) overnight, then blocked for non-specific binding for 1 hour. Bound c-Rel was detected using 4 \( \mu \)g/ml monoclonal anti-c-Rel (1RELAH5). Binding was revealed by SA-HRP (Vector Laboratories).
+
+Viral-load rtPCR. Infected mice were euthanized and lung tissue was extracted and homogenized using Gentle Macs (Miltenyi) in 1 ml PBS. Tissue was pelleted and supernatant was aliquoted and frozen. Viral RNA was purified from aliquots using the QIAamp viral RNA mini-kit (Qiagen). Presence of influenza was detected through amplification of influenza M gene using rtPCR. Primers used were AM-151 (5'-CATGCAATGGCTAAAGACAAGACC-3') and AM-397 (5'-AAGTGCACCAGCAGAATAACTGAG-3') and primer/probe AM-245 (6FAM-5'-CTGCAGCGTAGAGCTTTGTCCAAAATG-3'-TAMRA). Reverse transcription and amplication were done using TaqPath Multiplex Master Mix (Thermo Fisher). Samples were quantified to a standard of A/PR8 virus stock.
+
+Calcium flux assay. To measure changes in cellular calcium concentrations, B cells were stained with 2\( \mu \)M cell-permeant Fluor-3 and 4\( \mu \)M FuraRed (both Thermo Fisher) according to manufacturer’s protocol and stimulated with 10 \( \mu \)g/ml anti-IgM(fab)\(_2\) fragments prior to analysis by flow cytometry. The ratio of the calcium-excitible (Fluor3) and calcium-quenched (FuraRed) dyes were calculated to determine free-intracellular concentrations.
+References
+
+Adachi, O., Kawai, T., Takeda, K., Matsumoto, M., Tsutsui, H., Sakagami, M., Nakanishi, K., and Akira, S. (1998). Targeted disruption of the MyD88 gene results in loss of IL-1- and IL-18-mediated function. Immunity 9, 143-150.
+
+Akkaya, M., Akkaya, B., Kim, A.S., Miozzo, P., Sohn, H., Pena, M., Roesler, A.S., Theall, B.P., Henke, T., Kabat, J., et al. (2018). Toll-like receptor 9 antagonizes antibody affinity maturation. Nat Immunol 19, 255-266.
+
+Aversa, G., Punnonen, J., and de Vries, J.E. (1993). The 26-kD transmembrane form of tumor necrosis factor alpha on activated CD4+ T cell clones provides a costimulatory signal for human B cell activation. J Exp Med 177, 1575-1585.
+
+Browne, E.P. (2011). Toll-like receptor 7 controls the anti-retroviral germinal center response. PLoS Pathog 7, e1002293.
+
+Carrasco, D., Cheng, J., Lewin, A., Warr, G., Yang, H., Rizzo, C., Rosas, F., Snapper, C., and Bravo, R. (1998). Multiple hemopoietic defects and lymphoid hyperplasia in mice lacking the transcriptional activation domain of the c-Rel protein. J Exp Med 187, 973-984.
+
+Chatziandreou, N., Farsakoglu, Y., Palomino-Segura, M., D'Antuono, R., Pizzagalli, D.U., Sallusto, F., Lukacs-Kornek, V., Uggucioni, M., Corti, D., Turley, S.J., et al. (2017). Macrophage Death following Influenza Vaccination Initiates the Inflammatory Response that Promotes Dendritic Cell Function in the Draining Lymph Node. Cell Rep 18, 2427-2440.
+
+Coro, E.S., Chang, W.L., and Baumgarth, N. (2006). Type I IFN receptor signals directly stimulate local B cells early following influenza virus infection. J Immunol 176, 4343-4351.
+
+Denton, A.E., Dooley, J., Cinti, I., Silva-Cayetano, A., Fra-Bido, S., Innocentin, S., Hill, D.L., Carr, E.J., McKenzie, A.N.J., Liston, A., and Linterman, M.A. (2022). Targeting TLR4 during vaccination boosts MadCAM-1(+) lymphoid stromal cell activation and promotes the aged germinal center response. Sci Immunol 7, eabk0018.
+
+Di Niro, R., Lee, S.J., Vander Heiden, J.A., Elsner, R.A., Trivedi, N., Bannock, J.M., Gupta, N.T., Kleinstein, S.H., Vigneault, F., Gilbert, T.J., et al. (2015). Salmonella Infection Drives Promiscuous B Cell Activation Followed by Extrafollicular Affinity Maturation. Immunity 43, 120-131.
+
+Diebold, S.S., Kaisho, T., Hemmi, H., Akira, S., and Reis e Sousa, C. (2004). Innate antiviral responses by means of TLR7-mediated recognition of single-stranded RNA. Science 303, 1529-1531.
+
+Donahue, A.C., and Fruman, D.A. (2007). Distinct signaling mechanisms activate the target of rapamycin in response to different B-cell stimuli. Eur J Immunol 37, 2923-2936.
+
+Doucett, V.P., Gerhard, W., Owler, K., Curry, D., Brown, L., and Baumgarth, N. (2005). Enumeration and characterization of virus-specific B cells by multicolor flow cytometry. J Immunol Methods 303, 40-52.
+
+Dubois, B., Massacrier, C., Vanbervliet, B., Fayette, J., Brière, F., Banchereau, J., and Caux, C. (1998). Critical role of IL-12 in dendritic cell-induced differentiation of naïve B lymphocytes. J Immunol 161, 2223-2231.
+
+Foote, J., and Milstein, C. (1991). Kinetic maturation of an immune response. Nature 352, 530-532.
+
+Glaros, V., Rauschmeier, R., Artemov, A.V., Reinhardt, A., Ols, S., Emmanouilidi, A., Gustafsson, C., You, Y., Mirabello, C., Bjorklund, A.K., et al. (2021). Limited access to antigen drives generation of early B cell memory while restraining the plasmablast response. Immunity 54, 2005-2023 e2010.
+
+Goodnow, C.C., Crosbie, J., Adelstein, S., Lavoie, T.B., Smith-Gill, S.J., Brink, R.A., Pritchard-Briscoe, H., Wotherspoon, J.S., Loblay, R.H., Raphael, K., and et al. (1988). Altered immunoglobulin expression and functional silencing of self-reactive B lymphocytes in transgenic mice. Nature 334, 676-682.
+
+Grumont, R.J., and Gerondakis, S. (2000). Rel induces interferon regulatory factor 4 (IRF-4) expression in lymphocytes: modulation of interferon-regulated gene expression by rel/nuclear factor kappaB. J Exp Med 191, 1281-1292.
+Harris, J., Olière, S., Sharma, S., Sun, Q., Lin, R., Hiscott, J., and Grandvaux, N. (2006). Nuclear accumulation of cRel following C-terminal phosphorylation by TBK1/IKK epsilon. J Immunol 177, 2527-2535.
+
+Hayden, F.G., Fritz, R., Lobo, M.C., Alvord, W., Strober, W., and Straus, S.E. (1998). Local and systemic cytokine responses during experimental human influenza A virus infection. Relation to symptom formation and host defense. J Clin Invest 101, 643-649.
+
+He, B., Santamaria, R., Xu, W., Cols, M., Chen, K., Puga, I., Shan, M., Xiong, H., Bussel, J.B., Chiu, A., et al. (2010). The transmembrane activator TACI triggers immunoglobulin class switching by activating B cells through the adaptor MyD88. Nat Immunol 11, 836-845.
+
+Heer, A.K., Shamshiev, A., Donda, A., Uematsu, S., Akira, S., Kopf, M., and Marsland, B.J. (2007). TLR signaling fine-tunes anti-influenza B cell responses without regulating effector T cell responses. J Immunol 178, 2182-2191.
+
+Hoshino, K., Takeuchi, O., Kawai, T., Sanjo, H., Ogawa, T., Takeda, Y., Takeda, K., and Akira, S. (1999). Cutting edge: Toll-like receptor 4 (TLR4)-deficient mice are hyporesponsive to lipopolysaccharide: evidence for TLR4 as the Lps gene product. J Immunol 162, 3749-3752.
+
+Jego, G., Palucka, A.K., Blanck, J.-P., Chalouni, C., Pascual, V., and Banchereau, J. (2003). Plasmacytoid Dendritic Cells Induce Plasma Cell Differentiation through Type I Interferon and Interleukin 6. Immunity 19, 225-234.
+
+Kalinke, U., Bucher, E.M., Ernst, B., Oxenius, A., Roost, H.P., Geley, S., Kofler, R., Zinkernagel, R.M., and Hengartner, H. (1996). The role of somatic mutation in the generation of the protective humoral immune response against vesicular stomatitis virus. Immunity 5, 639-652.
+
+Kasturi, S.P., Skountzou, I., Albrecht, R.A., Koutsonanos, D., Hua, T., Nakaya, H.I., Ravindran, R., Stewart, S., Alam, M., Kwissa, M., et al. (2011). Programming the magnitude and persistence of antibody responses with innate immunity. Nature 470, 543-547.
+
+Kato, Y., Abbott, R.K., Freeman, B.L., Haupt, S., Groschel, B., Silva, M., Menis, S., Irvine, D.J., Schief, W.R., and Crotty, S. (2020). Multifaceted Effects of Antigen Valency on B Cell Response Composition and Differentiation In Vivo. Immunity 53, 548-563 e548.
+
+Kavalier, J., Caton, A.J., Staudt, L.M., Schwartz, D., and Gerhard, W. (1990). A set of closely related antibodies dominates the primary antibody response to the antigenic site CB of the A/PR/8/34 influenza virus hemagglutinin. J Immunol 145, 2312-2321.
+
+Khiem, D., Cyster, J.G., Schwarz, J.J., and Black, B.L. (2008). A p38 MAPK-MEF2C pathway regulates B-cell proliferation. Proc Natl Acad Sci U S A 105, 17067-17072.
+
+Kurosaki, T., Takata, M., Yamanashi, Y., Inazu, T., Taniguchi, T., Yamamoto, T., and Yamamura, H. (1994). Syk activation by the Src-family tyrosine kinase in the B cell receptor signaling. J Exp Med 179, 1725-1729.
+
+Lam, J.H., and Baumgarth, N. (2019). The Multifaceted B Cell Response to Influenza Virus. J Immunol 202, 351-359.
+
+Le Goffic, R., Pothlichet, J., Vitour, D., Fujita, T., Meurs, E., Chignard, M., and Si-Tahar, M. (2007). Cutting Edge: Influenza A virus activates TLR3-dependent inflammatory and RIG-I-dependent antiviral responses in human lung epithelial cells. J Immunol 178, 3368-3372.
+
+Liu, J.L., Chiles, T.C., Sen, R.J., and Rothstein, T.L. (1991). Inducible nuclear expression of NF-kappa B in primary B cells stimulated through the surface Ig receptor. J Immunol 146, 1685-1691.
+
+Loetsch, C., Warren, J., Laskowski, A., Vazquez-Lombardi, R., Jandl, C., Langley, D.B., Christ, D., Thorburn, D.R., Ryugo, D.K., Sprent, J., et al. (2017). Cytosolic Recognition of RNA Drives the Immune Response to Heterologous Erythrocytes. Cell Rep 21, 1624-1638.
+
+Miyauchi, K., Sugimoto-Ishige, A., Harada, Y., Adachi, Y., Usami, Y., Kaji, T., Inoue, K., Hasegawa, H., Watanabe, T., Hijikata, A., et al. (2016). Protective neutralizing influenza antibody response in the absence of T follicular helper cells. Nat Immunol 17, 1447-1458.
+Nhu, Q.M., Shirey, K., Teijaro, J.R., Farber, D.L., Netzle-Arnett, S., Antalis, T.M., Fasano, A., and Vogel, S.N. (2010). Novel signaling interactions between proteinase-activated receptor 2 and Toll-like receptors in vitro and in vivo. Mucosal Immunol 3, 29-39.
+
+Ochiai, K., Maienschein-Cline, M., Simonetti, G., Chen, J., Rosenthal, R., Brink, R., Chong, A.S., Klein, U., Dinner, A.R., Singh, H., and Sciammas, R. (2013). Transcriptional regulation of germinal center B and plasma cell fates by dynamical control of IRF4. Immunity 38, 918-929.
+
+Ozaki, K., Spolski, R., Feng, C.G., Qi, C.F., Cheng, J., Sher, A., Morse, H.C., 3rd, Liu, C., Schwartzberg, P.L., and Leonard, W.J. (2002). A critical role for IL-21 in regulating immunoglobulin production. Science 298, 1630-1634.
+
+Paus, D., Phan, T.G., Chan, T.D., Gardam, S., Basten, A., and Brink, R. (2006). Antigen recognition strength regulates the choice between extrafollicular plasma cell and germinal center B cell differentiation. J Exp Med 203, 1081-1091.
+
+Pone, E.J., Zhang, J., Mai, T., White, C.A., Li, G., Sakakura, J.K., Patel, P.J., Al-Qahtani, A., Zan, H., Xu, Z., and Casali, P. (2012). BCR-signalling synergizes with TLR-signalling for induction of AID and immunoglobulin class-switching through the non-canonical NF-kappaB pathway. Nat Commun 3, 767.
+
+Roost, H.P., Bachmann, M.F., Haag, A., Kalinke, U., Pliska, V., Hengartner, H., and Zinkernagel, R.M. (1995). Early high-affinity neutralizing anti-viral IgG responses without further overall improvements of affinity. Proc Natl Acad Sci U S A 92, 1257-1261.
+
+Rothaeusler, K., and Baumgarth, N. (2010). B-cell fate decisions following influenza virus infection. Eur J Immunol 40, 366-377.
+
+Roy, K., Mitchell, S., Liu, Y., Ohta, S., Lin, Y.S., Metzig, M.O., Nutt, S.L., and Hoffmann, A. (2019). A Regulatory Circuit Controlling the Dynamics of NFkappaB cRel Transitions B Cells from Proliferation to Plasma Cell Differentiation. Immunity 50, 616-628 e616.
+
+Sanders, C.J., Doherty, P.C., and Thomas, P.G. (2011). Respiratory epithelial cells in innate immunity to influenza virus infection. Cell Tissue Res 343, 13-21.
+
+Schweighoffer, E., Nys, J., Vanes, L., Smithers, N., and Tybulewicz, V.L.J. (2017). TLR4 signals in B lymphocytes are transduced via the B cell antigen receptor and SYK. J Exp Med 214, 1269-1280.
+
+Tabeta, K., Hoebe, K., Janssen, E.M., Du, X., George, P., Crozat, K., Mudd, S., Mann, N., Sovath, S., Goode, J., et al. (2006). The Unc93b1 mutation 3d disrupts exogenous antigen presentation and signaling via Toll-like receptors 3, 7 and 9. Nat Immunol 7, 156-164.
+
+Takeuchi, O., Hoshino, K., Kawai, T., Sanjo, H., Takada, H., Ogawa, T., Takeda, K., and Akira, S. (1999). Differential roles of TLR2 and TLR4 in recognition of gram-negative and gram-positive bacterial cell wall components. Immunity 11, 443-451.
+
+Tian, M., Hua, Z., Hong, S., Zhang, Z., Liu, C., Lin, L., Chen, J., Zhang, W., Zhou, X., Zhang, F., et al. (2018). B Cell-Intrinsic MyD88 Signaling Promotes Initial Cell Proliferation and Differentiation To Enhance the Germinal Center Response to a Virus-like Particle. J Immunol 200, 937-948.
+
+Trivedi, N., Weisel, F., Smita, S., Joachim, S., Kader, M., Radhakrishnan, A., Clouser, C., Rosenfeld, A.M., Chikina, M., Vigneault, F., et al. (2019). Liver Is a Generative Site for the B Cell Response to Ehrlichia muris. Immunity 51, 1088-1101 e1085.
+
+Tsai, S.Y., Segovia, J.A., Chang, T.H., Morris, I.R., Berton, M.T., Tessier, P.A., Tardif, M.R., Cesaro, A., and Bose, S. (2014). DAMP molecule S100A9 acts as a molecular pattern to enhance inflammation during influenza A virus infection: role of DDX21-TRIF-TLR4-MyD88 pathway. PLoS Pathog 10, e1003848.
+
+Wong, R., Belk, J.A., Govero, J., Uhrlaub, J.L., Reinartz, D., Zhao, H., Errico, J.M., D'Souza, L., Ripperger, T.J., Nikolic-Zugich, J., et al. (2020). Affinity-Restricted Memory B Cells Dominate Recall Responses to Heterologous Flaviviruses. Immunity 53, 1078-1094.e1077.
+
+Woodruff, M.C., Ramonell, R.P., Nguyen, D.C., Cashman, K.S., Saini, A.S., Haddad, N.S., Ley, A.M., Kyu, S., Howell, J.C., Ozturk, T., et al. (2020). Extrafollicular B cell responses correlate with neutralizing antibodies and morbidity in COVID-19. Nat Immunol 21, 1506-1516.
+Xu, H., Chaudhri, V.K., Wu, Z., Biliouris, K., Dienger-Stambaugh, K., Rochman, Y., and Singh, H. (2015). Regulation of bifurcating B cell trajectories by mutual antagonism between transcription factors IRF4 and IRF8. Nat Immunol 16, 1274-1281.
+
+Yamamoto, M., Sato, S., Hemmi, H., Hoshino, K., Kaisho, T., Sanjo, H., Takeuchi, O., Sugiyama, M., Okabe, M., Takeda, K., and Akira, S. (2003). Role of adaptor TRIF in the MyD88-independent toll-like receptor signaling pathway. Science 301, 640-643.
+
+Acknowledgements
+
+This work was supported by research grants from the NIH/NIAID, R01AI117890, R01AI085568 and U19AI109962 and an institutional NIH training grant from the NIH/NHLBI, T-32 HL007013.
+
+We thank Ms. Zheng Luo and Jacqueline Dieter for expert technical support, Drs. Gregory Barton (UC Berkeley) and Manuela Raffatellu (UC San Diego) for mice, and Dr. Frances Lund (UAB) for HA-baits. We further thank Tracy Rourke of the California National Primate Research Center (UC Davis) for technical assistance with flow cytometry and the UC Davis TRACS personnel for animal care and husbandry.
+
+Author Contributions. J.H.L. designed and conducted experiments, analyzed data, and wrote the manuscript. N.B. designed and supervised experiments, data analysis, and wrote the manuscript.
+
+Competing Interests. No competing interests are declared by either author.
+Figure 1. Primary influenza infection induces strong early EFRs prior to GC formation.
+
+Shown are flow cytometric analyses of mediastinal lymph nodes (medLN) from C57BL/6 mice infected with influenza A/PR8 intra-nasally (i. n.) at seven days post-infection (dpi). (a) Identification of extrafollicular plasmablasts (EF PBs) and pre-GC/GC B cells by flow cytometry. (b) IgM and IgD expression on EF PBs, pre-GC/GC B cells, and non-EF/non-GC B cells. (c-e) C57BL/6 mice were infected and medLN were collected on the days specified, measuring B cell frequencies of total cells (c), pre-GC/GC frequency of B cells (d), EF frequency of B cells (e).
+Figure 2. EFRs generate influenza-specific antibody-secreting cells. (a) Influenza-specific ELISPOTS of sorted EF PBs and pooled non-EF cells for total Ig (left) and IgG2c (right). (b) Flow plots of HA-specific B cells using double HA-tetramer staining. (c-e) Time course of HA-specific B cell subsets during influenza infection as in (c-e), measuring frequency of HA-specific clones (i), HA-specific pre-GC/GC clones (j), and HA-specific EF PBs (k). Graphs are representative of two experiments (n>/=3). Error bars represent 95% confidence interval (CI), statistical significance determined by unpaired Student’s t-test with Welch’s correction. **: p<0.01
+Figure 3. Subcutaneous immunization with influenza and alum does not elicit EFRs. (a-e) C57BL/6 mice were immunized s.c. with 1x \(10^7\) PFU influenza A/PR8 in alum and inguinal LNs were analyzed on days indicated. (a) Flow plots comparing immunization to infection EF and GC formation. (b) Kinetics of EF and pre-GC/GC B cells compared to infection. (c) Fold-difference of EF and GC responses compared to infection. (d) Flow plots comparing HA-specific B cell populations during immunization and infection. (e) Kinetics of total (left) and proliferating (right) HA-specific B cells compared to infection. Graphs are representative of two experiments (n=4). Error bars represent 95% CI, statistical significance determined by one-way ANOVA (b, e). ****: p<0.0001
+Figure 4. Optimal EFR kinetics and protective antibodies require MyD88 and TRIF.
+
+Knockout and WT mice were infected with 10 PFU A/PR8 and medLNs were collected at 7 days post-infection (dpi). (a) Fold-difference of B cell subsets in TLR-deficient versus WT mice at 7 dpi. (b-c) Sera from influenza-infected MyD88/TRIF-deficient (DKO) mice (b) or TLR2/4/unc93b-deficient (TKO) mice (c) at 10 dpi were transferred to C57BL/6 mice prior to infection with a lethal dose (100 PFU) of influenza A/PR8 the next day. Shown is % change in weight over the course of infection. Graphs are representative of two or more experiments (n>/=3 (a), n=10 (b,c)). Error bars represent 95% CI, statistical significance determined by one-way ANOVA (a) and unpaired Student’s t-test with Welch’s correction. *: p<0.05, **: p<0.01, ***: p<0.001, ****: p<0.0001 or indicated in subfigures.
+Figure 5. BCR-mediated survival and proliferation are defective in the absence of TLR signaling. (a) Mixed bone-marrow chimeras (BMC) established with irradiated CD45.1 C57BL/6 host mice reconstituted with μMT donor BM and BM from either DKO or TKO, then infected with 10 PFU A/PR8 6 weeks later. (b) Quantification of DKO and TKO BMC compared to WT BMC controls of B cell subsets at 7 dpi. (c) Pooled splenic and LN B cells from WT, DKO, or TKO B cells negatively enriched (>98% purity) were pulsed with graded levels of anti-IgM for 3 hours, then stimulated with CD40L and BAFF for 48 hours. (d) Quantification of cell viability (top) and cell proliferation (bottom). (e) Ki67+ non-EF/GC B cells in chimeras from 5 dpi. Graphs are representative of two experiments (n>/=4). Error bars represent 95% CI, statistical significance determined by one-way ANOVA and unpaired Student’s t-test with Welch’s correction. *: p<0.05, **: p<0.001, ****: p<0.0001.
+Figure 6. Lack of functional TLR signaling leads to altered BCR complex dynamics and failure to upregulate IRF4. (a) Representative flow plots showing IRF4 and IRF8 expression in infected mice, highlighting clustering of EF PBs (left). Fold-difference in IRF4 and IRF8 of non-EF/GC B cells from chimeras at 5 dpi (right). (b) Pre-enrichment baseline of IRF4 and IRF8 in B cells of each strain (left) and representative IRF4 versus IRF8 flow plots from cells stimulated with indicated anti-IgM concentrations (right). Colored numbers in plots correspond to each like-colored axis. (c-d) Fold-change compared to non-stimulated WT B cells in IRF4 (c) and IRF8 expression (d) after treatment outlined in Fig. 5c. (e) Fold-change in cytoplasmic c-Rel
+measured by flow cytometry after 30-minute anti-IgM or LPS treatment. (f) Fold-differences in total c-Rel expression after a 3h anti-IgM pulse and 48h culture in complete media only. Error bars represent 95% CI, statistical significance determined by one-way ANOVA and unpaired Student’s t-test with Welch’s correction. *: p<0.05 **: p<0.01 ***: p<0.001, ****: p<0.0001. Stars in (g,h) are Student’s t-test comparison to respective WT control.
+Figure 7. Sustained TLR-mediated inflammation generates strong EFRs in the draining LN after immunization. (a) Mice were immunized s.c. with or without influenza in alum and with or without LPS, then boosted with either LPS or PBS on days specified, followed by analysis of draining LN. (b) Counts of major B cell subsets. (c) Quantification of HA-specific B cell subsets as in (b). (d) Flow plots of HA-specific B cells from each regimen in terms of proliferation and plasma cell differentiation (left) and IRF4 vs IRF8 signature (right, HA-sp. highlighted in red). (e) Quantification of HA-specific EF PBs, proliferation, and relative expression of IRF4. Graphs are representative of two experiments (n>/=4). Error bars represent 95% CI. Statistical significance determined by one-way ANOVA and unpaired Student’s t-test with Welch’s correction. *: p<0.05, **: p<0.01 ***: p<0.001, ****: p<0.0001.
+Figure 8. Repeated antigen exposure alone biases antigen-specific B cells towards a GC fate, requires sustained LPS exposure to polarize towards an EF fate. (a) Mice were immunized s.c. with influenza and LPS in alum, then boosted with antigen alone or antigen with LPS and LPS alone on days specified, followed by analysis of draining LN. (b-e) Quantification of total HA B cells (b), Ki67+ HA B cells (c), HA GC B cells (d) and HA EF PBs (e). (f) Concentration of influenza-specific serum IgG at 10 days post-prime. (g,h) Serum from primed/boosted mice at 10 days post-prime was transferred to C57BL/6 mice prior to infection with a lethal dose (100 PFU) of influenza A/PR8 the next day. Shown is survival probability (g) and percent change in weight (h) by average (left) and individually (right) over the course of infection. Graphs are representative of two experiments (n>/=7, g, h n=10). Error bars represent 95% CI. Statistical significance determined by one-way ANOVA and unpaired Student’s t-test with Welch’s correction. *: p<0.05, **: p<0.01 ***: p<0.001, ****: p<0.0001.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SUPPLFIGSPluSTextFinal.pdf

02aa991c737fa072591798edf4f86f075938ab38dd6da2c55da7d15affbd0ada/preprint/preprint.md

@@ -0,0 +1,543 @@
+Control of Polymers’ Amorphous-crystalline Transition for Hydrogel Bioelectronics Miniaturization and Multifunctional Integration
+
+Siyuan Rao
+syrao@umass.edu
+
+University of Massachusetts, Amherst   https://orcid.org/0000-0002-1555-487X
+
+Sizhe Huang
+UMass Amherst
+
+Xinyue Liu
+Massachusetts Institute of Technology
+
+Shaoting Lin
+https://orcid.org/0000-0002-1308-9628
+
+Christopher Glynn
+UMass Amherst
+
+Kayla Felix
+UMass Amherst
+
+Atharva Sahasrabudhe
+Massachusetts Institute of Technology
+
+Collin Maley
+UMass Amherst
+
+Jingyi Xu
+UMass Amherst
+
+Weixuan Chen
+UMass Amherst
+
+Eunji Hong
+UMass Amherst
+
+Alfred Crosby
+University of Massachusetts Amherst   https://orcid.org/0000-0001-8850-8869
+
+Qianbin Wang
+UMass Amherst
+Keywords:
+
+Posted Date: May 9th, 2023
+
+DOI: https://doi.org/10.21203/rs.3.rs-2864872/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on April 25th, 2024. See the published version at https://doi.org/10.1038/s41467-024-47988-w.
+Control of Polymers’ Amorphous-crystalline Transition for Hydrogel Bioelectronics Miniaturization and Multifunctional Integration
+
+Sizhe Huang¹, Xinyue Liu², Shaoting, Lin³, Christopher Glynn¹, Kayla Felix¹, Atharva Sahasrabudhe⁴, Collin Maley¹, Jingyi Xu¹, Weixuan Chen¹, Eunji Hong¹, Alfred J. Crosby⁵, Qianbin Wang¹,*, Siyuan Rao¹,6,7,*
+
+¹ Department of Biomedical Engineering, University of Massachusetts, Amherst, MA 01003, United States
+² Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States
+³ Department of Mechanical Engineering, Michigan State University, MI, 48824, United States
+⁴ Research Laboratory of Electronics, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States
+⁵ Department of Polymer Science and Engineering, University of Massachusetts, Amherst, MA 01003, United States
+⁶ Institute for Applied Life Sciences, University of Massachusetts, Amherst, MA 01003, United States
+⁷ Neuroscience and Behavior Graduate Program, University of Massachusetts, Amherst, MA 01003, United States
+
+Corresponding authors: Qianbin Wang (qianbinwang@umass.edu), Siyuan Rao (syrao@umass.edu)
+Abstract:
+
+Bioelectronic devices made of soft elastic materials exhibit motion-adaptive properties suitable for brain-machine interfaces and for investigating complex neural circuits. While two-dimensional microfabrication strategies enable miniaturizing devices to access delicate nerve structures, creating 3D architecture for expansive implementation requires more accessible and scalable manufacturing approaches. Here we present a fabrication strategy through the control of metamorphic polymers’ amorphous-crystalline transition (COMPACT), for hydrogel bioelectronics with miniaturized fiber shape and multifunctional interrogation of neural circuits. By introducing multiple cross-linkers, acidification treatment, and oriented polymeric crystalline growth under deformation, we observed about an 80% diameter decrease in chemically cross-linked polyvinyl alcohol (PVA) hydrogel fibers, stably maintained in a fully hydrated state. We revealed that the addition of cross-linkers and acidification facilitated the oriented polymetric crystalline growth under mechanical stretching, which contributed to the desired hydrogel fiber diameter decrease. Our approach enabled the control of hydrogels’ properties, including refractive index (RI 1.37-1.40 at 480 nm), light transmission (> 96%), stretchability (95% - 111%), and elastic modulus (10-63 MPa). To exploit these properties, we fabricated step-index hydrogel optical probes with contrasting RIs and applied them in optogenetics and photometric recordings in the mouse brain region of the ventral tegmental area (VTA) with concurrent social behavioral assessment. To extend COMPACT hydrogel multifunctional scaffolds to assimilate conductive nanomaterials and integrate multiple components of optical waveguide and electrodes, we developed carbon nanotubes (CNTs)-PVA hydrogel microelectrodes for hindlimb muscle electromyographic and brain electrophysiological recordings of light-triggered neural activities in transgenic mice expressing Channelrhodopsin-2 (ChR2).
+Main
+
+Soft and elastic bioelectronics enable multifunctional interrogation of cell function from single-cell to organ-level resolution while providing tissue-like interfaces. In dynamically moving in vivo environments, such soft bio-interfaces can adapt to the persistent mechanical deformations of the living tissues, and consequently provide chronic, reliable access to biological systems. For the sophisticated yet delicate nervous system interfaces, elastic polymer materials, including polydimethylsiloxane (PDMS)\(^1\), cyclic olefin copolymer elastomer (COCE)\(^2\), polyurethane (PU) \(^3\), alginate hydrogels\(^4,\)^5, have been deployed as the suitably elastic substrate for multifunctional devices that enable neural optogenetics stimulation\(^{1,6,7}\), electrophysiological recording\(^{8,9}\), drug infusion\(^{10}\) and neurotransmitter detection\(^{11}\). However, fabricating dedicated microstructures in soft and elastic devices is limited to 2D architectures and heavily relies on successive and sophisticated manufacturing approaches such as lithography\(^{12,13}\) and micro-printing \(^{14}\).
+
+Thermal pulling yields multiple-step scaling-down feasibility for multifunctional polymer fibers\(^{10,15}\); however, this approach requires coherent parameters of the constituent materials, such as glass transition temperature (\(T_g\)), melting temperature (\(T_m\)) and thermal expansion coefficients (\(\alpha\)) to be drawn into an integrated fiber. Moreover, the high-temperature process narrows the selections of available polymers for high-water-content bioelectronics. Assisted with hydrogel cross-linking as a soft material matrix, hybrid multifunction fibers permit adaptive bending stiffness for long-term sensing and neural modulation\(^{4,16}\).
+
+Besides mechanical stiffness change in the hydrated state and the desiccated state, hydrogel materials permit tunable volumetric control as the supporting scaffold. Employing hydrogel swelling behaviors in the solvated state, the expansion microscopy technique utilized hydrogel volumetric increase to enhance microimaging resolution for intact biological tissues\(^{17}\). In contrast,
+hydrogel shrinking behaviors in a desiccated state have been applied to densify patterned materials in volumetric scaffold deposition and obtain nanoscale feature sizes in three dimensions\(^{18,19}\). However, the hydrogel swelling and shrinking behaviors in these techniques are based on reversible polymer chains collapse in the desiccated state and expansion upon hydration. When applied to an aqueous in vivo environment, the shrunk hydrogels will expand and lose the miniaturized structures from the original manufacturing.
+
+Inspired by the volumetric change resulting from polymer chains' folding and expansion, we hypothesize that control of the amorphous-crystalline transition in semi-crystalline hydrogels can enable intervention in polymer chain folding and crystallization. Consequently, this process prevents polymer chains’ expansion from their designed nanocrystalline structure in order to maintain hydrogels’ volumes under a solvated state. Hydrogel bioelectronics, miniaturized by the polymeric crystallization approaches, can stably maintain their designed architectures in vivo.
+
+Here, we developed a set of cross-linking chemistry and micro-fabrication processes to control polymeric crystalline domain growth with cross-linked polyvinyl alcohol (PVA) hydrogels. A stable and tunable volumetric decrease of hydrogels was consistently achieved in a hydrated state under physiological conditions (pH 6-8, 37 °C). Through acidification treatment that increases polymer chain mobility while introducing dual cross-linkers of the inorganic binder tetraethyl orthosilicate (TEOS) and the generic glutaraldehyde (GA), we minimized the polymetric crystalline scattering (crystal size around 3.5 nm) and increased the hydrogels’ refractive indices (RI). Further nanocrystalline orientation induced by uniaxial deformation promoted the generation of nanoscale anisotropic architectures. This control of metamorphic polymers’ amorphous-crystalline transition (COMPACT) strategy enabled a 79.7% diameter decrease of hydrogel fibers in the hydrated state while maintaining high stretchability (94.5% - 111.2%) and low elastic moduli
+(9.7-62.5 MPa). Since COMPACT hydrogels provide a variety of RI options, we developed core-cladding hydrogel fibers with distinct RI contrast (\( n_{core}=1.40,\ n_{cladding}=1.34 \)). These core-cladding structured hydrogel fibers were applied for concurrent photometry recordings from mouse brain ventral tegmental area (VTA) in the context of social interactions. Taking advantage of these tunable hydrogel matrix scaffolds, we loaded conductive nanomaterials, carbon nanotubes, into COMPACT hydrogels for hybrid microelectrodes. Integrated with an optical core, we produced multifunctional hydrogel optoelectronic devices for in vivo electrophysiological recording of optically triggered neural activities.
+
+Results
+
+COMPACT strategy for hydrogels controllable shrinking
+
+Chemically cross-linked PVA hydrogels have been widely employed with superior optical properties\(^{20}\), fatigue-resistance\(^{21,22}\), and biocompatibility for bioelectronics applications\(^{23,24}\). To further explore PVA hydrogels’ controllable miniaturization properties while preserving these advantageous features, we designed new hydrogels fabrication approaches by control of metamorphic polymers’ amorphous-crystalline transition (COMPACT) with the following aspects: (i) polymer chains folding and immobilization with multiple cross-linkers, (ii) intervention on intermolecular chain interactions in the hydrogel matrix, (iii) inducing the oriented growth of nanocrystalline domains. We implemented the COMPACT strategy following three major procedures to control individual polymer chain folding, polymer chain network interactions and nanocrystalline growth. We first introduced the hydrolysis of TEOS in PVA solutions through homogenization (**Fig. 1a** and **Supplementary Note 1**), followed by the addition of a generic cross-linker, GA. A combination of two types of cross-linkers is chosen to allow the control of polymer chain mobility via covalent bonding and parallel tuning of hydrogels’ refractive index. We then
+acidified the cross-linked hydrogels to promote intermolecular chain interactions and to facilitate the formation of nanocrystalline domains in hydrogels. External mechanical stretching was applied to the fully acidified hydrogels and maintained during the desiccating process. After the removal of water molecules from hydrogels, high-temperature (100 °C) annealing was employed to further promote the growth and orientation of the nanocrystalline domains. To test whether polymeric nanocrystalline domains created through the COMPACT strategy can preserve hydrogels volumetric shrinking under hydrated status, we next examined the dimensions and water fractions of cross-linked hydrogels under pristine, desiccated, and re-hydrated states (**Fig. 1b-e**).
+
+We prepared fiber-shaped hydrogels via molding and extrusion methods (**Supplementary Note 2**). At the pristine (**Fig. 1b**) and desiccated states (**Fig. 1c**), the two hydrogel fibers with TEOS-GA cross-linking (COMPACT+) and GA cross-linking (COMPACT-) exhibited comparable geometries and water fractions (**Fig. 1e**); however, only the TEOS-GA cross-linked PVA hydrogel fiber with acidification and mechanical stretching maintained the reduced diameters in the re-hydrated state (**Fig. 1d, e**).
+
+After we confirmed that hydrogels retained shrinking behaviors in the re-hydrated state with COMPACT treatment, we tested whether size reduction is dependent on the materials' geometries and external constraints. We prepared hydrogels with the shapes of thin film, fiber, and block, and examined the changes of COMPACT hydrogel film thickness (\( T \), **Fig. 1f**), fiber diameter (\( D \), **Fig. 1g**) and volume (\( V \), **Fig. 1h**). TEOS-GA cross-linked PVA hydrogel thin films with acidification treatment exhibited a thickness reduction ratio of \( 93.4 \pm 3.6\% \) (pristine thickness: \( 501 \pm 134 \) μm; re-hydrated thickness: \( 33 \pm 18 \) μm) under optical microscopy examination (**Fig. 1f**). TEOS-GA cross-linked PVA hydrogel fibers, with applied acidification and mechanical strain (200%) treatments, reached the maximum diameter shrinking ratio of \( 79.7 \pm 2.3\% \), by increasing the
+content of the TEOS cross-linker (Fig. 1g). In three-dimensional free shrinking structures, we observed 80.9± 0.7% volumetric shrinking in acidified TEOS-GA cross-linked cylinders as compared to pristine ones (Fig. 1h).
+
+We then investigated the mechanisms of the sustained hydrogel volume decrease and the design of amorphous and crystalline architectures. Fourier transform infrared spectroscopy (FTIR) results indicated covalent bonds (Si-O-Si and Si-OH) generated in the COMPACT hydrogel network (Fig. 1i). The new Si-O-Si (1080 cm^{-1}) and Si-OH (950 cm^{-1}) bonds came from hydrolyzed TEOS Si-OR groups’ reactions with the hydroxyl groups on PVA chains. The generic cross-linker GA reactions were confirmed by the observation of C=O bond (1740 cm^{-1}) and the Si-O-C bond (1140 cm^{-1}) from the reaction with TEOS Si-OR groups. Besides confirming covalent bonds generated among hydrogel polymer chains, differential scanning calorimetry (DSC) results exhibited the change of polymer chain interactions and polymeric crystallinity after COMPACT treatment. Undissolved PVA powders showed 28.4 ± 3.5% crystallinity (Fig. 1j and Supplementary Fig.2), similar to the reported crystallinity percentage of semi-crystalline PVA polymers^{25}. GA-cross-linked PVA hydrogels exhibited 21.6 ± 1.1% crystallinity while the additional TEOS cross-linking, and acidification suppressed the polymer chain folding to form crystalline domains (crystallinity: 12.7 ± 1.5%). We further examined the nanocrystalline domains and orientation with X-ray scattering techniques. The size of PVA nanocrystals was measured as 3.5 ± 0.1 nm while the nano-crystalline spacing increased from 8.4 nm to 10.2 nm after 200% axial stretching (Fig. 1k and Supplementary Fig. 2-4). Wide-angle X-ray scattering (WAXS) 2D patterns suggested that the lamellae crystal domains were re-oriented along the axial stretching direction (Fig. 1k and Supplementary Fig. 4).
+
+COMPACT hydrogel fibers’ tunable properties
+With COMPACT-enabled hydrated hydrogel size reduction, we expanded this methodology to develop a series of hydrogel fibers with controlled diameters and tunable optical and mechanical properties for biomedical use. We mapped a rational and comprehensive shrinking diagram by varying the content of inorganic cross-linker (TEOS), acidification, and external mechanical stretching (**Fig. 2a**). Generally, increasing cross-linking density with more cross-linkers yielded less ductile polymer chains with reduced dimension upon hydration. Acidification treatment dramatically boosted shrinking percentages across different cross-linking densities while mechanical static stretching further decreased hydrogel fibers in diameters (79.7 ± 2.3%). To fit COMPACT into a practical molding-extrusion fabrication process (**Supplementary Note 2**)26, we examined a series of hydrogel fibers made with different sizes of silicone molds (**Fig. 2b** and **Supplementary Fig. 5**). Independent from the mold size, all COMPACT hydrogel fibers reached reduced diameters more than 79%, which is consistent with the shrinking diagram (**Fig. 2a**). As an example, using 300 μm (inner diameter, ID) silicone molds, thin hydrogel fibers were fabricated with diameters of 80 ± 4 μm.
+
+Considering their fiber optic in vivo applications27, we examined the optical, mechanical and biocompatible properties of COMPACT hydrogel fibers. To ensure efficient light transmission for optical stimulation and recordings, we considered two important parameters of the hydrogel fiber core: refractive index (RI) and light transmittance. We observed that hydrogels’ refractive indices can be tuned by increasing TEOS contents. COMPACT hydrogels with 0 wt. % to 4 wt. % TEOS contents exhibited refractive indices ranging from 1.48 to 1.60 in the desiccated state (**Fig. 2c**) and 1.37 to 1.40 in the hydrated state (**Supplementary Fig. 6a-b**), which is comparable with the RI of other conventional polymer hydrogels28. Although all the transmittance remained above 96%, increasing TEOS content also led to decreased transmittance (**Fig. 2c** and **Supplementary Fig.
+6c), and increased autofluorescence (17.8% increase of 4 wt.% TEOS hydrogels compared to 0 wt. % TEOS hydrogels, excitation wavelength: 485 nm, excitation peak: 520 nm, Supplementary Fig. 6d). The optimal TEOS content was chosen as 3 wt. %, which resulted in hydrogels with 1.54 ± 0.01 of refractive index (Fig. 2c), > 96% of transmittance ((Fig. 2c, for 0.15 ± 0.02 mm thick membranes), and 6.13 ± 0.16 relative fluorescent units (RFU)/mm of autofluorescence (for 0.15 ± 0.02 mm thick membranes. water: 3.70 RFU/mm, Supplementary Fig. 6d).
+
+We then examined whether COMPACT hydrogels maintained tissue-like elasticity. COMPACT hydrogel fibers exhibited relatively low elastic moduli while maintaining high stretchability (Fig.2d and Supplementray Fig. 7a-b). The optimized COMPACT hydrogel fiber (3 wt.% TEOS, 12 mM HCl acidification treatment and 200% stretching, diameter: 227 ± 18 \( \mu \)m) exhibited an elastic modulus of 34.03 ± 7.38 MPa. Compared to silica fibers (~20GPa elastic modulus)\(^{29}\) and polymer fibers (~1GPa elastic modulus)\(^{2,4}\), COMPACT hydrogel fibers offer enhanced mechanical matching to the nervous tissues (1-4 kPa)\(^{30}\) and lead to less neural tissue damage from micro-motion involved in vivo studies\(^{31}\).
+
+We then tested whether crystalline-enabled size reduction of COMPACT hydrogels can overcome the intrinsic hydrogel swelling exhibited upon hydration and maintain structural stability in vivo, we incubated COMPACT hydrogel fibers in ex vivo physiological conditions (pH: 6-8, 37 °C, saline solution) and monitored fibers’ dimension over time. We observed the shrinking percentage maintained above 74% over 3 months (Fig. 2e and Supplementary Fig. 8). Cytotoxicity tests with human embryonic kidney cells (HEK293) exhibited no significant cell death in the presence of COMPACT hydrogels (Fig. 2f and Supplementary Fig. 9).
+
+Step-index hydrogel optical fibers
+COMPACT hydrogels were first fabricated into step-index optical fibers (Supplementary Note 3). Increased RI contrast between optical core and cladding layers ensures light transmission and the consequent photodetection sensitivity (Fig. 3a). Based on tunable refractive indices of COMPACT hydrogels (Fig. 2c and Supplementary Fig. 6 a-b), we designed step-index hydrogel fibers with high-RI core (\( n_{core}=1.40 \)) and low-RI cladding (\( n_{cladding}=1.34 \)).
+
+Hydrogel fibers were connected to a silica segment embedded in an optical ferrule, which provides a strong connection while preventing directly exposed hydrogel dehydration out of tissues and light loss (Supplementary Note 3). We validated the function of RI-contrasting core-cladding structures by comparing the light transmission between bare core fibers, step-index fibers with plain cladding and those with light-protective cladding (Fig. 3b-c, and Supplementary Note 3-4).
+
+The bare core fibers (diameter of \( 329 \pm 17 \) μm) exhibited a relatively high attenuation (\( 1.87 \pm 0.53 \) dB/cm) while introducing a thin low-RI cladding layer (thickness of \( 84 \pm 4 \) μm on the surface of \( 372 \pm 10 \) μm cores, \( n_{cladding}=1.34 \)) decreased the light transmission attenuation to \( 1.75 \pm 0.08 \) dB/cm (Fig. 3c). A representative light-absorption nanomaterial\(^{32,33}\), reduced graphene oxide (rGO) was loaded into low-RI cladding to further protect light leakage from fibers’ lateral surface and consequently reduced the light attenuation to \( 0.94 \pm 0.25 \) dB/cm (core \( 339 \pm 35 \) μm, cladding: 36 ± 11 μm of 5 wt.% PVA with 0.21 wt.% rGO) (Fig. 3c and Supplementary Fig. 10).
+
+To validate their functionality for in vivo optical interrogation, we tested COMPACT hydrogel fibers with fiber photometric recording in the context of mouse social behaviors. Activation of VTA region and its related circuits has been studied with various techniques, including optogenetics\(^{34}\), electrical stimulation and chemogenetics\(^{35}\), related to social behaviors in mice\(^{36}\). As a proof-of-concept application, we applied COMPACT hydrogel fibers to record mouse deep brain structure, VTA, with concurrent social behavior observation. We unilaterally implanted
+COMPACT optical fibers (580 ± 35 μm) in VTA after injecting of adeno-associated virus (AAV) containing genetically encoded calcium indicator (*hSyn*::GCaMP6s) (**Fig. 3e**). A home-built fiber photometry system (wavelengths: \( \lambda_{isosbestic\ point}=405\ \text{nm} \), \( \lambda_{excitation}=470\ \text{nm} \), \( \lambda_{emission}=510\ \text{nm} \)) based on the previous design was used to collect GCaMP fluorescent change as a proxy to reflect the neural activity\(^{37}\) (**Fig. 3g** and **Supplementary Fig. 13**). We utilized the stiffness change of hydrogel fibers from a desiccated state (stiff) to a hydrated state (soft) and implanted the hydrogel fiber in the desiccated state with calibrated coordinates (**Supplementary Figure. 11-12**). After an incubation period of 4 weeks for AAV virus expression, we subjected mice to a social behavioral test with concurrent photometric recordings. Mouse social interactions were analyzed with DeepLabCut markless pose estimation and a custom-developed MatLab algorithm (**Fig. 3f**). We found that increased fluorescent intensity of GCaMP was correlated with mouse social interaction epochs (**Fig. 3h**).
+
+**COMPACT multifunctional hydrogel neural probes**
+
+Hydrogel matrix can support various nanoscale materials to extend the functionalities while maintaining desired mechanical properties\(^{35,38}\). To enrich hydrogel neural probes’ modality for electrical recordings, we incorporated conductive carbon nanotubes (CNTs, 12 ± 6 nm diameter) into PVA hydrogel scaffolds during hydrogel cross-linking (**Fig. 4a** and **Supplementary Note 5**). Acidification and mechanical stretching facilitated CNT plaiting into polymer matrices and ensured entanglement with PVA chains and consequently augmented electrical conductivity as a percolated network\(^{39,40}\). CNTs-PVA hydrogel electrodes (86 ± 5 μm diameter) exhibited stable impedances of 658 ± 277 kΩ at 1kHz (PBS, 25 °C, **Fig. 4c** and **Supplementary Fig. 15**) and impedance was tunable with designed mold sizes and CNT loadings (**Fig. 4d** and f). CNTs-PVA hydrogel electrodes were insulated with a viscoelastic coating of styrene-ethylene-butylene-
+styrene (SEBS) (Supplementary Note 5 and Supplementary Fig. 16). To verify the stability of CNTs-PVA hydrogel electrodes, we incubated them in PBS solutions and characterized the impedance over 6 weeks (**Fig.4e**). No significant increase of impedance at 1kHz was found.
+
+Then we deployed CNT-PVA hydrogel electrodes for electromyographic (EMG) recordings of mouse hindlimb muscles in response to the pulsed blue light illumination. CNT-PVA hydrogel electrodes detected hindlimb muscle electrical signals upon transdermal optical stimulation (wavelength \( \lambda = 473 \) nm, 200 mW/mm\(^2\), 0.5 Hz, pulse width 50ms) in *Thy1::ChR2-EYFP* mice, which express photo-excitatory opsin, Channelrhodopsin 2 (ChR2), in the nervous system (**Fig. 4f**). EMG signals exhibited repeatable amplitude and signal-to-noise ratios, which indicates the reliability of CNT-PVA hydrogel microelectrodes.
+
+When extending hydrogel miniaturization from bulk materials to interfaces, the COMPACT strategy offers a new avenue for multiple components integration. Since RI-distinct core-cladding structures ensure light transmission in optical cores, we introduced two CNT-PVA electrodes into the cladding layers with a COMPACT hydrogel core (**Fig. 4b**). A hydrogel optoelectrical device (optrode), is designed to enable optical modulation with simultaneous electrophysiological recording (**Supplementary Note 6**). In *Thy1::ChR2-EYFP* mice, blue light pulses (\( \lambda = 473 \) nm, 0.5 Hz, pulse width 50 ms, 10 mW/mm\(^2\)), delivered through the hydrogel optical core, consistently activated ChR2-expressing neurons in VTA while the neural electrical signals were collected through CNT-PVA electrodes (**Fig. 4l**). The optical evoked potentials were repeatedly captured with correlation with the onset of light stimulation over two weeks post-implantation.
+
+**Discussion**
+In this study, we developed a set of hydrogel cross-linking chemistry and fiber-shaped device microfabrication approaches through a bottom-up strategy of tuning polymers’ amorphous-crystalline transition for hydrogel bioelectronics miniaturization and integration. COMPACT provides an accessible, scalable, and controllable fabrication method for micro-structured hydrogel fibers as small as 80 \( \mu \)m with consistently low asperity. These hydrogels provide a platform for functionally augmented interfaces through loadings of additional nanomaterials. COMPACT hydrogels can be further designed into step-index optical probes and optoelectronic devices (optrodes) which are well-suited for neural modulation and recordings concurrent with behavioral assays in mice.
+
+Unlike established approaches to shrink hydrogels via desiccation, where collapse of polymer chain during drying leads to reversible swelling upon hydration, COMPACT hydrogels’ polymeric nanocrystalline and enhanced interpolymer chain interactions maintained stable folding in the hydrated state and therefore permit retained volumetric size reduction. Over 3 months of incubations under physiological temperature and osmolarity, the shrunk COMPACT hydrogel fibers maintained the designed diameters within less than 1% variance (**Fig. 2e**), which illustrates COMPACT bioelectronics’ volumetric stability of their miniaturized size in vivo. In contrast, COMPACT hydrogel fibers incubated at PVA dissolution temperature (100 \( ^\circ \)C) in water for several hours resumed their pristine swollen size; this volume reversion demonstrates the crystalline impact on size reduction through control of local free volume in hydrogel matrices. This crystalline-dominated hydrogel miniaturization phenomenon can be extended to other semi-crystalline polymers at different material interfaces, where volumetric stability is important, such as the proton-exchange membrane in packed fuel cells.
+In COMPACT hydrogels, chemical cross-linkers and acidification treatment both contribute to the retained volumetric decrease upon re-hydration while mechanical deformation induced the orientated nanocrystalline growth. An increased number of chemical cross-linkers, TEOS (0 wt.% to 4 wt.%, Fig. 1g), enhanced the anchoring of amorphous PVA chains through covalent cross-linking and prevent swelling in the hydrated state. Under the same cross-linking degree, acidification treatment granted polymer chains enhanced interactions and suppressed crystallinity (Fig. 1j and Supplementary Fig. 1 and 7c). Nanocrystalline domains maintained the nanoscale size (~3.5 nm) without compromising the transmittance in the visible range. Axial mechanical deformation re-orientated nanocrystalline and created anisotropic nanostructures (Fig. 1k), which enabled hydrogel fibers’ desired decrease in diameter while causing a minimal effect on crystallinity degree (Supplementary Fig. 1c) or nanocrystalline size (Fig. 1k).
+
+Controllable hydrogel shrinking provides an effective methodology for miniaturization and integration for neural probe fabrication. The molding and extrusion approaches offer a series of precisely controlled hydrogel fiber diameters with structural homogeneity and low surface asperity to avoid diffuse reflection at the hydrogel interfaces. COMPACT hydrogel fibers’ tissue-like mechanical properties exhibit improved immune response compared to stiff silica fibers (Fig. 2d and Supplementary Fig. 14). Although the mold sizes are commercially limited, COMPACT procedures, including regulating polymer and crosslinker constituent content and fiber extensions can expand the range of available fiber sizes. Successive rounds of molding with strong polymer chain infiltration at the interfaces enable the design of multimodal microstructures, including core-cladding (30-80 μm) in step-index optical probes and electrode integration in the cladding layer of optrodes. Currently, the number of integrated components, such as electrodes and microfluidic channels, is limited by the coaxial alignment in the secondary molding step; the accessibility and
+throughput of multimodal fabrication can be further improved with guiding devices to facilitate integration and alignment, or alternative coating approaches.
+
+COMPACT strategy is generalizable for soft and stretchable bioelectronics. Polymer matrices provide sufficient free volume for water access as well as nanomaterials’ incorporation. High aspect-ratio nanomaterials, such as silver nanowires and carbon nanotubes, can be effectively entangled with polymer chains through cross-linking and condensation during acidification and stretching. This procedure augments electrical conductivity while maintaining viscoelasticity. The colloidal stability of nanomaterials in viscous polymer precursor solutions is important to create a homogeneous composite after cross-linking to prevent phase separation and ensure stable electrical conductivity.
+
+Compared to other soft bioelectronics fabrication approaches, such as lithography and micro-printing, COMPACT technique offers scalable and efficient multimodal hydrogel fibers manufacturing without the need for expensive and sophisticated facilities. COMPACT multifunctional neural probes have been employed for bi-directional optical interrogation concomitant with mouse social behaviors and electrical recordings of light-triggered neural activity in mice. Extended functionalities, such as drug or viral vector delivery, can be further achieved by integrating additional microfluidic channels in the cladding layer and retains light transmission efficiency in the optical core. COMPACT multifunctional neural probes involve independent components alignment and miniaturization steps, which potentiates the integration of multiple components with various lengths to target multiple depths of tissue within single-step implantation. This adaptability will increase the density of functional interfaces and overcome the traditional limitation of fiber-shaped neural probes with single-target interfaces at the tip.
+Control over semi-crystalline polymers’ amorphous-crystalline transition creates a direct fabrication methodology for elastic soft materials. Extending it to the manufacture of sophisticated optoelectronic devices, the COMPACT strategy imparts a generalizable and modular platform for hydrogel bioelectronics’ miniaturization and integration, which consequently enables multimodal interrogation of complex biological systems.
+
+Methods
+
+Hydrogel synthesis. The chemicals used in this study included tetraethyl orthosilicate (TEOS, Sigma-Aldrich 86578, 99%), hydrochloric acid (HCl, Sigma-Aldrich, 258148, 37%), glutaraldehyde solution (GA, Sigma-Aldrich G6257, 25% in water), and polyvinyl alcohol (PVA) with an average molecular weight of 146,000 to 186,000 Da and 99+% hydrolyzed (Sigma-Aldrich, 363065). MilliQ water with a resistivity of 18 MΩ·cm at 25 °C was used throughout the experiments. To prepare the PVA (10 wt. %) solution, PVA was dissolved in MilliQ water and stirred in a water bath at 100 °C for at least 4 hours until a clear and transparent solution was obtained. The hydrolysis of TEOS was carried out using HCl as a catalyst in PVA solutions with a molar ratio of TEOS: HCl: H2O=x: 4: y, where x was between 1 to 4, and y started from 4 to 16. TEOS solutions with concentrations ranging from 2 wt.% to 8 wt.% were added to the PVA solutions, which were then homogenized at two different levels. A mixture of HCl and MilliQ water in a molar ratio of 4: y, where y was in the range of 4 to 16, was added dropwise to the PVA-TEOS emulsion while homogenizing at 12000 rpm using a portable homogenizer until a stable emulsion was formed. The resulting emulsion was further homogenized using a high-speed homogenizer (FSH2A lab). The mixed solutions were stirred in a water bath at 100 °C for 1 hour
+until transparent solutions were obtained, followed by an additional 12 hours of stirring at 60 °C.
+
+The composition of all solutions used in this study is provided in Table 1.
+
+<table>
+  <tr>
+    <th colspan="5">Table 1. TEOS and PVA concentrations of PVA-TEOS solutions</th>
+  </tr>
+  <tr>
+    <th>TEOS: HCl: H<sub>2</sub>O (molar ratio)</th>
+    <th>TEOS wt.% in PVA pre-solutions</th>
+    <th>HCl wt.% in solutions</th>
+    <th>PVA wt.% in solutions</th>
+  </tr>
+  <tr>
+    <td>1: 4: 4</td>
+    <td>2</td>
+    <td>0.014</td>
+    <td>10</td>
+  </tr>
+  <tr>
+    <td>2: 4: 8</td>
+    <td>4</td>
+    <td>0.014</td>
+    <td>10</td>
+  </tr>
+  <tr>
+    <td>3: 4: 12</td>
+    <td>6</td>
+    <td>0.014</td>
+    <td>10</td>
+  </tr>
+  <tr>
+    <td>4: 4: 16</td>
+    <td>8</td>
+    <td>0.014</td>
+    <td>10</td>
+  </tr>
+</table>
+
+Optical hydrogel probe fabrication. A step-index multimode silica fiber (core diameter 400 μm, NA 0.5, Thorlabs FP400URT) was prepared by removing the protective coating using a fiber stripping tool (Micro-strip, Micro Electronics, Inc). The stripped fiber was then divided into 13-mm segments using a diamond cutter. These fiber segments were inserted and extruded from one end of an optical ferrule (bore diameter 400 μm, Thorlabs CFX440-10) with a length of 2.5 mm and secured with EccoBond F adhesive (Loctite). Both ends of the silica fibers in the ferrules were polished using a polish kit (Thorlabs D50-F, NRS913A, and CTG913). The light transmission of all silica fibers and ferrules was tested by coupling with a 470 nm blue light-emitting diode (LED) (Thorlabs M470F3) after polishing. To remove the plastic coatings on the extruded silica fibers, they were treated with 2M sodium hydroxide solution (Sigma-Aldrich, 1064980500) for 2 hours followed by an additional treatment with chloroform (Sigma-Aldrich, 472476) for 30 minutes. A thin layer of 10 wt.% PVA was then coated on the extruded silica fibers via dip coating, and the PVA-coated silica fibers were air-dried at room temperature for 12 hours and annealed at 100 °C for 2 hours. A vacuum planetary mixer (Musashi ARV-310, 2000 rpm, and 16 kPa vacuum) was utilized for the mixing and degassing of all solutions. For degassing and mixing, 100 μL of GA was added to 10 g of 10 wt.% PVA pre-solution and agitated for 1 minute. 10 g of pre-made PVA-
+TEOS solution was also degassed and mixed for 1 minute. Subsequently, the above two solutions were combined (weight ratio of 1:1) and mixed for another minute. The resulting PVA-TEOS-GA solution was infused into silicone tubes (McMaster-Carr 5236k204, 80 mm in length), and the optic ferrules were inserted into the silicone tubes, with the silica fiber end connected to the PVA mixture. After curing at room temperature for 4 hours, the PVA-TEOS-GA fibers were demolded using dichloromethane (DCM, Sigma-Aldrich, 270997, 99.8%) and washed with a large amount of water to remove residual chemicals for two days. Ferrule-connected fibers were air-dried at room temperature for 12 hours and annealed at 100 °C for 20 minutes. Finally, the hydrogel fibers were rehydrated with MilliQ water before use. The compositions of all fabricated fibers are listed in Table 2.
+
+<table>
+  <tr>
+    <th rowspan="2">Nomenclatura</th>
+    <th colspan="2">TEOS: HCl: H<sub>2</sub>O<br>(molar ratio)</th>
+    <th colspan="2">TEOS wt.% in fibers</th>
+    <th colspan="2">GA wt.% in fibers</th>
+    <th colspan="2">PVA wt.% in fibers</th>
+  </tr>
+  <tr>
+    <th>TEOS</th>
+    <th>HCl</th>
+    <th>TEOS</th>
+    <th>HCl</th>
+    <th>GA</th>
+    <th>PVA</th>
+  </tr>
+  <tr>
+    <td>10P-1T-GA</td>
+    <td>1</td>
+    <td>4</td>
+    <td>1</td>
+    <td>0.007</td>
+    <td>0.005</td>
+    <td>10</td>
+  </tr>
+  <tr>
+    <td>10P-2T-GA</td>
+    <td>2</td>
+    <td>4</td>
+    <td>8</td>
+    <td>0.007</td>
+    <td>0.005</td>
+    <td>10</td>
+  </tr>
+  <tr>
+    <td>10P-3T-GA</td>
+    <td>3</td>
+    <td>4</td>
+    <td>12</td>
+    <td>0.007</td>
+    <td>0.005</td>
+    <td>10</td>
+  </tr>
+  <tr>
+    <td>10P-1T-GA</td>
+    <td>4</td>
+    <td>4</td>
+    <td>16</td>
+    <td>0.007</td>
+    <td>0.005</td>
+    <td>10</td>
+  </tr>
+</table>
+
+Core-cladding optical probe fabrication. A vacuum planetary mixer (Musashi ARV-310, 2000 rpm, and 16 kPa vacuum) was employed for mixing and degassing of all solutions. The optical fiber probes were first dried and then re-inserted into silicone tubing (McMaster-Carr 51845K66) and reswelled in water. For the preparation of the core-cladding optical fiber probes, 100 μL of GA was added to 10g of 5 wt.% PVA pre-solution, which was then degassed and mixed for 1 minute. Additionally, 150 μL of HCl was added to 10g of 5 wt.% PVA pre-solution, which was also degassed and mixed for 1 minute. The two solutions were combined (weight ratio of 1:1) and mixed for 1 minute. The resulting mixed solution was infused into the silicone tubing and allowed
+to cross-link for 4 hours at room temperature. The core-cladding optical fiber probes were extruded by immersing them in DCM and stored in MilliQ water until further use.
+
+XRD characterization of hydrogel materials. X-ray scattering measurements were conducted using the SAXSLAB GANESHA 300XL instrument, equipped with a Dectris Pilatus 300K 2D CMOS photon counting detector (size: 83.8 x 106.5 mm^2). A small-angle 2 mm beamstop was utilized for SAXS measurements, while a wide-angle 2 mm beamstop was employed for WAXS measurements. The exposure time was set at 600 seconds. The average size of the nanocrystalline domain was determined using Scherrer’s equation, which is expressed as \( D = \frac{k\lambda}{\beta \cos \theta} \), where \( k \) is a dimensionless shape factor that varies based on the actual shape of the nanocrystalline domain (\( k = 1 \), approximating the spherical shape of the nanocrystalline domains), \( \lambda \) is the wavelength of X-ray diffraction (\( \lambda = 1.54 \) Å), \( \theta \) is the peak of the Bragg angle, and \( \beta \) is the full width at half maximum (FWHM) of the WAXS peaks. The d-spacing between nanocrystalline domains was calculated using \( d = \frac{2\pi}{q_{max}} \), where \( q_{max} \) is the q value at its maximum intensity from SAXS patterns. The FWHM (\( \beta \)) and \( q_{max} \) were obtained by curve fitting of the WAXS and SAXS patterns, respectively, in Origin software (OriginLab Corporation).
+
+DSC characterization of hydrogel materials. The degree of crystallinity of hydrogel fibers and materials was assessed using a DSC instrument (2920 TA instrument). The PVA hydrogels were analyzed in the desiccated state. A small quantity of sample (1-15 mg) was loaded into a crucible (TA instrument T81006) and inserted into a temperature-controlled DSC cell. A blank crucible served as a reference. The sample was heated from 30 °C to 300 °C in air, with a heating rate of 20 °C/min. The differential heat flow to the sample and reference was recorded by the instrument.
+To determine the melting fusion enthalpy of endothermic peaks, heat flow (mW) over sample weight (mg) was plotted against time (s). The areas of melting endothermic peaks were integrated using TA analyze software (TA Universal Analysis). The degree of crystallinity \( \alpha \) was estimated using the equation: \( \alpha = \frac{\Delta H_m}{\Delta H_{m}} \cdot 100\% \), where \( \Delta H_m \) (J/g) was calculated from the integration of melting endothermic peaks and \( \Delta H_m \) (150 J/g) was the enthalpy of melting 100% of PVA crystallites. The crystallinity outcomes of PVA samples are presented in Supplementary Table 1.
+
+Hydrogel Refractive Index Measurement. A series of hydrogel membranes were prepared via spin coating using a spin coating instrument (SETCAS, KW-4A) on silicon (Si) substrates (University Wafer, Inc., Model 447). The Si substrates were cut into square wafers (13.5 mm x 17.5 mm) using a diamond cutter and then subjected to a rigorous cleaning process. The cleaning process involved washing and ultrasonication in Acetone (Sigma-Aldrich 179124, 99.5%) for 3 minutes, followed by rinsing with MilliQ water. The Si wafers were then washed and ultrasonicated in 30 wt.% H$_2$SO$_4$ solution (Fisher Chemical 210524, 95.0%) for 3 minutes, followed by rinsing with MilliQ water. Finally, the Si wafers were washed and ultrasonicated in 10 wt.% of H$_2$O$_2$ solution (Sigma-Aldrich 216763, 30 wt.% in water) for 3 minutes, followed by rinsing with 95% ethanol (Fisher Chemical A962P4, 95.0%). The Si wafers were mounted on the spin coater and coated with 10P-GA, 10P-1T-GA, 10P-2T-GA, 10P-3T-GA, and 10P-4T-GA membranes (n=4 for each group) at 1000 rpm for 10s, and at 5000rpm 50s. PVA solutions used for the membranes were prepared using the same method as discussed previously. After spin-coating, the PVA membrane-coated Si wafers were allowed to cross-link and dry in the air for at least 12 hours and then annealed at 100 °C for 20 minutes. The refractive index (RI) of the PVA membrane-coated Si wafers was measured using an ellipsometer (J.A. Woollam RC2) in the range
+of 400 nm to 700 nm. The measurements were carried out on the membranes in their desiccated states. A series of COMPACT hydrogel membranes (0-4 wt.% TEOS) were prepared using a similar procedure as described above but using a rectangular mold (21.5 × 21.5 × 1 mm). The membranes were demolded after cross-linking, dried at room temperature for 12 hours, and cut into small sheets (2 × 2 mm). The sheets were then annealed at 100 °C for 20 minutes and reswelled in MilliQ water for 1 hour. The RI of the membranes in their hydrated states was measured using a refractometer (Sper Scientific 300034) with water used for calibration.
+
+Hydrogel Absorbance and Fluorescence Measurement. A set of hydrogel membranes (designated as 10P-GA, 10P-1T-GA, 10P-2T-GA, 10P-3T-GA, and 10P-4T-GA, comprising 4 replicates for each group) were synthesized and cross-linked in a 96-well plate using established techniques. Subsequently, 1 mL of PVA solution was added to each well and allowed to cross-link and air dry for at least 12 hours, followed by annealing at 100 °C for 20 minutes. Rehydration of the membranes was achieved by the addition of 100 μL of MilliQ water to each well. To obtain transmittance spectra in the range of 400 nm to 700 nm, the 96-well plate was subjected to analysis using a plate reader (Biotek Synergy 2). Autofluorescence measurements were acquired using excitation/emission wavelengths of 470 nm/510 nm and 485 nm/520 nm, respectively. Membrane thickness was determined by caliper measurements and recorded three times to normalize the transmittance spectra and autofluorescence readings with respect to thickness. A blank control consisting of 200 μL of MilliQ water was included for comparison purposes.
+
+Mechanical characterization of hydrogel fibers. To ensure consistency, all hydrogel fibers were hydrated prior to the extension test. Tensile tests were conducted using a tensile instrument
+equipped with a 50N load cell (Stable Micro System TA, XT plusC). The fibers were stretched at a constant rate of 1 mm/second. The nominal stress was calculated from the formula \( \sigma = \frac{F}{A} \), where \( F \) represents the force measured by the instrument, and \( A \) represents the cross-sectional area of the fibers in their hydrated state. The strain was calculated using \( \varepsilon = \frac{\Delta L}{L} \), where \( \Delta L \) represents the displacement and \( L \) represents the initial gauge length. Two marks were labeled on the fibers using a sharpie pen to determine the initial gauge length \( L \) prior to the tensile test. A high-resolution camera was used to capture the entire tensile process and track displacement. The stress-strain curve was generated based on the calculated nominal stress and strain. The elastic moduli (E) were determined by calculating the average slope of the stress-strain relationship in the first 10% of applied strain. The average slope was determined by linear regression analysis (OriginLab Corporation). The stretchability of the fibers was reported as a percentage of the strain at the fracture point obtained from the stress-strain curves.
+
+Light attenuation of hydrogel fibers. The light transmission loss of hydrogel fibers was tested by the cutback method. Ferrule-connected hydrogel fibers were inserted into a plastic tube (5 cm in length and 3 mm in diameter) and injected with 1 wt.% agar gel to maintain their hydrated state. The ferrule was connected to a 470 nm LED light (Thorlabs M470F3) via an adaptor (Thorlabs SM1FCM). The power (in dB) of transmitted light through the hydrogel fiber was measured using a power meter (Thorlabs, PM16-122). The original power reading was recorded, and a 5 mm interval of cutting was adapted. Starting from the far end of the ferrule, the output power was measured after each cut using a cutter. The attenuation coefficient (\( \alpha \)) was calculated using the formula \( \alpha = (\frac{10^4}{L_1 - L_2}) \cdot log \left( \frac{P_1}{P_2} \right) \), where \( L_1 \) and \( L_2 \) represent the original and cut lengths of the fiber
+in meters, respectively. \( P_1 \) and \( P_2 \) are the transmitted power readings before and after the cut, respectively.
+
+Dimension measurements of hydrogel fibers. Microscopic images of hydrogel fibers were captured using a bright field mode microscope (AmScope) in MilliQ water. Three distinct regions of each fiber, namely two ends and the middle part, were imaged. The diameter of each fiber was measured using ImageJ software, with nine measurements taken for each fiber. The length of the fibers was measured using a caliper, with three measurements taken for each fiber.
+
+SEM imaging. SEM was performed on dried samples using an FEI Magellan 400 XHR instrument. To analyze the cross-sectional morphology of the integrated hydrogel optrode probe, the probe was sectioned into thin pillars (0.1 mm in height) and subsequently mounted on carbon tape for imaging.
+
+TEM imaging. The TME images were acquired under a transmission electron microscope (FEI Tecnai 12). The carbon nanotubes were diluted (1:10) in MilliQ water and deposited on a copper grid (Sigma-Aldrich, FCF200-Cu) for imaging.
+
+Stability tests of hydrogel fibers. The fabricated COMPACT hydrogel fibers (3 wt.% TEOS) were incubated at 37 °C under physiological-like solutions (saline, ionic strength 305~310 mOsm, pH from 6.0 to 8.0) over 3 months to validate the stability of hydrogel materials. The dimensions of fiber were measured before and after the incubation and statistical analysis was performed on the dimensions between pre-incubation and post-incubation each week.
+Cell culture and biocompatibility tests. The HEK 293FT cell line was maintained in DMEM (with GlutaMax, Sigma Aldrich, D5796) + 10% fetal bovine serum and seeded in a 24-well plate. COMPACT hydrogel fibers (3 wt.% TEOS) were incubated in DMEM for 24 hours at 37 °C. Hydrogel-incubated DMEM was then added to the well plate and incubated for 24 hours. Calcein-AM (green, 2 μL of 1 mg/mL per well, Sigma-Aldrich 17783) was added to indicate living cells, and ethidium homodimer-1 (red, 2 μL of 1 mg/mL per well, Sigma-Aldrich 46043) was added to indicate dead cells. A fluorescent microscope (Nikon TiU with SOLA Light Engine Gen III illumination hardware and PCO panda sCMOS camera) was used to take images of cells with and without hydrogel incubation. Image J was utilized to count living cells and dead cells. Cell death rate (%) was calculated by using the formula: \( death\ rate\ (\%) = \frac{dead\ cell\ numbers}{total\ cell\ numbers} \cdot 100\% \).
+
+Electrochemical impedance spectroscopy (EIS) of COMPACT hydrogel electrodes. The impedance of COMPACT hydrogel electrodes was assessed using an Electrochemical working station (Princeton Applied Research, PARSTAT 2273) by applying a sinusoidal driving voltage (10 mV, 10 Hz ~1 MHz). Impedance spectra of COMPACT hydrogel electrodes were acquired in PBS solutions.
+
+Virus package. pAAV-hSyn-GCaMP6s-WPRE-SV40 was a gift from The Genetically Encoded Neuronal Indicator and Effector Project (GENIE) and D. Kim (Addgene viral preparation no. 100843-AAV9). AAV9- hSyn-GCaMP6s were prepared in Rao Lab at UMass Amherst with Beckman Coulter Ultracentrifuge Optima XL70 with VTi 50.1 rotor. Before use, the viral vector was diluted to a titer of \( 10^{12} \) transducing units per milliliter.
+Animals. All animal surgeries were reviewed and approved by the Committee on Animal Care at the University of Massachusetts Amherst. Wild-type (C57BL/6J) mice and Thy1::ChR2-EYFP mice were purchased from the Jackson Laboratory. Mice were given ad libitum access to food and water and were housed at 24 °C ± 1 °C, with 50% relative humidity, and on a 12-h light/12-h dark cycle. All experiments were conducted during the light cycle.
+
+In vivo hydrogel optical fiber implantation into the mouse brain. C57BL/6J mice were anesthetized using 1.0% isoflurane administered in a chamber and subsequently secured onto a stereotactic frame (RWD Life Science) with a heating pad to maintain their body temperature. All surgical procedures were conducted in sterile conditions with 1% isoflurane used to maintain anesthesia. The Allen Brain Atlas was used to align the skull and determine the coordinates for viral injection and fiber implantation, specifically targeting the ventral tegmental area (VTA) at coordinates AP: -2.95 mm, ML: ± 0.50 mm, DV: -4.80 mm. An opening was made in the skull using a micro drill (RWD Life Science) at the designated coordinates. A total of 600 nL of adeno-associated virus (AAV) carrying hSyn::GCaMP6s was injected into the target region via a micro syringe and pump (World Precision Instruments, Micro 4). The viral injection device was held in place in the VTA region for 15 minutes to facilitate virus diffusion. Following fiber probe insertion, the probes were lifted by 0.1 mm to accommodate for the viral volume. Finally, the fiber probes were secured to the skull using an adhesive (Parkell, C&B METABOND) and reinforced using dental cement (Jet Set-4). The mice were monitored on the heating pad following removal of isoflurane until they were fully awake.
+In vivo optrode device implantation into the mouse brain. *Thy1::ChR2-EYFP* mice were anesthetized with 1.0% isoflurane and placed on a stereotactic frame (RWD Life Science) equipped with a heating pad to maintain body temperature. Surgery was conducted under sterile conditions, and 1% isoflurane was continuously administered to maintain anesthesia. Allen Brain Atlas was utilized to align the skull and establish optrode device coordinates (VTA, AP: -3.00 mm, ML: + (or -) 0.45 mm, DV: -4.80 mm) based on the mouse brain atlas. Prior to optrode implantation, a ground screw was implanted (AP: -3.50 mm, ML: - (or +) 1.50 mm, DV: -0.20 mm) and cerebrospinal fluid was contacted with the screw. The optrode devices were fixed on the skull with adhesive (Parkell, C&B METABOND) and reinforced with dental cement (Jet Set-4). Following the removal of isoflurane, the mice were monitored on the heating pad until fully awakened.
+
+Fiber photometry recording. Following a four-week recovery period, hSyn::GCaMP6s injected mice were tethered to a fiber photometry (FIP) system using a silica fiber (with a core diameter of 400 μm and a numerical aperture of 0.5, Thorlabs FP400URT). The silica fiber was connected to the FIP system using an adaptor (Thorlabs SM1SMA), and a ferrule (Thorlabs CF440) was fixed to the other end of the fiber. The ferrule was coupled to the implanted fiber probe using a connecting sleeve (Thorlabs ADAF1). The mice were placed in a custom-made chamber (20 × 20 × 20 cm) for social preference tests, and fluorescent signals were computed using custom-written Python code. To excite the fluorescent signal, a custom setup consisting of a 470 nm LED (Thorlabs M470F3), a 405 nm LED (Thorlabs M405F3), and dichroic mirrors (Thorlabs DMLP425R) were used. Illumination periods were determined by detecting synchronization ON/OFF pulses for each LED, with each illumination containing pulses at 10 Hz. To eliminate moving artifacts, the fitted 470 nm signals were subtracted from the fitted 405 nm signals.
+Social behavioral assay. For all behavioral experiments, adult C57BL/6 mice implanted with optical fiber probes were utilized during the dark phase of the light/dark cycle and were given at least 30 minutes of acclimatization in the behavior chamber before testing. Adult male C57BL/6 mice aged 5-6 weeks were used as strangers, and tests were performed in a dark environment. A chamber box (20 × 20 × 20 cm) containing a social cage was utilized for social interactions. Subsequently, a novel mouse was introduced to the social zone, and the test mouse was exposed to the novel mouse and allowed to interact freely. Concurrently, GCaMP fluorescence changes were recorded during social tests. A dark-vision camera was installed above the social chamber to record video footage during the social tests. The time spent interacting and the distance of social interaction were analyzed using customized algorithms for social interaction assessment with DeepLabCut. The analyzed social interaction epochs were then correlated with GCaMP signals.
+
+Immunohistology. The mice were euthanized using fatal plus (Vortech Pharmaceuticals, LTD) and transcardiac perfusion was carried out using 20 mL of PBS (Sigma-Aldrich P3813) solution followed by 20 mL of 4% paraformaldehyde (PFA, Sigma-Aldrich 8187151000) solution. The brains were then dissected from the bodies and fixed in 4% PFA solution at 4 °C overnight. After fixation, the brain tissues were treated with 30% sucrose in PBS for 2 days and subsequently frozen at -20 °C in an O.C.T. cube (21.5 × 21.5 × 22 mm) and sectioned on a cryostat (Leica CM1900) with a thickness of 20 μm. The sectioned tissues were then permeabilized in PBST (0.3% Triton-X-100 in PBS, Sigma-Aldrich 93443) for 15 minutes at room temperature and blocked with 1% bovine serum albumin in PBS (Sigma-Aldrich A9647) for 30 minutes prior to staining. Primary antibody solutions (Iba1 Rabbit and GFAP Rabbit, Agilent Dako, Z0334, at a dilution of 1:400 in
+PBS) were applied to stain the tissues and incubated overnight at room temperature. After washing the tissues with PBS three times, secondary antibody solutions (GFAP: Thermo Fisher Scientific, Donkey anti-Rabbit IgG (H+L) Highly Cross-Absorbed Secondary Antibody Alexa Fluor 488 Invitrogen, #A-21206; Iba1: Thermo Fisher Scientific, Donkey anti-Rabbit IgG (H+L) Highly Cross-Absorbed Secondary Antibody Alexa Fluor 555, # A-31572; dilution: 1:200 in PBS) were applied and incubated at room temperature for 2 hours. The tissues were then washed with PBS three times and mounted on glass slides. DAPI mounting medium (Southernbiotech, Fluoromount-G, Cat. No. 0100-01) was used to mount the coverglass on top of the glass slide with the sections. The slides were left to dry in air at room temperature overnight before images were acquired using a confocal microscope (Leica SP2).
+
+Electromyography. EMG signals were recorded from the gastrocnemius muscle with one reference needle electrode, one hydrogel working electrode (287 ± 14 μm) and one ground electrode. A 473 nm laser (200 mW/mm², 0.5 Hz, pulse width 50ms) was used for transdermal optical stimulation. EMG data triggered by optogenetic activation were collected through a DAM50 system.
+
+In vivo electrophysiology. Electrophysiological recordings were performed by connecting the pin connectors of optrode devices to a DAM50 recording system. Optical illumination was carried out using a 473 nm laser connected to the implanted optrode devices via a ferrule-sleeve-ferrule connecting system. The laser (10 mW/mm²) was pulsed at a frequency of 0.5 Hz with a pulse width of 50 ms during optical stimulation. Signals were sampled at 50 kHz and filtered between 1-1000 Hz. The amplitude and noise level of evoked potentials were analyzed using a MATLAB algorithm.
+Code availability
+
+The custom code used in this study is available from the corresponding author upon reasonable request.
+
+Data availability
+
+The data supporting the findings of this study are available from the corresponding author upon reasonable request.
+
+Acknowledgments
+
+We thank D. Kim and P. Anikeeva for the generous gifts of the plasmids and cell lines, Y. Liu for his assistance on electrochemical characterization, H. Kim for her assistance on mechanical property characterization and R. Chen for his thoughtful comments on our manuscript. This work was funded in part by the UMass Amherst Faculty Research Grant (P1FRG0000000295), the Brain&Behavior Research Foundation Young Investigator Grant (29878) and the National Institutes of Health (R00MH120279). This work made use of the UMass Amherst core facilities of Electron Microscopy, Light Microscopy, Raman, IR and XRF Spectroscopy, Roll-to-Roll Fabrication and Processing, and X-Ray Scattering, and Animal Care Service.
+a Control of metamorphic polymers’ amorphous-crystalline transition (COMPACT)
+
+Chemical crosslinking - Tunable optical properties
+Acidification - Supressed polymerchain interactions
+Induced nanocrystalline growth - oriented nanocrystalline growth under deformation
+
+Polyvinyl alcohol (PVA)    Glutaraldehyde (GA)    Si-OR group    Nanocrystalline domain    Water
+
+TEOS hydrolysis    Crosslinking    H^+    Drying under stretching    Annealing-Reswelling
+PVA dissolving    GA
+
+Force    Force
+Size
+
+b
+(1) Pristine    COMPACT    (2) Desiccated    Rehydration    (3) Re-hydrated
+
+c
+
+d
+
+e
+Water content fraction (%)
+(1)    (2)    (3)    (1)    (2)    (3)
+COMPACT
+
+f
+Thickness (T)
+Substrate
+
+g
+Length (L)
+Diameter (D)
+
+h
+Volume (V)
+Pristine    H^+(-)    H^+(+)
+
+i
+% Transmittance
+FTIR Peaks Assignments
+820    Si-O-Si bond
+559    Si-OH bending
+1080    Si-OH
+1140    Si-O-C bond
+heat flow (mW)
+Crystallinity percentage (%)
+PVA    28.4 ± 3.5
+(-)    21.6 ± 1.1
+(+)    12.7 ± 1.5
+Temperature (°C)
+
+j
+
+k
+Nanocrystalline Spacing (nm)
+Stretching percentages (%)
+Figure 1. COMPACT strategy for hydrogel materials miniaturization. a, Schematic illustrations of hydrogel network of metamorphic polymers’ amorphous-crystal transition (COMPACT). COMPACT treatment includes cross-linking with both glutaraldehyde (GA) and tetraethyl orthosilicate (TEOS), acidification and mechanical stretching. b-e, Representative photographs and water contents of TEOS-GA cross-linked polyvinyl alcohol (PVA) hydrogel with COMPACT treatment (+) and GA cross-linked hydrogel without acidification and stretching (-) at the pristine state (b), desiccated state (c) and re-hydrated state (d). Grid size: 5 mm. f, Shrinking behaviors of TEOS-GA cross-linked PVA (4 wt.% TEOS) hydrogel film with acidification treatment. Film thickness is quantified as mean ± standard deviation (s.d., paired student’s t-test, ***p=0.0004). Each dot represents one individual film. g, Shrinking behaviors of COMPACT hydrogel fibers (1-4 wt.% TEOS and 200% stretching). Hydrogel fibers’ length (black) and diameter (red) are quantified as mean ± s.d. Each dot represents one independent fiber. h, Shrinking behaviors of cross-linked hydrogel cylinders. The volume of TEOS-GA cross-linked hydrogel cylinders (4 wt.% TEOS) and with acidification treatment and GA cross-linked hydrogel cylinders without acidification treatment are compared with mean ± s.d. (unpaired student's t-test, F_{3,3}=6.084, *p=0.0161). Each dot represents one independent hydrogel cylinder. i, Fourier transform infrared (FTIR) spectroscopy of COMPACT (-) and COMPACT (+) hydrogels. j, Differential scanning calorimetry (DSC) profiles of COMPACT (-) and COMPACT (+) and their crystallinity percentages. k, Small-angle X-ray (SAXS) and wide-angle X-ray (WAXS) results of hydrogel materials in the desiccated state (mean ± s.d.). Inset: SAXS and WAXS 2D patterns.
+Figure 2. Controllable hydrogel fiber fabrication and its properties. a, A shrinking diagram of COMPACT (+) hydrogel fibers. Each dot (mean ± s.d.) represents an independent hydrogel fiber sample. The samples shaded in red areas are treated with acidification. b, Shrinking behaviors of COMPACT hydrogel fibers (4wt.% TEOS) prepared in different sizes of molds. Each dot (mean ± s.d.) represents one independent fiber (one(One-way ANOVA and Tukey's multiple comparisons test, F_{3,12}=0.9543, n.s.: not significant. p=0.4455). c, COMPACT hydrogel fibers’ optical properties of refractive index (blue) and normalized light transmittance (red) (mean ± s.d). Inset: representative photographs of 0 wt.% TEOS and 4 wt.% TEOS hydrogel membranes. Grid size: 1 mm. d, COMPACT hydrogel fibers’ mechanical properties of elastic modulus (blue) and stretchability percentage (red). Each dot represents one independent fiber sample. One-way
+ANOVA and Tukey’s multiple comparisons test were used to determine the statistical significance of elastic modulus: (F_{4,15}=20.51, ****p<0.0001;) and stretchability: (F_{4,15}=1.492, n.s. p=0.2543), respectively. **e**, Stability assessment of diameter reduction of COMPACT hydrogel fibers (3wt.% TEOS). Each dot (mean ± s.d.) represents one independent fiber (two-way ANOVA and Tukey's multiple comparisons tests). **f**, Cytotoxicity assessment of COMPACT (+) hydrogels. Hydrogel incubated media was used to culture with HEK293 cell cultures. Calcein-AM (green) was used to stain living cells and ethidium homodimer-1 (red) was used to stain dead cells. Cell death rates are presented as mean ± standard error (s.e.m., unpaired student’s t-test).
+a
+Light ray
+Inner core High RI
+Outer cladding Low RI
+Total internal reflection
+Cross section
+Nanocrystallites
+Neural Tissue
+Low light loss
+D<λ/4
+
+b
+Cladding Low RI
+Core High RI
+PVA+rGO Cladding
+Core High RI
+Light leakage
+
+c
+COMPACT (-)
+COMPACT (+)
+
+d
+Attenuation coefficient (dB/cm)
+Core
+Core plain cladding
+Core rGO cladding
+n.s. p=0.8771
+** p=0.0029
+** p=0.0056
+
+e
+AAV
+hydrogel fiber
+4 weeks
+(1) Viral injection+implantation
+(2) Photometric recording +behavior tests
+VTA
+
+f
+No social interaction
+Social interaction
+
+g
+Camera
+DAQ
+PC
+Objective
+Behavior camera
+405 nm
+470 nm
+Dichroic mirror
+Behavior chamber
+t
+
+h
+GCaMP6s fluorescence signals
+F (z-score)
+20 s
+Figure 3. Hydrogel optical neural probes for photometric recording with behavioral assessment. **a**, A schematic illustration of light transmission in a step-index hydrogel fiber. **b**, Schematic illustrations and representative photographs of a COMPACT core hydrogel fiber, a COMPACT core-plain-cladding hydrogel fiber, and a COMPACT core-rGO-cladding fiber. Scale: 200 μm. **c**, Representative photographs of blue light (480 nm) transmission from a COMPACT (-) core hydrogel fiber and a COMPACT (+) core hydrogel fiber into solutions containing Calcein fluorescent dye. **d**, Light attenuation coefficients of COMPACT core hydrogel fibers, COMPACT core-plain-cladding hydrogel fibers, and COMPACT core-rGO-cladding fibers (mean ± s.d., one-way ANOVA and Tukey's multiple comparisons test, F_{2,9}=13.3, **p=0.0021**). Each dot presents one independent hydrogel fiber sample. **e**, Experimental scheme for the viral injection, optical fiber implantation, photometric recording and social behavior tests. **f**, Representative images in mouse social interaction tests. **g**, A schematic illustration of fiber photometry recording setup with concurrent mouse social behavior tests. **h**, Normalized fluorescence intensity change (\( \Delta F/F_0 \)) of GCaMP6s in the VTA from mice social interactions. Blue bars indicate social interaction time.
+a
+CNTs-PVA electrode
+Hair
+
+b
+
+c
+
+d
+Specific Impedance (10^4 Ω·cm)
+CNT concentrations (wt.%)
+
+e
+Impedance @ 1 KHz (kΩ)
+Mold size (μm)
+Electrode diameter (μm)
+
+f
+Impedance @ 1 KHz (kΩ)
+Incubating time (weeks)
+Shrinking percentage (%)
+
+g
+Laser (473 nm)
+ChR2 expression
+scatic nerve
+Reference electrode
+Ground electrode
+Hydrogel electrode
+
+h
+Times (s)
+1 nV
+
+i
+Times (ms)
+1 nV
+
+j
+CNTs-PVA electrode
+Optical waveguide
+Waxing
+Casting
+
+k
+Optical ferrule
+Electrical pin
+Hydrogel waveguide
+Hydrogel electrode
+Ground screw
+
+l
+Electrical pin
+Optical ferrule
+
+m
+EYFP
+DAPI
+Overlay
+
+n
+Day 5
+Day 7
+Day 14
+300 μV
+1 s
+
+o
+Amplitudes (μV)
+Days
+Figure 4. Integrated multifunctional hydrogel neural probes. a, A Representative photograph of a carbon nanotube (CNT)-PVA hydrogel electrode as compared with a piece of human hair. Scale: 300 μm. b, A transmission electron microscopy (TEM) image of CNTs. Scale: 200 nm. c, Impedance at 1 kHz (red dots) and diameters of the electrodes (blue dots) fabricated different stretching percentages (mean ± s.d.). Each dot represents one independent hydrogel electrode. d, Impedance at 1 kHz of electrodes fabricated with different CNT concentrations (mean ± s.d.). Each dot represents one independent hydrogel electrode. e, Impedance at 1 kHz of electrodes (red dots) and diameters of the electrode (blue dots) fabricated with different sizes of molds (mean ± s.d.). Each dot represents one independent hydrogel electrode. f, Stability assessment on impedance (red dots) and diameters (blue dots) of hydrogel electrodes (mean ± s.d.). Each dot represents one independent hydrogel electrode. g, A schematic illustration of electrical recordings from mouse gastrocnemius muscles with a CNTs-PVA electrode in the presence of transdermal optical stimulation. h, Representative EMG signals recorded with CNT-PVA hydrogel electrodes upon transdermal optogenetic stimulations in Thy1::ChR2-EYFP mice (\( \lambda = 473 \) nm, 0.5 Hz, pulse width 50 ms, 200 mW/mm\(^2\)). Blue bars indicate the light illumination periods. i, Overlay plot of EMG peaks. j, A scanning electron microscopy (SEM) image at the cross-section of an integrated multifunctional neural probe containing an optical core and two CNT-PVA hydrogel electrodes. Scale: 100 μm. k-l, Photographs of a hydrogel optoelectronic device (optrode) before implantation and after implantation in a Thy1::ChR2-EYFP mouse brain. Scale: 2 mm. m, Confocal images of the expression of ChR2-EYFP in the VTA region of mouse. Scale: 50 μm. n, Representative in vivo electrophysiological signals recorded with optrodes upon optical stimulation (blue bars, \( \lambda = 473 \) nm, 0.5 Hz, pulse width 50 ms, 10 mW/mm\(^2\)). l, Amplitudes of electrophysiological signals
+recorded with optical stimulation on day 3, day 5, day 7, and day 14 post-implantation (\( \lambda = 473 \) nm,
+0.5 Hz, pulse width 50 ms, 10 mW/mm\(^2\), mean \( \pm \) s.e.m.).
+Reference
+
+1. Park, S. Il et al. Soft, stretchable, fully implantable miniaturized optoelectronic systems for wireless optogenetics. Nat. Biotechnol. 33, 1280–1286 (2015).
+2. Lu, C. et al. Flexible and stretchable nanowire-coated fibers for optoelectronic probing of spinal cord circuits. Sci. Adv. 3, (2017).
+3. Yang, Q. et al. High-speed, scanned laser structuring of multi-layered eco/bioresorbable materials for advanced electronic systems. Nat. Commun. 13, 6518 (2022).
+4. Park, S. et al. Adaptive and multifunctional hydrogel hybrid probes for long-term sensing and modulation of neural activity. Nat. Commun. 12, 3435 (2021).
+5. Tringides, C. M. et al. Viscoelastic surface electrode arrays to interface with viscoelastic tissues. Nat. Nanotechnol. 16, 1019–1029 (2021).
+6. Wu, Y. et al. Wireless multi-lateral optofluidic microsystems for real-time programmable optogenetics and photopharmacology. Nat. Commun. 13, 5571 (2022).
+7. Kathe, C. et al. Wireless closed-loop optogenetics across the entire dorsoventral spinal cord in mice. Nat. Biotechnol. 40, 198–208 (2022).
+8. Yoon, Y. et al. Neural probe system for behavioral neuropharmacology by bi-directional wireless drug delivery and electrophysiology in socially interacting mice. Nat. Commun. 13, 5521 (2022).
+9. Bonaccini Calia, A. et al. Full-bandwidth electrophysiology of seizures and epileptiform activity enabled by flexible graphene microtransistor depth neural probes. Nat. Nanotechnol. 17, 301–309 (2022).
+10. Canales, A. et al. Multifunctional fibers for simultaneous optical, electrical and chemical interrogation of neural circuits in vivo. Nat. Biotechnol. 33, 277–284 (2015).
+11. Li, J. et al. A tissue-like neurotransmitter sensor for the brain and gut. Nature **606**, 94–101 (2022).
+12. Cho, H. et al. Multiplex lithography for multilevel multiscale architectures and its application to polymer electrolyte membrane fuel cell. Nat. Commun. **6**, 8484 (2015).
+13. Eichelsdoerfer, D. J. et al. Large-area molecular patterning with polymer pen lithography. Nat. Protoc. **8**, 2548–2560 (2013).
+14. Saccone, M. A., Gallivan, R. A., Narita, K., Yee, D. W. & Greer, J. R. Additive manufacturing of micro-architected metals via hydrogel infusion. Nature **612**, 685–690 (2022).
+15. Park, S., Loke, G., Fink, Y. & Anikeeva, P. Flexible fiber-based optoelectronics for neural interfaces. Chem. Soc. Rev. **48**, 1826–1852 (2019).
+16. Tabet, A. et al. Modular Integration of Hydrogel Neural Interfaces. ACS Cent. Sci. **7**, 1516–1523 (2021).
+17. Wassie, A. T., Zhao, Y. & Boyden, E. S. Expansion microscopy: principles and uses in biological research. Nat. Methods **16**, 33–41 (2019).
+18. Oran, D. et al. 3D nanofabrication by volumetric deposition and controlled shrinkage of patterned scaffolds. Science. **362**, 1281–1285 (2018).
+19. Han, F. et al. Three-dimensional nanofabrication via ultrafast laser patterning and kinetically regulated material assembly. Science. **378**, 1325–1331 (2022).
+20. Guo, J. et al. Highly Stretchable, Strain Sensing Hydrogel Optical Fibers. Adv. Mater. **28**, 10244–10249 (2016).
+21. Liu, J. et al. Fatigue-resistant adhesion of hydrogels. Nat. Commun. **11**, 1–9 (2020).
+22. Lin, S. et al. Anti-fatigue-fracture hydrogels. Sci. Adv. **5**, (2019).
+23. Deng, J. et al. Electrical bioadhesive interface for bioelectronics. Nat. Mater. **20**, 229–236 (2021).
+24. Tan, P. et al. Solution-processable, soft, self-adhesive, and conductive polymer composites for soft electronics. Nat. Commun. **13**, 358 (2022).
+25. Peppas, N. A. & Merrill, E. W. Development of semicrystalline poly(vinyl alcohol) hydrogels for biomedical applications. J. Biomed. Mater. Res. **11**, 423–434 (1977).
+26. Choi, M., Humar, M., Kim, S. & Yun, S. H. Step-Index Optical Fiber Made of Biocompatible Hydrogels. Adv. Mater. **27**, 4081–4086 (2015).
+27. Guo, J. et al. Highly Stretchable, Strain Sensing Hydrogel Optical Fibers. Adv. Mater. **28**, 10244–10249 (2016).
+28. Beecroft, L. L. & Ober, C. K. High Refractive Index Polymers for Optical Applications. J. Macromol. Sci. Part A **34**, 573–586 (1997).
+29. Kurkjian, C. R., Krause, J. T. & Matthewson, M. J. Strength and fatigue of silica optical fibers. J. Light. Technol. **7**, 1360–1370 (1989).
+30. Lacour, S. P., Courtine, G. & Guck, J. Materials and technologies for soft implantable neuroprostheses. Nat. Rev. Mater. **1**, 16063 (2016).
+31. Király, B. et al. In vivo localization of chronically implanted electrodes and optic fibers in mice. Nat. Commun. **11**, 4686 (2020).
+32. Yi, J., Choe, G., Park, J. & Lee, J. Y. Graphene oxide-incorporated hydrogels for biomedical applications. Polym. J. **52**, 823–837 (2020).
+33. Kashyap, S., Pratihar, S. K. & Behera, S. K. Strong and ductile graphene oxide reinforced PVA nanocomposites. J. Alloys Compd. **684**, 254–260 (2016).
+34. Lammel, S. et al. Input-specific control of reward and aversion in the ventral tegmental area.
+Nature **491**, 212–217 (2012).
+
+35. Markovic, T. *et al.* Pain induces adaptations in ventral tegmental area dopamine neurons to drive anhedonia-like behavior. *Nat. Neurosci.* **24**, 1601–1613 (2021).
+
+36. Gunaydin, L. A. *et al.* Natural Neural Projection Dynamics Underlying Social Behavior. *Cell* **157**, 1535–1551 (2014).
+
+37. Kim, C. K. *et al.* Simultaneous fast measurement of circuit dynamics at multiple sites across the mammalian brain. *Nat. Methods* **13**, 325–328 (2016).
+
+38. Freedman, B. R. *et al.* Enhanced tendon healing by a tough hydrogel with an adhesive side and high drug-loading capacity. *Nat. Biomed. Eng.* **6**, 1167–1179 (2022).
+
+39. Law, S. S. Y. *et al.* Polymer-coated carbon nanotube hybrids with functional peptides for gene delivery into plant mitochondria. *Nat. Commun.* **13**, 2417 (2022).
+
+40. Gao, F., Viry, L., Maugey, M., Poulin, P. & Mano, N. Engineering hybrid nanotube wires for high-power biofuel cells. *Nat. Commun.* **1**, 2 (2010).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SupplementaryInformation.pdf

02ed829a2b0d25d6841e8c42a5b49a127bb3f49651373ce20617894c907bc9bb/peer_review/peer_review.md

@@ -0,0 +1,150 @@
+Peer Review File
+
+Human Neutralizing Antibodies Target a Conserved Lateral Patch on H7N9 Hemagglutinin Head
+
+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/.
+Reviewers' Comments:
+
+Reviewer #1:
+Remarks to the Author:
+The mechanisms by which antibodies neutralize and protect against viral infection remains an important focus of vaccinologist. In immunologically and structurally characterizing monoclonal antibodies that target H7 hemagglutinin, Jia and colleagues revealed how two antibodies work in concert to increase their antiviral potency in vivo (mouse model). It reflects how an optimal polyclonal antibody response may work to protect against disease by targeting different regions of the same molecule. It also highlights advantages of using a cocktail of monoclonal antibodies as a therapeutic.
+
+The strength of this study shows the robustness of protection of mAbs H7K1 and H7.HK2 in pre-exposure prophylactic experiments against live H7N9 challenges and, through structural studies, where the mAbs recognize the beta14-centered surface of H7 HA1. Moreover, antigenic changes in the globular head in more recent years (2016 to 2017) have rendered a lot of published neutralizing mAbs against ineffective in recognizing (although, this should have been demonstrated in the present study). The finding that both H7.HK1, H7.HK2 and the HA2 mAb H7.HK4 are still effective against more contemporary strains is promising.
+
+However, there are several weaknesses in the study that should be noted. The current work does not really add any significant insight into novel mechanisms of antibody inhibitory activity. There are already data to suggest and show that combination therapy can result in more pronounced protection in vivo. Post-exposure prophylactic experiments (one day post infection) are not robust enough to warrant its use as a therapeutic for either H7.HK1, H7.HK2 or H7HK.4. Given the inability of generating escape mutants against non-neutralizing antibodies that protect via Fc-mediated immunity, it would have been great if the present study had structurally resolved how H7HK.4 bound to the HA2 region – this would have increased the impact of this paper. It would also have benefited the reader if the authors had included past published (if they had access to the sequence or reached out to other authors) mAbs and show that they do not recognize more contemporary strains of H7 in the current study. Lastly, there is weak data to suggest there is ‘allosteric’ neutralization that is occurring. It is not clear if ‘allosteric’ is the best term to use in this case. One could argue that the only data is an in vivo experiment where two mAbs were mixed together to provide protection – no neutralization or structural experiment to warrant using the term allosteric.
+
+Other comments:
+
+Page 2; line 17: The data as presented arguably does not show allosteric mechanism of mAb neutralization. However, there is data to support augmented protection when mAbs are given as a cocktail.
+Page 4; line 10: The phrase ‘HA2-directed mAbs typically lacked neutralizing activity’ is arguably not accurate. Data generated from multiple investigators/groups in the past decade have clearly demonstrated that a large number of (murine and human) HA2 mAbs do have neutralizing activity against divergent subtypes. What is probably more typical is that HA2 mAbs are not as robust in neutralizing activity when compared to HA1 mAbs. The sentence should be rephrased.
+Page 10; line 19: The authors might mean, ‘this analysis is consistent with’
+
+Reviewer #2:
+Remarks to the Author:
+The authors discovered and characterized four human monoclonal antibodies, named H7.HK1, H7.HK2, H7.HK3 and H7/HK4, from a convalescent patient infected with A/Hong Kong/470129/2013 (H7N9) virus that the authors reported before (ref. 14). IgG B-cells of the PBMC samples from the patient were sorted against a soluble recombinant H7 HA protein from A/Shanghai/2/2013 (H7N9),
+and finally these four antibodies were recovered for further studies.
+
+Measured by ELISA, all four antibodies were found to bind to full-length H7 HA ectodomain. All antibodies except H7.HK4 also bound H7 HA head, and it was proposed H7.HK4 likely bound to stem region of H7 HA. H7.HK3 cross-reacted with an H15 HA, and H7.HK4 cross-reacted with H10 and H15 HAs.
+
+Neutralization studies were conducted with H7N9 pseudo-viruses and live replicating H7N9 viruses. H7.HK1 and H7.HK2 which shared the same VH and VL germline genes as well as the same CDR lengths neutralized H7N9 viruses isolated in original 2013 and later 2016 strains. But H7.HK3 nd H7.HK4 did not show any H7N9 neutralization under experimental conditions. All four antibodies did not neutralize the tested H3N2, H1N1, H5N1 and H9N2 viruses.
+
+Using cryo-EM, Fabs of H7.HK1 and H7.HK2 with H7 HA ectodomain from A/Shanghai/2/2013 (H7N9) were determined to 3.62 and 3.69 Å resolution, respectively. Both antibodies bound to similar H7 epitopes that are located in the HA head but away from HA receptor binding site. It was interesting that upon H7.HK1 and H7.HK2 binding to the H7 HA, the 220-loop from an adjacent H7 HA protomer became disordered under cryo-EM experiment conditions. Based on this finding, the authors proposed an allosteric mechanism of neutralization employed by these two antibodies.
+
+In mouse models for prophylactic and therapeutic studies, H7.HK1 and H7.HK2 protected H7N9 virus infection, while H7.HK3 and H7.HK4 showed no protection. Interestingly, for H7.HK4 which is human IgG1, the engineered antibody H7.Hk4.mIgG2a with a mouse Fc region showed protection against H7N9 virus.
+
+Overall, this is a systematic study of human antibodies against H7N9 viruses, and antibodies H7.HK1 and H7.HK2 were proposed to be one of the best human antibodies for neutralization potency and mouse protection efficacy, as well as their breath against the original 2013 H7N9 virus and recent 2016 H7N9 viruses.
+
+The major points:
+
+1. Title “Allosteric Neutralization by Human H7N9 Antibodies”. In cryo-EM structures, upon H7.HK1 and H7.HK2 binding to H7HA protomer, the 220-loop of an adjacent protomer became disordered under cryo-EM conditions and the antibodies might clash with the 220-loop if it was ordered. But more evidence appears to be needed to claim allosteric neutralization by these two antibodies. First of all, does the H7 HA bound with H7.HK1 or H7.HK2 still bind sialic acid receptors? Since the disordered 220-loop is in an adjacent protomer and with the documented information about HA heads being able to open and close in a breathing motion, such when binding FluA-20 binds at the HA head trimer interface, the 220-loop might still be functional for receptor binding in a temporal manner. It may also be possible that steric block of HA binding to receptors on the cell surface could contribute to virus neutralization.
+
+2. Abstract, Page 2, line 17-19: “Our data demonstrated an allosteric mechanism of mAb neutralization and augmented protection against H7N9 when a HA1-directed neutralizing mAb and a HA2-directed non-neutralizing mAb were combined”. In addition to the question of whether an allosteric mechanism is proved, the augmented protection may need to be caveated against H7N9 by H7.HK2 plus H7.HK4 (Fig. 3). Actually, in mouse prophylactic experiments, the mouse survival rate and body weight loss were about the same for H7.HK2 plus H7.HK4.mIgG2a vs H7.HK2 itself (Fig. 3A), although the combination of H7.HK2 and H7.HK4.mIgG2a was a little better than the control combination of H7.HK2 and H7.HK4.mIgG1 with different mouse Fc regions.
+
+3. The binding epitopes of H7.HK1 and H7.HK2 are actually on H7 HA lateral patch, as recently proposed in H1 HA structure (PMID 29255041, Raymond D.D. et al, PNAS, 2018, 115(1): 168-173).
+The authors may focus the discussion on this novel H7 lateral patch epitope, and potency and breath of H7.HK1 and H7.HK2.
+
+Minor points:
+
+1. Neuraminidase inhibitors Tamiflu and Relenza as well as endonuclease inhibitor Xofluza were approved to treat influenza infection, and were tested to be effective for H7N9 viruses. Since human H7N9 infection cases are rare, these might not yet have been reported for treatment efficacy. While antibody research for H7N9 virus is very important, these small molecule inhibitors are based on different targets. Accordingly, the relevant introduction in this manuscript could be revised.
+
+2. For clarity and comparison, please use H3 numbering for the H7 HA and Kabat numbering for the antibodies.
+
+3. Page 7, Table 1. Please add to the notes full strain names of all viruses.
+
+4. Fig. 2. Figs. 2D and 2E could be made clearer by including only key residues and excluding unrelated secondary structures and others.
+
+5. Page 23, line 17. "GenBank under accession # xxxxxxxxx to xxxxxxxxx" Please add.
+
+6. Page 24, Reference 6. Incomplete journal reference. Please add page number.
+
+7. Page 28. Please be consistent in use of abbreviations for the journal names.
+
+8. Supplementary Fig. 2. "(F) Representative density is shown for the interface of H7.HK1 heavy chain, light chain, and H7 HA. (G) Representative density is shown for the interface of H7.HK2 heavy chain, light chain, and H7 HA." Please specify what chains are in the different colors.
+
+9. Page 38. Supplementary Table 1. "Electron exposure (e-/Å2)" Please replace with " e-/Å2 "
+
+10. Page 38. Supplementary Table 1. Please truncate B-values to integers as decimal points not meaningful. Is the ligand cited here the antibody Fab? If so, please indicate as antibody or Fab. For the second entry, the B value of the Fab is lower than the HA? This would be unusual as the constant regions of the Fab are usually disordered as indicated in fig. S2E.
+Point-by-Point Response to the Reviewers
+
+Reviewer #1:
+
+The mechanisms by which antibodies neutralize and protect against viral infection remains an important focus of vaccinologist. In immunologically and structurally characterizing monoclonal antibodies that target H7 hemagglutinin, Jia and colleagues revealed how two antibodies work in concert to increase their antiviral potency in vivo (mouse model). It reflects how an optimal polyclonal antibody response may work to protect against disease by targeting different regions of the same molecule. It also highlights advantages of using a cocktail of monoclonal antibodies as a therapeutic.
+
+Response: We appreciate the Reviewer’s overall positive view of our study.
+
+The strength of this study shows the robustness of protection of mAbs H7.HK1 and H7.HK2 in pre-exposure prophylactic experiments against live H7N9 challenges and, through structural studies, where the mAbs recognize the beta14-centered surface of H7 HA1. Moreover, antigenic changes in the globular head in more recent years (2016 to 2017) have rendered a lot of published neutralizing mAbs against ineffective in recognizing (although, this should have been demonstrated in the present study). The finding that both H7.HK1, H7.HK2 and the HA2 mAb H7.HK4 are still effective against more contemporary strains is promising.
+
+Response: We agree with the Reviewer and added 3 previous mAbs, L4A-14, H7.167, and 07-5F01, for direct comparison with H7.HK2 (Results, page 7-8, Supplement Fig. 1). We corrected some mistakes and more accurately determined the mAb neutralization IC_{50}s (Fig. 1d and Table 1). Since the GD2016 pseudo virus did not generate high enough titers, we used the HK2017 H7N9 pseudo virus for neutralization comparison. Evaluated both by H7 antigen binding and pseudo virus neutralization, H7.HK2 is superior to the two best RBS-directed mAbs L4A-14 and H7.167 and matches the one best non-RBS mAb 07-5F01 against H7N9.
+
+However, there are several weaknesses in the study that should be noted. The current work does not really add any significant insight into novel mechanisms of antibody inhibitory activity. There are already data to suggest and show that combination therapy can result in more pronounced protection in vivo. Post-exposure prophylactic experiments (one day post infection) are not robust enough to warrant its use as a therapeutic for either H7.HK1, H7.HK2 or H7HK.4. Given the inability of generating escape mutants against non-neutralizing antibodies that protect via Fc-mediated immunity, it would have been great if the present study had structurally resolved how H7HK.4 bound to the HA2 region – this would have increased the impact of this paper. It would also have benefited the reader if the authors had included past published (if they had access to the sequence or reached out to other authors) mAbs and show that they do not recognize more contemporary strains of H7 in the current study. Lastly, there is weak data to suggest there is ‘allosteric’ neutralization that is occurring. It is not clear if ‘allosteric’ is the best term to use in this case. One could argue that the only data is an in vivo experiment where two mAbs were mixed together to provide protection – no neutralization or structural experiment to warrant using the term allosteric.
+Response: We appreciate the Reviewer’s view on the less studied HA2-directed antibodies. We confirmed the H7.HK4 epitope on HA2 by Western blot (Fig. 1c). We have also attempted to obtain the cryoEM structures of H7.HK3 and H7.HK4, but those Fabs did not bind the stabilized H7 trimer under typical cryoEM conditions. We have added the ELISA binding data of all four mAbs to the stabilized H7 trimer in Supplementary Fig. 3a. H7.HK4 did not bind the stabilized H7 trimer, likely explaining its lack of neutralization. For the two HA1-directed neutralizing mAbs H7.HK1 and H7.HK2, their epitopes are distinct from previous H7N9 neutralizing mAbs. Hence, these mAbs are valuable to complement other neutralizing antibodies for combination therapy. Regarding previous mAbs, we have included three for direct comparison with H7.HK2 – see response above. We also agree with the Reviewer that our data does not warrant using the term “allosteric” and have thus removed it.
+
+Other comments:
+
+Page 2; line 17: The data as presented arguably does not show allosteric mechanism of mAb neutralization. However, there is data to support augmented protection when mAbs are given as a cocktail. – We agree and removed “allosteric” from Abstract.
+
+Page 4; line 10: The phrase ‘HA2-directed mAbs typically lacked neutralizing activity’ is arguably not accurate. Data generated from multiple investigators/groups in the past decade have clearly demonstrated that a large number of (murine and human) HA2 mAbs do have neutralizing activity against divergent subtypes. What is probably more typical is that HA2 mAbs are not as robust in neutralizing activity when compared to HA1 mAbs. The sentence should be rephrased. – We agree and rephrased the sentence to: HA2-directed mAbs have also demonstrated neutralizing activity against divergent subtypes, although typically not as robust in neutralizing activity when compared to HA1 mAbs (page 4).
+
+Page 10; line 19: The authors might mean, ‘this analysis is consistent with’ – Yes, we revised the sentence to “this analysis is consistent with” (page 12).
+
+Reviewer #2:
+
+The authors discovered and characterized four human monoclonal antibodies, named H7.HK1, H7.HK2, H7.HK3 and H7/HK4, from a convalescent patient infected with A/Hong Kong/470129/2013 (H7N9) virus that the authors reported before (ref. 14). IgG B-cells of the PBMC samples from the patient were sorted against a soluble recombinant H7 HA protein from A/Shanghai/2/2013 (H7N9), and finally these four antibodies were recovered for further studies.
+
+Measured by ELISA, all four antibodies were found to bind to full-length H7 HA ectodomain. All antibodies except H7.HK4 also bound H7 HA head, and it was proposed H7.HK4 likely bound to stem region of H7 HA. H7.HK3 cross-reacted with an H15 HA, and H7.HK4 cross-reacted with H10 and H15 HAs.
+
+Neutralization studies were conducted with H7N9 pseudo-viruses and live replicating H7N9 viruses. H7.HK1 and H7.HK2 which shared the same VH and VL germline genes as well as the
+same CDR lengths neutralized H7N9 viruses isolated in original 2013 and later 2016 strains. But H7.HK3 nd H7.HK4 did not show any H7N9 neutralization under experimental conditions. All four antibodies did not neutralize the tested H3N2, H1N1, H5N1 and H9N2 viruses.
+
+Using cryo-EM, Fabs of H7.HK1 and H7.HK2 with H7 HA ectodomain from A/Shanghai/2/2013 (H7N9) were determined to 3.62 and 3.69 Å resolution, respectively. Both antibodies bound to similar H7 epitopes that are located in the HA head but away from HA receptor binding site. It was interesting that upon H7.HK1 and H7.HK2 binding to the H7 HA, the 220-loop from an adjacent H7 HA protomer became disordered under cryo-EM experiment conditions. Based on this finding, the authors proposed an allosteric mechanism of neutralization employed by these two antibodies.
+
+In mouse models for prophylactic and therapeutic studies, H7.HK1 and H7.HK2 protected H7N9 virus infection, while H7.HK3 and H7.HK4 showed no protection. Interestingly, for H7.HK4 which is human IgG1, the engineered antibody H7.Hk4.mIgG2a with a mouse Fc region showed protection against H7N9 virus.
+
+Overall, this is a systematic study of human antibodies against H7N9 viruses, and antibodies H7.HK1 and H7.HK2 were proposed to be one of the best human antibodies for neutralization potency and mouse protection efficacy, as well as their breath against the original 2013 H7N9 virus and recent 2016 H7N9 viruses.
+
+Response: We appreciate the Reviewer’s detailed review and summary of our study.
+
+The major points:
+
+1. Title “Allosteric Neutralization by Human H7N9 Antibodies”. In cryo-EM structures, upon H7.HK1 and H7.HK2 binding to H7HA protomer, the 220-loop of an adjacent protomer became disordered under cryo-EM conditions and the antibodies might clash with the 220-loop if it was ordered. But more evidence appears to be needed to claim allosteric neutralization by these two antibodies. First of all, does the H7 HA bound with H7.HK1 or H7.HK2 still bind sialic acid receptors? Since the disordered 220-loop is in an adjacent protomer and with the documented information about HA heads being able to open and close in a breathing motion, such when binding FluA-20 binds at the HA head trimer interface, the 220-loop might still be functional for receptor binding in a temporal manner. It may also be possible that steric block of HA binding to receptors on the cell surface could contribute to virus neutralization.
+
+Response: We agree with the Reviewer’s interpretation of our data and appreciate the likely mechanism of “steric occlusion”. We have thus removed the term “allosteric”.
+
+2. Abstract, Page 2, line 17-19: “Our data demonstrated an allosteric mechanism of mAb neutralization and augmented protection against H7N9 when a HA1-directed neutralizing mAb and a HA2-directed non-neutralizing mAb were combined”. In addition to the question of whether an allosteric mechanism is proved, the augmented protection may need to be caveated against H7N9 by H7.HK2 plus H7.HK4 (Fig. 3). Actually, in mouse prophylactic experiments, the mouse survival rate and body weight loss were about the same for H7.HK2 plus
+H7.HK4.mIgG2a vs H7.HK2 itself (Fig. 3A), although the combination of H7.HK2 and H7.HK4.mIgG2a was a little better than the control combination of H7.HK2 and H7.HK4.mIgG1 with different mouse Fc regions.
+
+Response: We appreciate the Reviewer’s careful assessment of our data. In Fig. 3a, we have now overlaid the survival and body weight data for H7.HK2 by itself and in combination (Page 14). The mAb combination did not improve the body weight trough as both groups fully protected mice from death with up to 7-8% weight loss. The mAb combination did improve the recovery of weight loss starting on day 7 post challenge (Fig. 3a, 4th row). We acknowledge the degree of improvement was moderate and noted it so both in the Abstract (page 2) and Discussion (page 19).
+
+3. The binding epitopes of H7.HK1 and H7.HK2 are actually on H7 HA lateral patch, as recently proposed in H1 HA structure (PMID 29255041, Raymond D.D. et al, PNAS, 2018, 115(1): 168-173). The authors may focus the discussion on this novel H7 lateral patch epitope, and potency and breath of H7.HK1 and H7.HK2.
+
+Response: We truly appreciate the Reviewer’s suggestion to focus on the epitopes of H7.HK1 and H7.HK2 that target the lateral patch of HA head initially identified with H1. We have now added Fig. 2g for the lateral patch analysis and a new Fig. 4 to define the HA1 lateral patch as a supersite for neutralizing antibodies (Results, page 14-15) and discussed this site of vulnerability (page 17-18).
+
+Minor points:
+
+1. Neuraminidase inhibitors Tamiflu and Relenza as well as endonuclease inhibitor Xofluzo were approved to treat influenza infection, and were tested to be effective for H7N9 viruses. Since human H7N9 infection cases are rare, these might not yet have been reported for treatment efficacy. While antibody research for H7N9 virus is very important, these small molecule inhibitors are based on different targets. Accordingly, the relevant introduction in this manuscript could be revised. – We appreciate this input and revised the relevant Introduction to include the small molecule inhibitors and removed “lack of treatment efficacy”.
+
+2. For clarity and comparison, please use H3 numbering for the H7 HA and Kabat numbering for the antibodies. – We have revised the manuscript to use H3 numbering for H7 HA and Kabat numbering for antibodies.
+
+3. Page 7, Table 1. Please add to the notes full strain names of all viruses. – Table 1 has been modified to include full names of all viral strains.
+
+4. Fig. 2. Figs. 2D and 2E could be made clearer by including only key residues and excluding unrelated secondary structures and others. – Figs. 2d and 2e have been modified to include only key residues.
+
+5. Page 23, line 17. "GenBank under accession # xxxxxxxx to xxxxxxxx" Please add. – GenBank accession numbers have been added.
+6. Page 24, Reference 6. Incomplete journal reference. Please add page number. – Reference 6 has been updated to include the page number.
+
+7. Page 28. Please be consistent in use of abbreviations for the journal names. – References have been updated to be consistent in journal name abbreviations.
+
+8. Supplementary Fig. 2. "(F) Representative density is shown for the interface of H7.HK1 heavy chain, light chain, and H7 HA. (G) Representative density is shown for the interface of H7.HK2 heavy chain, light chain, and H7 HA." Please specify what chains are in the different colors. – We have now specified the antibody chains.
+
+9. Page 38. Supplementary Table 1. "Electron exposure (e-/Å2)" Please replace with “ e–/Å2 ” – Done.
+
+10. Page 38. Supplementary Table 1. Please truncate B-values to integers as decimal points not meaningful. Is the ligand cited here the antibody Fab? If so, please indicate as antibody or Fab. For the second entry, the B value of the Fab is lower than the HA? This would be unusual as the constant regions of the Fab are usually disordered as indicated in fig. S2E. – B-values are now displayed as integers. “Ligand” here referred to N-linked glycans, not to Fab. This has been corrected.
+Reviewers' Comments:
+
+Reviewer #1:
+Remarks to the Author:
+The authors had done a very good job of addressing past concerns/suggestions to their manuscript. they have also done a very good job in addressing the second reviewer's concerns. Thank you.
+
+Reviewer #2:
+Remarks to the Author:
+My previous comments have been satisfactorily addressed.

02ed829a2b0d25d6841e8c42a5b49a127bb3f49651373ce20617894c907bc9bb/preprint/preprint.md

@@ -0,0 +1,642 @@
+Allosteric Neutralization by Human H7N9 Antibodies
+
+Xueling Wu
+xw2702@cumc.columbia.edu
+
+Columbia University   https://orcid.org/0000-0002-9752-3734
+
+Manxue Jia
+Aaron Diamond AIDS Research Center
+
+HanJun Zhao
+The University of Hong Kong   https://orcid.org/0009-0001-9014-6972
+
+Nicholas Morano
+Department of Biochemistry, Zuckerman Mind Brain Behavior Institute, Columbia University
+
+Hong Lu
+Aaron Diamond AIDS Research Center, Columbia University Vagelos College of Physicians and Surgeons
+
+Yin-Ming Lui
+State Key Laboratory for Emerging Infectious Diseases, Carol Yu Centre for Infection, Department of Microbiology, School of Clinical Medicine, Li Ka Shing Faculty of Medicine, University of Hong Kon
+
+Haijuan Du
+Vaccine Research Center, National Institute of Allergy and Infectious Diseases, National Institutes of Health
+
+Jordan Becker
+Department of Biochemistry, Zuckerman Mind Brain Behavior Institute, Columbia University
+
+Kwok-Yung Yuen
+The University of Hong Kong   https://orcid.org/0000-0002-2083-1552
+
+David Ho
+Columbia University Irving Medical Center   https://orcid.org/0000-0003-1627-149X
+
+Peter Kwong
+Vaccine Research Center, National Institute of Allergy and Infectious Diseases   https://orcid.org/0000-0003-3560-232X
+
+Lawrence Shapiro
+Department of Biochemistry, Zuckerman Mind Brain Behavior Institute, Columbia University
+
+Kelvin Kai-Wang To
+The University of Hong Kong   https://orcid.org/0000-0002-1921-5824
+Article
+
+Keywords:
+
+Posted Date: November 7th, 2023
+
+DOI: https://doi.org/10.21203/rs.3.rs-3429355/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+Additional Declarations: Yes there is potential Competing Interest. An U.S. provisional patent titled “Human Protective Neutralizing and Non-neutralizing Antibodies and Their Use against Influenza Viruses” was filed with filing No. 63/578,505 and XW, MJ, NCM, HL, DDH, KY, KKT, PDK, and LS as co-inventors.
+
+Version of Record: A version of this preprint was published at Nature Communications on May 27th, 2024. See the published version at https://doi.org/10.1038/s41467-024-48758-4.
+Allosteric Neutralization by Human H7N9 Antibodies
+
+Manxue Jia1†, Hanjun Zhao2,3‡, Nicholas C. Morano4,5†, Hong Lu1,5, Yin-Ming Lui2, Haijuan Du6, Jordan E. Becker4,5, Kwok-Yung Yuen2,3,7, David D. Ho1,5, Peter D. Kwong4,6, Lawrence Shapiro4,5, Kelvin Kai-Wang To2,3,7*, Xueling Wu1,5*
+
+1Aaron Diamond AIDS Research Center, Affiliate of Rockefeller University, New York, NY 10016, USA.
+
+2State Key Laboratory for Emerging Infectious Diseases, Carol Yu Centre for Infection, Department of Microbiology, School of Clinical Medicine, Li Ka Shing Faculty of Medicine, University of Hong Kong, Pokfulam, Hong Kong Special Administrative Region, China.
+
+3Centre for Virology, Vaccinology and Therapeutics, Hong Kong Science and Technology Park, Sha Tin, Hong Kong Special Administrative Region, China.
+
+4Department of Biochemistry, Zuckerman Mind Brain Behavior Institute, Columbia University, New York, NY 10027, USA.
+
+5Aaron Diamond AIDS Research Center, Columbia University Vagelos College of Physicians and Surgeons, New York, NY 10032, USA.
+
+6Vaccine Research Center, National Institute of Allergy and Infectious Diseases, National Institutes of Health, Bethesda, MD 20892, USA.
+
+7Department of Clinical Microbiology and Infection, University of Hong Kong-Shenzhen Hospital, Shenzhen, Guangdong, China.
+
+†These authors contributed equally to this work.
+
+*Correspondence to: Xueling Wu (email: xw2702@cumc.columbia.edu)
+
+Kelvin Kai-Wang To (email: kelvinto@hku.hk)
+Abstract: The avian influenza A virus H7N9 causes severe human infections with more than 30% fatality despite the use of neuraminidase inhibitors. Currently there is no H7N9-specific prevention or treatment for humans. From a 2013 H7N9 convalescent case occurred in Hong Kong, we isolated four H7 hemagglutinin (HA)-reactive monoclonal antibodies (mAbs) by single B cell cloning, with three mAbs directed to the HA globular head domain (HA1) and one to the HA stem region (HA2). Two clonally related HA1-directed mAbs, H7.HK1 and H7.HK2, potently neutralized H7N9 and protected mice from a lethal H7N9/AH1 challenge. Cryo-EM structures revealed that H7.HK1 and H7.HK2 bind to a β14-centered surface partially overlapping with the antigenic site D of HA1 and disrupt the 220-loop that makes hydrophobic contacts with sialic acid on the adjacent protomer, thus affectively blocking viral entry. The more potent mAb H7.HK2 retained full HA1 binding and neutralization capacity to later H7N9 isolates from 2016-2017, which is consistent with structural data showing that the antigenic mutations of 2016-2017 from the 2013 H7N9 only occurred at the periphery of the mAb epitope. The HA2-directed mAb H7.HK4 lacked neutralizing activity but protected mice from the lethal H7N9/AH1 challenge when engineered to mouse IgG2a enabling Fc effector function in mice. Used in combination with H7.HK2 at a suboptimal dose, H7.HK4 augmented mouse protection. Our data demonstrated an allosteric mechanism of mAb neutralization and augmented protection against H7N9 when a HA1-directed neutralizing mAb and a HA2-directed non-neutralizing mAb were combined.
+INTRODUCTION
+
+H7N9 is an avian influenza A group 2 virus first transmitted to humans in the spring of 2013 most likely through live poultry market exposure in China (1-3). The virus reemerged in the fall of 2013 and in the winter of later years, with the largest epidemic reported as the 5th wave in 2016-2017 (4-6). Though there is limited evidence for human-to-human transmission, few mutations in the hemagglutinin (HA) gene of the virus might be sufficient to overcome its inefficiency for human transmission (7-10). Like other influenza virus infections, the most common treatment against H7N9 is the neuraminidase inhibitor oseltamivir, but oseltamivir-resistant strains have emerged (11-13). Intravenous (i.v.) zanamivir, though not clinically approved, has been used on a compassionate basis in some severe cases because of favorable pharmacokinetics and in vitro susceptibility against oseltamivir-resistant strains (14, 15); however, the effectiveness of i.v. zanamivir against H7N9 has not been validated in large clinical trials. Despite the use of neuraminidase inhibitors, H7N9 case-fatality rate remains higher than 30%, and currently there is no licensed vaccine against H7N9 for humans. An endonuclease inhibitor baloxavir marboxil, targeting the virus polymerase acid, protected mice from lethal H7N9 challenge (16), but treatment for human H7N9 infection with this inhibitor has not been reported. Concerns for a major outbreak and lack of effective treatment warrant further studies to identify and develop human monoclonal antibodies (mAbs) with potent antiviral functions against H7N9.
+
+Because HA is the major target for influenza neutralizing antibodies, H7-reactive human mAbs have been isolated and characterized from H7N9 acute infections (17), convalescent cases (18), and H7N9 experimental vaccinees (19-21). The binding sites of these mAbs have been mapped to the HA globular head (HA1) and stem (HA2) domains. A subset of HA1-directed mAbs
+potently neutralized H7N9 and protected mice from H7N9 challenges at doses of 0.3, 1, 5 mg/kg or higher (17-20). These HA1-directed mAbs typically neutralized H7N9 by direct interference with or around the receptor (sialic acid) binding site (17, 19, 22). These epitopes correspond to the antigenic sites of A and B as previously mapped on the surface of H3 HA (23-25). Of note, significant antigenic drift has been documented in the HA gene of 2016-2017 H7N9 from the initial 2013 isolates (17, 26, 27). For example, Huang et al isolated 17 neutralizing mAbs from four cases infected in 2013 and 2014, yet only three of these mAbs were active against viral isolates from 2016-2017 (17). A broad mAb FluA-20 targeting the HA1 trimer interface did not mediate neutralization *in vitro*, but protected mice from viral challenges by disrupting HA trimers and inhibiting cell-to-cell spread of virus (21). HA2-directed mAbs typically lacked neutralizing activity, yet a few of them protected mice from H7N9 challenges at 5 mg/kg (20), especially when the mAbs were engineered as mouse IgG2a that has the highest Fc-mediated effector functions in mouse (28). These studies have not tested the combination of two or more mAbs that target different regions of H7N9 HA.
+
+In the post COVID-19 era, preparedness for future pandemics has risen with high enthusiasm. We aim to facilitate the development of human mAbs against H7N9, which has been considered one of the most serious pandemic threats. We have obtained peripheral blood mononuclear cells (PBMCs) from a 2013 H7N9 convalescent case in Hong Kong with the virus isolated as A/Hong Kong/470129/2013 H7N9 (14). The course of this infection lasted for about one month and the treatment required extracorporeal membrane oxygenation (ECMO) and i.v. zanamivir (14). Development of plasma neutralizing antibodies was evident at recovery. The PBMC sample we used to isolate mAbs was collected one year post recovery.
+RESULTS
+
+H7-reactive mAb isolation
+
+For H7-specific mAb isolation, we purchased a soluble recombinant H7 HA protein based on A/Shanghai2/2013 H7N9 for biotinylation, followed by streptavidin-PE conjugation. With this H7-PE bait, we stained 5 million PBMCs from the H7N9_HK2013 donor and sorted a total of 68 IgG+ B cells (defined as CD3+CD19+CD20+IgG+) that are H7-PE+ (Fig. 1A). Most of the sorted cells were at the borderline of H7-PE staining, but a few stained brightly for H7-PE. From the sorted B cells, we performed single B cell RT-PCR and recovered four H7-reactive mAbs – namely, H7.HK1, H7.HK2, H7.HK3, and H7.HK4.
+
+Measured by ELISA, the four reconstituted mAbs bound tightly to the H7N9 HA antigen used for H7-PE staining and to a recombinant H7N7 HA antigen based on A/Netherlands/219/2003 H7N7 (Fig. 1B, upper panels). Pre-treating the H7N9 HA with Endoglycosidase H (Endo H) had no effect on the mAb binding profiles, indicating that these mAbs do not rely on H7 glycans for binding (Fig. 1B, upper panels). After switching the ELISA coating antigen to HA1 of the matching H7N9 HA from A/Shanghai2/2013, the binding curves of H7.HK1, H7.HK2, and H7.HK3 were fully retained, indicating that these mAbs bind to the globular head domain HA1; in contrast, H7.HK4 lost binding to H7N9 HA1, indicating that its binding epitope is likely located in the HA2 stem domain (Fig. 1B, middle panels). Because of the documented antigenic drift for 2016-2017 H7N9 isolates, we also tested the mAb binding to HA1s from A/Guangdong/17SF003/2016 H7N9 and A/Hong Kong/125/2017 H7N9. The binding curves of H7.HK1, H7.HK2, and H7.HK3 to both 2016 and 2017 HA1s were fully retained, and H7.HK4 did not bind to any HA1s (Fig. 1B, middle panels). Additionally, we tested these mAbs for binding to 6 other non-H7 HA proteins. Though H7.HK1 and H7.HK2 did not react with any non-H7 HA,
+H7.HK3 cross-reacted with H15N8 HA, and H7.HK4 cross-reacted with H10N8 and H15N8 HAs (Fig. 1B, lower panels), which sequence-wise are the closest to H7 in group 2 influenza HA genes (29).
+
+H7-reactive mAb neutralization
+
+Using expression plasmids separately encoding H7 and N9 genes from A/Shanghai/4664T/2013 to pseudotype with HIV-1 NL4-3-lucΔenv backbone (30), we generated the H7N9 2013 pseudotype particles and tested mAb neutralization by a luciferase readout from single round infection of MDCK cells (Fig. 1C, left). H7.HK1 and H7.HK2 each potently neutralized the H7N9 2013 pseudovirus with IC_{50}s of 5 and 2 ng/mL respectively, while the other two mAbs H7.HK3 and H7.HK4 did not neutralize at up to 10 μg/mL (Fig. 1C, left, Table 1). Similarly, we generated pseudovirus using an expression plasmid encoding H7 from A/Guangdong/17SF003/2016 H7N9. H7.HK2 fully retained its potent neutralization against the H7N9 2016 pseudovirus with an IC_{50} of 2 ng/mL, and H7.HK1’s neutralization was reduced to an IC_{50} of 16 ng/mL, while the other two mAbs H7.HK3 and H7.HK4 did not neutralize (Fig. 1C, right, Table 1). We further assessed the mAb neutralization against three live replicating H7N9 viruses, Anhui1 (AH1), Zhejiang (ZJ), and the donor’s autologous isolate A/Hong Kong/470129/2013, for multiple rounds of infection in MDCK cells. Scored by the presence of cytopathic effect, mAbs H7.HK1 and H7.HK2 neutralized all three H7N9 live isolates with IC_{50}s ranging 0.3-1 μg/mL; however, they did not neutralize any non-H7N9 influenza isolates tested, indicating that these mAbs are specific to H7N9 (Table 1). The other two mAbs H7.HK3 and H7.HK4 did not neutralize any of the tested H7N9 and therefore were not tested against non-H7N9 viruses. The neutralization IC_{50}s of H7.HK1 and H7.HK2 using the pseudovirus were about 100-fold more potent than those using the live replicating viruses, suggesting that the pseudovirus neutralization
+is more sensitive thus useful for initial screening of neutralizing mAbs, which could then be confirmed with live replicating viruses. Similar differences in IC_{50} values have been reported for other HA-reactive mAbs tested by both pseudovirus and live replicating virus (31).
+
+<table>
+  <tr>
+    <th rowspan="2">mAb ID</th>
+    <th colspan="10">Neutralization IC_{50} (\mu g/mL) in MDCK cells</th>
+  </tr>
+  <tr>
+    <th>H7N9 2013 pseudovirus</th>
+    <th>H7N9 2016 pseudovirus</th>
+    <th>H7N9/ AH1</th>
+    <th>H7N9/ ZJ</th>
+    <th>H7N9/HK 470129</th>
+    <th>H3N2/ 400500</th>
+    <th>H1N1/ 415742</th>
+    <th>H5N1/ 459094</th>
+    <th>H5N1/ 1194</th>
+    <th>H9N2/ 1073</th>
+  </tr>
+  <tr>
+    <td>H7.HK1</td>
+    <td>0.005</td>
+    <td>0.016</td>
+    <td>0.3</td>
+    <td>0.3</td>
+    <td>0.4</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+  </tr>
+  <tr>
+    <td>H7.HK2</td>
+    <td>0.002</td>
+    <td>0.002</td>
+    <td>0.3</td>
+    <td>1.0</td>
+    <td>0.9</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+  </tr>
+  <tr>
+    <td>H7.HK3</td>
+    <td>&gt;10</td>
+    <td>&gt;10</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+    <td>ND</td>
+    <td>ND</td>
+    <td>ND</td>
+    <td>ND</td>
+    <td>ND</td>
+    <td>ND</td>
+  </tr>
+  <tr>
+    <td>H7.HK4</td>
+    <td>&gt;10</td>
+    <td>&gt;10</td>
+    <td>&gt;30</td>
+    <td>&gt;30</td>
+    <td>ND</td>
+    <td>ND</td>
+    <td>ND</td>
+    <td>ND</td>
+    <td>ND</td>
+    <td>ND</td>
+  </tr>
+</table>
+
+"ND" indicates "not done".
+
+H7-reactive mAb sequences
+
+Sequence analysis revealed that all four H7.HK mAbs are IgG1 (Table 2). H7.HK1 and H7.HK2 are clonal variants using IGHV4-59 for heavy chain with 8-10% somatic hypermutation (SHM) and a complementarity-determining region (CDR) H3 of 11 amino acids according to the Chothia definition (32-34), and IGKV2-28 for light chain with 6% SHM and a CDR L3 of 9 amino acids. Though clonally related, H7.HK1 and H7.HK2 share only 3 out of 13-15 amino acid SHMs in the heavy chain V-gene and 1 out of 8 amino acid SHMs in the light chain V-gene (Supplementary Fig. 1). A putative N-linked glycosylation site is present in the light chain CDR L1 of H7.HK1 and H7.HK2. H7.HK3 uses IGHV7-4-1 for heavy chain with 7% SHM and a CDR H3 of 14 amino acids, and IGKV1-5 for light chain with 5% SHM and a CDR L3 of 8 amino acids. A putative N-linked glycosylation site is also present in H7.HK3 at the heavy chain CDR H2. H7.HK4 uses IGHV4-61 for heavy chain with 7% SHM and a CDR H3 of 13 amino acids, and IGKV1-16 for light chain with 5% SHM and a CDR L3 of 9 amino acids (Table 2, Supplementary Fig. 1).
+Table 2 Genetic composition, epitope, and neutralization function of H7.HK mAbs
+
+<table>
+  <tr>
+    <th>mAb ID</th>
+    <th>Origin</th>
+    <th>Time point</th>
+    <th>Isotype</th>
+    <th>V-gene (SHM%)</th>
+    <th>CDR3 length in amino acid</th>
+    <th>Epitope</th>
+    <th>Neutralization</th>
+  </tr>
+  <tr>
+    <td>H7.HK1</td>
+    <td>Human</td>
+    <td>1 year post recovery</td>
+    <td>IgG1</td>
+    <td>HV4-59 (8%)<br>KV2-28 (6%)</td>
+    <td>H3: 11, L3: 9</td>
+    <td>H7 HA1</td>
+    <td>Yes</td>
+  </tr>
+  <tr>
+    <td>H7.HK2</td>
+    <td>Human</td>
+    <td>1 year post recovery</td>
+    <td>IgG1</td>
+    <td>HV4-59 (10%)<br>KV2-28 (6%)</td>
+    <td>H3: 11, L3: 9</td>
+    <td>H7 HA1</td>
+    <td>Yes</td>
+  </tr>
+  <tr>
+    <td>H7.HK3</td>
+    <td>Human</td>
+    <td>1 year post recovery</td>
+    <td>IgG1</td>
+    <td>HV7-4-1 (5%)<br>KV1-5 (7%)</td>
+    <td>H3: 14, L3: 8</td>
+    <td>H7 HA1</td>
+    <td>No</td>
+  </tr>
+  <tr>
+    <td>H7.HK4</td>
+    <td>Human</td>
+    <td>1 year post recovery</td>
+    <td>IgG1</td>
+    <td>HV4-61 (7%)<br>KV1-16 (5%)</td>
+    <td>H3: 13, L3: 9</td>
+    <td>H7 HA2</td>
+    <td>No</td>
+  </tr>
+</table>
+
+H7-reactive mAb structures
+
+For structural analysis, we generated the antibody fragments for antigen binding (Fabs) and expressed the H7 HA trimer by transient transfection of Expi293F cells. We froze grids containing the Fab:HA complexes and determined cryo-EM structures of each Fab bound to an H7 HA trimer. A resolution of 3.62 Å for H7.HK1 and 3.69 Å for H7.HK2 was achieved (Fig. 2A, Supplementary Fig. 2, Table S1). These complex structures demonstrate that H7.HK1 and H7.HK2 are highly superimposable (Fig. 2B) and their interactions with H7 are centered at β14 and extended to the surfaces of β10 and β19 (Fig. 2C). This β14-targeting surface partially overlaps with the antigenic site D towards sites A and B as previously mapped on H3 (23, 25). Analysis of the H7.HK1 epitope demonstrates that most interactions are driven by the heavy chain and consist of seven hydrogen bonds (Y52:E111, R97:G114, G102:S158, D103:T116, Y104:T156, Y104:S158, S106:T116) and one salt bridge (H53:E111) (Fig. 2D). The light chain is less involved in binding, making only one hydrogen bond (Y54:Q154) and weak hydrophobic interactions (Fig. 2E). The light chain of both H7.HK1 and H7.HK2 are glycosylated in CDR L1; this glycan plays no role in binding, but there is good density to support its presence. The epitope of H7.HK2 is similar to that of H7.HK1, only differing in slight contacts on the periphery (Supplementary Fig. 3A). Additionally, nearly all hydrogen bonds are conserved
+between the two antibodies (Supplementary Fig. 3B). However, the substitution of F61S in CDR L2 of H7.HK2 results in an additional hydrogen bond with HA G119. This substitution also shifts the orientation of H7.HK2 CDR L2 slightly so that Y54 interacts with T156 for H7.HK2 instead of Q154 for H7.HK1 (Supplementary Fig. 3C). Finally, as H53 is substituted with tyrosine in the heavy chain of H7.HK2, it does not make the H53:E111 salt bridge.
+
+To analyze the mechanism of neutralization, we first compared the binding site of H7.HK1 to that of four other H7-reactive antibodies with published structures, L4A-14, L4B-18, L3A-44 (PDB: 6II4, 6II8, 6II9) (17) and H7.167 (PDB: 5V2A) (19). This analysis demonstrates that the binding site of H7.HK1 is almost completely distinct from that of these previously published antibodies, which compete for the receptor binding site (RBS) (Fig. 2F). The binding site of H7.HK1 is also distant from that of 07-5F01, which was mapped to an escape mutation R65K (corresponding to R47K here) of HA1 (20) (Fig. 2F). Strikingly, the epitope of H7.HK1 (\( \beta 14 \)-centered) is extremely distal to the RBS of the protomer it interacts with and is closer to the RBS on the adjacent protomer. To further examine the relationship between the mAb binding site and RBS, the human receptor analogue Sialylneolacto-N-tetraose c (LSTc) was modeled into the RBS of H7 (PDB: 4BSE) (35) in the H7.HK1 complex. Interestingly, there were no steric clashes between H7.HK1 and sialic acid bound to the adjacent protomer, and no mAb interaction with RBS (Fig. 2G). However, the HA 220-loop (G209-G219) that makes hydrophobic contacts with sialic acid has no density present in the structure of H7.HK1 or H7.HK2 bound to HA, suggesting that these antibody binding causes 220-loop to become disordered. All previously examined H7 structures, as well as an additional cryo-EM structure in which Fab 1D12 is bound to the stem region of H7 HA (PDB: 6WXL) (36) have consistent electron density accounting for
+this loop. Alignments of the H7.HK1 complex structure with the crystal structure of H7 HA bound to LSTc (PDB: 4BSE) (35) demonstrate where the 220-loop would be when receptor is bound and that the light chain of H7.HK1 would clash with this loop (**Fig. 2H**), further supporting that H7.HK1 and H7.HK2 act by causing 220-loop to become disordered, thus preventing its interactions with the sialic acid receptor. The HA1 trimer interface mAb FluA-20 interacts with the non-RBS side of 220-loop on the protomer it interacts with (21). To our knowledge, the allosteric mechanism of neutralization employed by H7.HK1 and H7.HK2 is distinct from previously reported HA1-directed H7N9 neutralizing mAbs, which all directly compete with sialic acid for binding to HA on the protomer they interact with (*17, 19, 21, 22, 37*).
+
+Since the H7N9 HA gene has significantly evolved and changed in 2016-2017 compared to that of 2013 (with up to 13 amino acid substitutions in HA1), we examined the locations of mutated residues in the epitopes of H7.HK1 and H7.HK2 that consist of 32 contacting residues in HA1 for both mAbs (**Supplementary Fig. 4A**). There are four mutations in the binding site of H7.HK1 and H7.HK2 – namely, A112T/P, S118N, G119E, and R163K, appeared in 2016-2017 compared to the 2013 H7N9, and all four mutations are located at one side edge of the epitopes (**Supplementary Fig. 4B**), thus not altering the mAb interactions with HA1. This analysis is in consistency with the intact binding of H7.HK1 and H7.HK2 to both 2016 and 2017 HA1s aligned to the 2013 HA1 (**Fig. 1B**, middle panels) and H7.HK2’s full retention of neutralization against the H7N9 2016 pseudovirus (**Fig. 1C**).
+H7-reactive mAb mouse protection
+
+We next assessed the prophylactic and therapeutic effect of H7.HK mAbs as human IgG1 in a mouse lethal challenge model. To assess mAb prophylactic effect, balb/c mice (n = 5-10 per group from 1-2 experiments) were injected intraperitoneally (i.p.) with human H7N9 mAbs one day before intranasal (i.n.) challenge of 10-fold 50% lethal dose (10 LD_{50}) of A/Anhui/1/2013 H7N9 virus. Given 100 \( \mu \)g per mouse (equivalent to 5 mg/kg), the neutralizing mAbs H7.HK1 and H7.HK2 each fully protected mice without apparent weight loss (Fig. 3A, top panels); given 20 \( \mu \)g per mouse (equivalent to 1 mg/kg), H7.HK2 still fully protected mice from death (defined as \( \geq 20\% \) weight loss), with up to 8% average weight loss; H7.HK1 protected 7 out of 10 mice from death, with up to 12% average weight loss for mice that survived (Fig. 3A, upper middle panels). By day 2 post challenge, the weight preservation was significantly better in mice receiving 20 \( \mu \)g of H7.HK1 or H7.HK2 than mice receiving the placebo mAb or phosphate buffered saline (PBS). Mice receiving the non-neutralizing mAbs H7.HK3 or H7.HK4 (100 \( \mu \)g or 20 \( \mu \)g) were not protected and showed no difference from placebo mAb and PBS controls (Fig. 3A, top and upper middle panels).
+
+Since anti-HA2 stem mAbs have demonstrated Fc-mediated protection against influenza (38), we converted the anti-HA2 non-neutralizing mAb H7.HK4 to mouse IgG2a (mIgG2a) – an isotype that mediates strong Fc effector function in mice, and tested it for prophylaxis in the mouse challenge model, along with mouse IgG1 (mIgG1), which lacks Fc effector function in mice (28). Given 100 \( \mu \)g per mouse, H7.HK4 mIgG2a but not mIgG1 protected 4 out of 5 mice from death, with up to 17% average weight loss for mice that survived (Fig. 3A, lower middle panels). By day 3 post challenge, the weight preservation was significantly better in mice receiving
+H7.HK4 mIgG2a than mice receiving H7.HK4 mIgG1 or placebo mIgG2a. Though survived, mice receiving 100 \( \mu \)g H7.HK4 mIgG2a lost more weight than those receiving 20 \( \mu \)g neutralizing mAbs H7.HK1 or H7.HK2 (Fig. 3A, upper middle panels), indicating less prophylaxis efficiency for H7.HK4 (as mIgG2a) than H7.HK1 and H7.HK2.
+
+Since the H7.HK2 and H7.HK4 mAbs bind to different sites on the HA and protect through different mechanisms, we tested the combination of suboptimal dose of 20 \( \mu \)g H7.HK2 (as human IgG1) with 100 \( \mu \)g H7.HK4 mIgG2a in the mouse challenge model, using 20 \( \mu \)g H7.HK2 (as human IgG1) with 100 \( \mu \)g H7.HK4 mIgG1 as a control. Compared to this control group, which protected 9 out of 10 mice from death and lost up to 11% body weight for mice that survived, the combination of 20 \( \mu \)g of H7.HK2 (as human IgG1) with 100 \( \mu \)g H7.HK4 mIgG2a fully protected mice from death, with only up to 7% weight loss, and the weight difference was statistically significant between these two groups since day 3 post challenge (Fig. 3A, bottom panels), indicating a beneficial role of H7.HK4 in the mAb combination regimen.
+
+To assess mAb therapeutic effects, we first i.n. challenged mice (n = 5-10 per group from 1-2 experiments) with 10 LD\(_{50}\) of A/Anhui/1/2013 H7N9 virus, waited for one day, and then on day 1 post challenge i.p. injected mice with 100 \( \mu \)g H7.HK1 or H7.HK2 as human IgG1, or H7.HK4 as mIgG2a (Fig. 3B). Twelve and 13 out of 15 mice receiving 100 \( \mu \)g H7.HK1 or H7.HK2 one day after viral challenge initially lost weight similarly to placebo and PBS controls but then started to recover on day 5 after challenge. Therefore, the neutralizing mAbs H7.HK1 and H7.HK2 showed both prophylactic and therapeutic efficacies in the mouse lethal challenge model. None of the 5 mice receiving 100 \( \mu \)g H7.HK4 mIgG2a one day after challenge survived
+(Fig. 3B), indicating that H7.HK4 as mIgG2a demonstrated measurable prophylactic effect but not therapeutic efficacy.
+
+DISCUSSION
+
+Already endemic, adapted, and evolved in humans for 10 years, H7N9 continues to post risk and infect human cases exposed to infected poultry in China. While the current risk to public health is low, the pandemic potential of H7N9 is especially concerning if it were to gain the ability of sustained human-to-human transmission. Based on its biological features such as dual affinity for avian and human receptors, high case-fatality rate, resistance to neuraminidase inhibitors, and lack of pre-existing immunity in the human populations, there is an immediate need and interest to develop human mAb prophylaxis and therapeutics against H7N9, to which a specific treatment or licensed vaccine (for humans) is not available.
+
+In this study, we identified two HA1-directed clonally related human mAbs, H7.HK1 and H7.HK2, that neutralized H7N9 with potencies and mouse protection efficacies (prophylactic and therapeutic) in line with the best of previously reported H7N9 mAbs. Specifically, a combined phage library from three H7N9 convalescent cases yielded a single neutralizing mAb clone (18). Despite possible nonnative heavy and light chain pairing from phage display, the best member of the mAb clone, HNIgGA6, neutralized H7N9 and protected mice against a lethal challenge at 5 mg/kg with up to about 10% weight loss (18). Likewise, from a study of four H7N9 acutely infected cases, the best mAb L4A-14 cloned from plasmablast protected mice against a lethal challenge at 10 mg/kg with up to about 10% weight loss (17). The most potent mAb H7.167 from a study of EBV transformed B cells from five representative H7N9
+experimental vaccinees neutralized H7N9 and protected mice against a sub-lethal challenge of H7-PR8 at 1.65 mg/kg without apparent weight loss (19). The best HA1-directed neutralizing mAb 07-5F01 from a study of H7N9 experimental vaccinees’ plasmablasts protected mice against a lethal challenge at 0.3 mg/kg without apparent weight loss (20). The broad HA1 trimer interface mAb FluA-20 from a healthy donor with extensive influenza vaccinations lacked in vitro neutralization but protected mice against a sub-lethal challenge of H7-PR8 at 10 mg/kg without apparent weight loss (21). In comparison, H7.HK1 and H7.HK2 protected mice against a lethal challenge at 1 mg/kg with up to 12% weight loss.
+
+We have also structurally defined the epitopes of H7.HK1 and H7.HK2 to the β14-centered surface of H7 HA1, partially overlapping with the antigenic site D rather than the commonly targeted RBS and trimer interface by previous H7N9 mAbs (37), including the best reported human mAbs discussed above. Structural alignments and comparisons demonstrated that H7.HK1 and H7.HK2 interacted with H7 completely differently from L4A-14, H7.167, 07-5F01, and FluA-20. By escape mutations, a previous H3 neutralizing mAb D1-8 was mapped to the lower part of antigenic site D towards site E (39); this epitope partially overlaps with the H7.HK1 and H7.HK2 epitope described here. However, without structural data, the action of neutralization by D1-8 cannot be determined. Importantly, D1-8 does not react to H7, and likewise, H7.HK1 and H7.HK2 do not react to H3. Hence, D1-8 cannot replace the anti-H7N9 function of H7.HK1 and H7.HK2. The unique β14-targeting epitope on HA1 would render H7.HK1 and H7.HK2 favorable candidates for combination prophylaxis and therapy against H7N9 to augment protection efficacy and increase the genetic barrier for viral escape.
+H7N9 has evolved over time and its HA gene has significantly changed in 2016-2017 compared to that of 2013. Consequently, most neutralizing mAbs isolated from individuals infected or vaccinated with the 2013 H7 HA lost reactivity to 2016-2017 isolates, requiring updated H7 immunogens for mAb and vaccine development (17). We show that four mutations appeared in 2016-2017 are located at the periphery of the H7.HK1 and H7.HK2 epitopes and confirmed that the binding profiles of H7.HK1 and H7.HK2 are intact to both 2016 and 2017 HA1s as compared to 2013 HA1. We also showed that H7.HK2 fully retained its potent neutralization (IC_{50} of 2 ng/mL) against the H7N9 2016 pseudovirus, while H7.HK1’s neutralization IC_{50} was weakened from 5 to 16 ng/mL. Previous protective mAbs such as HNIgGA6 (18), H7.167 (19), and 07-5F01 (20) were not evaluated for reactivity to H7N9 2016-2017 isolates. L4A-14 was active against A/Guangdong/TH005/2017 (an avian virus related to A/Guangdong/17SF003/2016) but required 10 mg/kg, compared to 1 mg/kg of H7.HK1 and H7.HK2, for mice protection with up to about 10% weight loss (17). Compared to a 2013 H7N9 isolate, the neutralization IC_{50} of 07-5F01 was reduced by more than 100-fold against A/mallard/Netherlands/12/2000 H7N7 (20), and H7.167 did not recognize H7 from A/Netherlands/219/2003 H7N7 (19), to which all four H7.HK mAbs from the present study bound tightly.
+
+Lastly, we tested a suboptimal dose of H7.HK2 combining with the HA2-directed non-neutralizing mAb H7.HK4 against mouse lethal challenge. Compared to HA1 (head region of HA), HA2 (stem region) is genetically more conserved. Hence, HA2-directed mAbs typically display broader recognition of HA subtypes than HA1-directed mAbs. This is indeed the case for H7.HK4, i.e., in addition to H7N9 and H7N7, it also recognized the HAs from H10N8 and H15N8, to which both H7.HK1 and H7.HK2 had no reactivity. When converted to mouse IgG2a
+enabling Fc effector function in mice, H7.HK4 demonstrated measurable prophylactic protection at 5 mg/kg and augmented mouse protection of H7.HK2, supporting the inclusion of HA2-directed antibodies in a mAb combination regimen against H7N9.
+
+In summary, from a 2013 H7N9 convalescent case occurred in Hong Kong, we isolated two clonally related HA1-directed neutralizing mAbs, H7.HK1 and H7.HK2, that demonstrated prophylactic and therapeutic efficacies in a mouse lethal challenge model. Cryo-EM structures revealed a β14-centered site of vulnerability targeted by H7.HK1 and H7.HK2, which allowed full binding and neutralization capacity of H7.HK2 to the later 2016-2017 H7N9 isolates. This unique epitope renders H7.HK2 a favorable candidate for combination prophylaxis and therapy against H7N9, which may include multiple HA1-directed neutralizing mAbs targeting different epitopes and benefit from the inclusion of HA2-directed mAbs as well.
+
+METHODS
+
+Collection of human specimens
+
+A blood specimen was collected from the H7N9_HK2013 patient about one year after recovery from a hospitalized severe H7N9 infection. Written informed consent was obtained from the patient. The study was approved by the Institutional Review Board (IRB) of the University of Hong Kong and the Hospital Authority (Reference number: UW-13-265).
+
+Plasmids, viruses, antibodies, and cells
+
+Expression plasmids encoding the H7 hemagglutinin and N9 neuraminidase based on A/Shanghai/4664T/2013 H7N9 strain were obtained from Dr. Jianqing Xu (30). Codon-optimized gene encoding the H7 hemagglutinin of A/Guangdong/17SF003/2016 H7N9 was
+synthesized (Twist Bioscience) and cloned into pcDNA3.1 (Invitrogen). HIV-1 pNL4-3.Luc.R-E- backbone was obtained through the NIH HIV Reagent Program, as contributed by Dr. Nathaniel Landau. These plasmids were used to co-transfect 293T cells (ATCC, Manassas, VA) to generate H7N9 2013 and 2016 pseudoviruses. All live replicating influenza A viruses used in this study were isolated from patients and include A/Hong Kong/470129/2013 H7N9 (14), A/Zhejiang/DTID-ZJU01/2013 H7N9 (3), A/Anhui/1/2013 H7N9 (obtained from the China Center for Disease Control and Prevention), A/Vietnam/1194/2004 H5N1, A/Hong Kong/459094/2010 H5N1, A/Hong Kong/1073/1999 H9N2, A/Hong Kong/415742/2009 H1N1, and A/Hong Kong/400500/2015 H3N2. The non-H7N9 placebo mAb used in this study, AD358_n1, has been described (40) and is specific to HIV-1 gp120. Human embryonic kidney 293 cell line, of which the sex is female, is the parental cell for 293T and Expi293F cell lines. 293T was obtained from ATCC and maintained as adherent cells in complete DMEM medium at 37°C. 293T is highly transfectable and contains SV40 T-antigen. Expi293F was obtained from ThermoFisher and adapted to suspension culture in Expi293 Expression Medium at 37°C. The Madin-Darby Canine Kidney (MDCK) cell line, of which the sex is female, was obtained from ATCC and maintained as adherent cells in complete DMEM medium at 37°C.
+
+Single B cell sorting by fluorescence activated cell sorter (FACS)
+
+A soluble recombinant HA antigen based on A/Shanghai/2/2013 H7N9 (Immune Technologies, New York, NY) was biotinylated, followed by streptavidin mediated conjugation of phycoerythrin (PE) (Invitrogen). PBMCs were stained with an antibody cocktail to CD3-PE-CF594 (BD Biosciences, San Jose, CA), CD19-PE-Cy7 (BioLegend, San Diego, CA), CD20-APC-Cy7 (BioLegend), IgG-FITC (BD Biosciences), and IgM-V450 (BD Biosciences). In
+addition, live/dead yellow stain (Invitrogen) was used to exclude dead cells. After washing, cells were sorted using a multi-laser MoFlo sorter (Beckman Coulter, Jersey City, NJ). Fluorescence compensation was performed with anti-mouse Ig kappa chain beads (BD Biosciences) stained with each antibody in a separate tube. Individual B cells were sorted into a 96-well PCR plate, each well containing 20 μL lysis buffer, composed of 0.5 μL RNaseOut (Invitrogen), 5 μL 5x first-strand buffer, 1.25 μL 0.1M DTT, and 0.0625 μL Igepal (Sigma, St. Louis, MO). The PCR plate with sorted cells was frozen on dry-ice and then stored at -80°C. The total cell sample passing through the sorter was analyzed with FlowJo (TreeStar, Cupertino, CA).
+
+Single B cell RT-PCR, sequencing, and cloning
+
+From each sorted cell, the variable regions of IgG heavy and light chains were amplified by RT-PCR and cloned into expression vectors as previously described (40). Briefly, frozen plates with single B-cell RNA were thawed at room temperature, and RT was carried out by adding into each well 3 μL random hexamers at 150 ng/μL (Gene Link, Hawthorne, NY), 2 μL dNTP (each at 10 mM), and 1 μL SuperScript II (Invitrogen), followed by incubation at 42°C for 2 h. We note that these RT parameters may be suboptimal to those described previously (41, 42). After RT, 25 μL water was added to each well to dilute cDNA, and the cDNA plates were stored at -20°C for later use. The variable regions of heavy, kappa, and lambda chains were amplified independently by nested PCR in 50 μL, using 5 μL cDNA as template, with HotStarTaq Plus DNA polymerase (Qiagen) and primer mixes as described (41, 43). Cycler parameters were 94°C for 5 m, 50 cycles of 94°C for 30 s, 52-55°C for 30 s, and 72°C for 1 m, followed by 72°C for 10 m. The PCR amplicons were subjected to direct Sanger sequencing, and the antibody sequences were analyzed using IMGT/V-QUEST. Selected PCR sequences that gave productive gamma,
+kappa, and lambda chain rearrangements were re-amplified with custom primers containing unique restriction digest sites and cloned into the corresponding human gamma, kappa, and lambda chain expression vectors as described (40-42). Full IgG1 was expressed by co-transfecting Expi293F cells (ThermoFisher) with equal amounts of paired heavy and light chain plasmids and purified using recombinant Protein A agarose (ThermoFisher).
+
+ELISA, with and without Endo H treatment
+
+H7N9 HA and HA1 based on A/Shanghai/2/2013, A/Guangdong/17SF003/2016, A/Hong Kong/125/2017, and H7N7 HA based on A/Netherlands/219/2003 were purchased (Immune Technologies, New York, NY). Other non-H7 HA proteins were also purchased (Sino Biological, Chesterbrook, PA). ELISA plates were coated with HA or HA1 antigens at 2 \( \mu \)g/mL in PBS overnight at 4°C. For Endo H treatment, the required amount of antigen was diluted in 10x buffer and mixed with 1 \( \mu \)L Endo H (New England BioLabs, Ipswich, MA) for 1 h at 37°C; an equal amount of antigen (untreated) was processed under identical condition without Endo H. Both treated and untreated antigens were then diluted in PBS to coat ELISA plates at 2 \( \mu \)g/mL. Coated plates were blocked with 1% BSA (bovine serum albumin) in PBS for 1 h at 37°C, followed by incubation with serially diluted mAbs for 1 h at 37°C. Horseradish peroxidase (HRP)-conjugated goat anti-human IgG Fc (Jackson ImmunoResearch, West Grove, PA) was added at 1:10,000 for 1 h at 37°C. All ELISA incubation volumes were 100 \( \mu \)L/well except that 200 \( \mu \)L/well was used for blocking. Plates were washed between steps with 0.1% Tween 20 in PBS and developed with 3,3',5,5'-tetramethylbenzidine (TMB) (Novex, Life Technologies) for 5 m, with 1 M H\(_2\)SO\(_4\) as terminator and read at 450 nm.
+H7N9 neutralization assays
+
+H7N9 neutralization was first measured with a single-round infection of MDCK cells using H7N9 2013 and 2016 pseudoviruses as described (30). Neutralization curves were fitted by a 5-parameter nonlinear regression built in Prism (GraphPad Software, La Jolla, CA). The 50% inhibitory titers (IC_{50}s) were reported as the antibody concentrations required to inhibit infection by 50%. H7N9 neutralization was next measured using live replicating influenza viruses to infect MDCK cells as described (44). Briefly, serially diluted mAbs were incubated with 100 TCID_{50} (50% tissue culture infective dose) of an influenza virus at 37°C for 2 h, and 100 μL virus-mAb mixture was added to MDCK cells. After 1 h incubation, the virus-mAb mixture was removed, and minimum-essential medium with 2 μg/mL L-1-tosylamide-2-phenylethylchloromethyl ketone-treated trypsin (TPCK-trypsin) was added to each well. The plates were then incubated for 72 h, and cytopathic effects were recorded. The mAb concentration that protected 50% of 5 replicate wells from cytopathology was reported as IC_{50}.
+
+H7 HA production
+
+Soluble, disulfide-stabilized, fully cleaved H7 HA trimers were produced by transient co-transfection of plasmids encoding H7 HA (H7 SH13 DS2 6R) and Furin of Expi293F cells (Life Technologies) using Turbo293 transfection Reagent (Speed biosystem). After 5 days at 37°C, culture supernatants were harvested by centrifugation and concentrated 5-fold by Tangential Flow Filtration. The recombinant HA trimer was captured by Ni-NTA (Sigma-Aldrich) through a C-terminal 6xHis-tag. The imidazole eluant was combined 1:1 (v/v) with saturated ammonium sulfate, centrifuged at 4°C, and pellet removed. The supernatant was dialyzed against 500 mM NaCl, 50 mM Tris pH 8, and purified by size exclusion chromatography on a Superdex 200 Increase 10/300 GL column (Cytiva).
+Human mAb Fab preparation
+
+Human mAb Fab fragments were produced by digestion of the full IgG antibodies with immobilized Papain (ThermoFisher) equilibrated with 25 mM phosphate, 150 mM NaCl, pH 10, and 2 mM EDTA for 3 h. The resulting Fabs were purified from the cleaved Fc domain by affinity chromatography using protein A. Fab purity was analyzed by SDS-PAGE. All Fabs were buffer-exchanged into 25 mM phosphate, 150 mM NaCl, pH 7.0 prior to cryo-EM experiments.
+
+Cryo-EM sample preparation, data collection, and structure determination
+
+To determine the structures of H7.HK1 and H7.HK2 with H7 HA trimer, trimer was mixed with the antibody Fab at 1 to 1.2 molar ratio at a final total protein concentration of ~1 mg/mL and adjusted to a final concentration of 0.005% (w/v) n-Dodecyl β-D-maltoside (DDM) to prevent preferred orientation and aggregation during vitrification. Cryo-EM grids were prepared by applying 3 μL of sample to a freshly glow discharged carbon-coated copper grid (CF 1.2/1.3 300 mesh). The sample was vitrified in liquid ethane using a Vitrobot Mark IV with a wait time of 30 s, a blot time of 3 s, and a blot force of 0. Cryo-EM data were collected on a Titan Krios operating at 300 keV, equipped with a K3 detector (Gatan) operating in counting mode. Data were acquired using Leginon (45). The dose was fractionated over 50 raw frames. For all structures, the movie frames were aligned and dose-weighted (46) using cryoSPARC 3.4 (47); the CTF estimation, particle picking, 2D classifications, ab initio model generation, heterogeneous refinements, homogeneous 3D refinements and non-uniform refinement calculations were carried out using cryoSPARC 3.4 (47).
+Atomic model building and refinement
+
+For structural determination, a model of the antibody Fab was generated using SAbPred (48).
+
+The Fab model and the crystal structure of an H7 HA mutant (PDB: 6IDD) (10) was docked into the cryo-EM density map using UCSF Chimera (49) to build an initial model of the complex.
+
+The model was then manually rebuilt to the best fit into the density using Coot (50) and refined using Phenix (51). Interface calculations were performed using PISA (52). Structures were analyzed and figures were generated using PyMOL (http://www.pymol.org) and UCSF Chimera (49). Final model statistics are summarized in Table S1.
+
+Mouse prophylactic and therapeutic studies
+
+The mouse prophylactic and therapeutic studies were approved by the Committee on the Use of Live Animals in Teaching and Research (CULATR) of the University of Hong Kong (Reference number: 4011-16) and conducted in biosafety level 3 animal facilities as described previously (53). Female BALB/c mice of 6-8 weeks of age were obtained from the Laboratory Animal Unit of The University of Hong Kong. For prophylactic study, one day before virus inoculation, each mouse was administered with 100 \( \mu \)L of mAb at 1 mg/mL intraperitoneally. For therapeutic study, infected mice were administered with 100 \( \mu \)L of mAb at 1 mg/mL intraperitoneally at day 1 post viral challenge. Mice in the control groups were administered with either PBS or with a non-H7N9 mAb. On the day of virus infection, each mouse was inoculated with 10 LD\(_{50}\) (40 \( \mu \)L) of H7N9/AH1 virus through intranasal route. Virus inoculation was performed under ketamine (100 mg/kg) and xylazine (10 mg/kg) anesthesia. The mice were monitored for 14 days with disease severity score and body weight recorded daily. Disease severity were scored as follow:
+
+Score 0, apparently healthy; Score 1 (mild disease symptom), ruffled fur but still active; Score 2
+(medium disease symptom), ruffled fur, reduced activity and no weight gain; Score 3 (severe disease symptoms), ruffled fur, hunched posture, labored breathing and weight loss; Score 4 (moribund): very inactive, showing difficulty moving around and accessing to food and water, and weight loss. The predefined humane endpoints were either a weight loss of \( \geq 20\% \) or a disease severity score of 4. Mice were euthanized if the humane endpoints were reached.
+
+Statistical analysis
+
+GraphPad Prism was used to plot the ELISA data using sigmoidal dose-response with variable slope for curve fitting and the neutralization data using 5-parameter nonlinear regression for curve fitting. All quantitative data are presented as mean \( \pm \) standard error (SEM). GraphPad Prism was also used to plot the mouse Survival curves. Unpaired student’s t-test in GraphPad Prism was used for comparisons between groups, and a \( P \) value of less than 0.05 was considered statistically significant.
+
+Data availability
+
+Sequences of the heavy and light chain variable regions of four H7N9 human mAbs are available in GenBank under accession # xxxxxxxx to xxxxxxxx. The Cryo-EM reconstruction of H7.HK1 and H7.HK2 Fabs in complex with H7 SH13 DS2 6R HA has been deposited in the Electron Microscopy Data Bank as EMD-41422 and EMD-41441 and the Protein Data Bank (PDB: 8TNL and 8TOA). Materials will be made available to researchers with appropriate materials transfer agreements (MTAs). All inquiries should be sent to the corresponding authors.
+References
+
+1. R. Gao, B. Cao, Y. Hu, Z. Feng, D. Wang, W. Hu, J. Chen, Z. Jie, H. Qiu, K. Xu, X. Xu, H. Lu, W. Zhu, Z. Gao, N. Xiang, Y. Shen, Z. He, Y. Gu, Z. Zhang, Y. Yang, X. Zhao, L. Zhou, X. Li, S. Zou, Y. Zhang, X. Li, L. Yang, J. Guo, J. Dong, Q. Li, L. Dong, Y. Zhu, T. Bai, S. Wang, P. Hao, W. Yang, Y. Zhang, J. Han, H. Yu, D. Li, G. F. Gao, G. Wu, Y. Wang, Z. Yuan, Y. Shu, Human infection with a novel avian-origin influenza A (H7N9) virus. *N Engl J Med* **368**, 1888-1897 (2013).
+2. C. J. Bao, L. B. Cui, M. H. Zhou, L. Hong, G. F. Gao, H. Wang, Live-animal markets and influenza A (H7N9) virus infection. *N Engl J Med* **368**, 2337-2339 (2013).
+3. Y. Chen, W. Liang, S. Yang, N. Wu, H. Gao, J. Sheng, H. Yao, J. Wo, Q. Fang, D. Cui, Y. Li, X. Yao, Y. Zhang, H. Wu, S. Zheng, H. Diao, S. Xia, K. H. Chan, H. W. Tsoi, J. L. Teng, W. Song, P. Wang, S. Y. Lau, M. Zheng, J. F. Chan, K. K. To, H. Chen, L. Li, K. Y. Yuen, Human infections with the emerging avian influenza A H7N9 virus from wet market poultry: clinical analysis and characterisation of viral genome. *Lancet* **381**, 1916-1925 (2013).
+4. S. Su, M. Gu, D. Liu, J. Cui, G. F. Gao, J. Zhou, X. Liu, Epidemiology, Evolution, and Pathogenesis of H7N9 Influenza Viruses in Five Epidemic Waves since 2013 in China. *Trends Microbiol* **25**, 713-728 (2017).
+5. X. Wang, H. Jiang, P. Wu, T. M. Uyeki, L. Feng, S. Lai, L. Wang, X. Huo, K. Xu, E. Chen, X. Wang, J. He, M. Kang, R. Zhang, J. Zhang, J. Wu, S. Hu, H. Zhang, X. Liu, W. Fu, J. Ou, S. Wu, Y. Qin, Z. Zhang, Y. Shi, J. Zhang, J. Artois, V. J. Fang, H. Zhu, Y. Guan, M. Gilbert, P. W. Horby, G. M. Leung, G. F. Gao, B. J. Cowling, H. Yu, Epidemiology of avian influenza A H7N9 virus in human beings across five epidemics in mainland China, 2013-17: an epidemiological study of laboratory-confirmed case series. *The Lancet Infectious diseases* **17**, 822-832 (2017).
+6. W. Qi, W. Jia, D. Liu, J. Li, Y. Bi, S. Xie, B. Li, T. Hu, Y. Du, L. Xing, J. Zhang, F. Zhang, X. Wei, J. S. Eden, H. Li, H. Tian, W. Li, G. Su, G. Lao, C. Xu, B. Xu, W. Liu, G. Zhang, T. Ren, E. C. Holmes, J. Cui, W. Shi, G. F. Gao, M. Liao, Emergence and Adaptation of a Novel Highly Pathogenic H7N9 Influenza Virus in Birds and Humans from a 2013 Human-Infecting Low-Pathogenic Ancestor. *Journal of virology* **92**, (2018).
+7. J. Zhou, D. Wang, R. Gao, B. Zhao, J. Song, X. Qi, Y. Zhang, Y. Shi, L. Yang, W. Zhu, T. Bai, K. Qin, Y. Lan, S. Zou, J. Guo, J. Dong, L. Dong, Y. Zhang, H. Wei, X. Li, J. Lu, L. Liu, X. Zhao, X. Li, W. Huang, L. Wen, H. Bo, L. Xin, Y. Chen, C. Xu, Y. Pei, Y. Yang, X. Zhang, S. Wang, Z. Feng, J. Han, W. Yang, G. F. Gao, G. Wu, D. Li, Y. Wang, Y. Shu, Biological features of novel avian influenza A (H7N9) virus. *Nature* **499**, 500-503 (2013).
+8. Y. Shi, W. Zhang, F. Wang, J. Qi, Y. Wu, H. Song, F. Gao, Y. Bi, Y. Zhang, Z. Fan, C. Qin, H. Sun, J. Liu, J. Haywood, W. Liu, W. Gong, D. Wang, Y. Shu, Y. Wang, J. Yan, G. F. Gao, Structures and receptor binding of hemagglutinins from human-infecting H7N9 influenza viruses. *Science* **342**, 243-247 (2013).
+9. R. Xu, R. P. de Vries, X. Zhu, C. M. Nycholat, R. McBride, W. Yu, J. C. Paulson, I. A. Wilson, Preferential recognition of avian-like receptors in human influenza A H7N9 viruses. *Science* **342**, 1230-1235 (2013).
+10. Y. Xu, R. Peng, W. Zhang, J. Qi, H. Song, S. Liu, H. Wang, M. Wang, H. Xiao, L. Fu, Z. Fan, Y. Bi, J. Yan, Y. Shi, G. F. Gao, Avian-to-Human Receptor-Binding Adaptation of Avian H7N9 Influenza Virus Hemagglutinin. Cell Rep 29, 2217-2228 e2215 (2019).
+11. Y. Hu, S. Lu, Z. Song, W. Wang, P. Hao, J. Li, X. Zhang, H. L. Yen, B. Shi, T. Li, W. Guan, L. Xu, Y. Liu, S. Wang, X. Zhang, D. Tian, Z. Zhu, J. He, K. Huang, H. Chen, L. Zheng, X. Li, J. Ping, B. Kang, X. Xi, L. Zha, Y. Li, Z. Zhang, M. Peiris, Z. Yuan, Association between adverse clinical outcome in human disease caused by novel influenza A H7N9 virus and sustained viral shedding and emergence of antiviral resistance. Lancet 381, 2273-2279 (2013).
+12. R. Hai, M. Schmolke, V. H. Leyva-Grado, R. R. Thangavel, I. Margine, E. L. Jaffe, F. Krammer, A. Solorzano, A. Garcia-Sastre, P. Palese, N. M. Bouvier, Influenza A(H7N9) virus gains neuraminidase inhibitor resistance without loss of in vivo virulence or transmissibility. Nature communications 4, 2854 (2013).
+13. H. Marjuki, V. P. Mishin, A. P. Chesnokov, J. A. De La Cruz, C. T. Davis, J. M. Villanueva, A. M. Fry, L. V. Gubareva, Neuraminidase Mutations Conferring Resistance to Oseltamivir in Influenza A(H7N9) Viruses. Journal of virology 89, 5419-5426 (2015).
+14. K. K. To, W. Song, S. Y. Lau, T. L. Que, D. C. Lung, I. F. Hung, H. Chen, K. Y. Yuen, Unique reassortant of influenza A(H7N9) virus associated with severe disease emerging in Hong Kong. The Journal of infection 69, 60-68 (2014).
+15. P. L. Ho, W. C. Sin, J. F. Chan, V. C. Cheng, K. H. Chan, Severe influenza A H7N9 pneumonia with rapid virological response to intravenous zanamivir. The European respiratory journal 44, 535-537 (2014).
+16. M. Kiso, S. Yamayoshi, Y. Furusawa, M. Imai, Y. Kawaoka, Treatment of Highly Pathogenic H7N9 Virus-Infected Mice with Baloxavir Marboxil. Viruses 11, (2019).
+17. K. A. Huang, P. Rijal, H. Jiang, B. Wang, L. Schimanski, T. Dong, Y. M. Liu, P. Chang, M. Iqbal, M. C. Wang, Z. Chen, R. Song, C. C. Huang, J. H. Yang, J. Qi, T. Y. Lin, A. Li, T. J. Powell, J. T. Jan, C. Ma, G. F. Gao, Y. Shi, A. R. Townsend, Structure-function analysis of neutralizing antibodies to H7N9 influenza from naturally infected humans. Nat Microbiol 4, 306-315 (2019).
+18. Z. Chen, J. Wang, L. Bao, L. Guo, W. Zhang, Y. Xue, H. Zhou, Y. Xiao, F. Wu, Y. Deng, C. Qin, Q. Jin, Human monoclonal antibodies targeting the haemagglutinin glycoprotein can neutralize H7N9 influenza virus. Nature communications 6, 6714 (2015).
+19. N. J. Thornburg, H. Zhang, S. Bangaru, G. Sapparapu, N. Kose, R. M. Lampley, R. G. Bombardi, Y. Yu, S. Graham, A. Branchizio, S. M. Yoder, M. T. Rock, C. B. Creech, K. M. Edwards, D. Lee, S. Li, I. A. Wilson, A. Garcia-Sastre, R. A. Albrecht, J. E. Crowe, Jr., H7N9 influenza virus neutralizing antibodies that possess few somatic mutations. The Journal of clinical investigation 126, 1482-1494 (2016).
+20. C. J. Henry Dunand, P. E. Leon, M. Huang, A. Choi, V. Chromikova, I. Y. Ho, G. S. Tan, J. Cruz, A. Hirsh, N. Y. Zheng, C. E. Mullarkey, F. A. Ennis, M. Terajima, J. J. Treanor, D. J. Topham, K. Subbarao, P. Palese, F. Krammer, P. C. Wilson, Both Neutralizing and Non-Neutralizing Human H7N9 Influenza Vaccine-Induced Monoclonal Antibodies Confer Protection. Cell host & microbe 19, 800-813 (2016).
+21. S. Bangaru, S. Lang, M. Schotsaert, H. A. Vanderven, X. Zhu, N. Kose, R. Bombardi, J. A. Finn, S. J. Kent, P. Gilchuk, I. Gilchuk, H. L. Turner, A. Garcia-Sastre, S. Li, A. B.
+Ward, I. A. Wilson, J. E. Crowe, Jr., A Site of Vulnerability on the Influenza Virus Hemagglutinin Head Domain Trimer Interface. Cell **177**, 1136-1152 e1118 (2019).
+22. J. Wang, Z. Chen, L. Bao, W. Zhang, Y. Xue, X. Pang, X. Zhang, C. Qin, Q. Jin, Characterization of Two Human Monoclonal Antibodies Neutralizing Influenza A H7N9 Viruses. Journal of virology **89**, 9115-9118 (2015).
+23. R. G. Webster, W. G. Laver, Determination of the number of nonoverlapping antigenic areas on Hong Kong (H3N2) influenza virus hemagglutinin with monoclonal antibodies and the selection of variants with potential epidemiological significance. Virology **104**, 139-148 (1980).
+24. J. J. Skehel, D. J. Stevens, R. S. Daniels, A. R. Douglas, M. Knossow, I. A. Wilson, D. C. Wiley, A carbohydrate side chain on hemagglutinins of Hong Kong influenza viruses inhibits recognition by a monoclonal antibody. Proc Natl Acad Sci U S A **81**, 1779-1783 (1984).
+25. L. Popova, K. Smith, A. H. West, P. C. Wilson, J. A. James, L. F. Thompson, G. M. Air, Immunodominance of antigenic site B over site A of hemagglutinin of recent H3N2 influenza viruses. PloS one **7**, e41895 (2012).
+26. W. Zhu, J. Zhou, Z. Li, L. Yang, X. Li, W. Huang, S. Zou, W. Chen, H. Wei, J. Tang, L. Liu, J. Dong, D. Wang, Y. Shu, Biological characterisation of the emerged highly pathogenic avian influenza (HPAI) A(H7N9) viruses in humans, in mainland China, 2016 to 2017. Euro Surveill **22**, (2017).
+27. J. C. Kile, R. Ren, L. Liu, C. M. Greene, K. Roguski, A. D. Iuliano, Y. Jang, J. Jones, S. Thor, Y. Song, S. Zhou, S. C. Trock, V. Dugan, D. E. Wentworth, M. Z. Levine, T. M. Uyeki, J. M. Katz, D. B. Jernigan, S. J. Olsen, A. M. Fry, E. Azziz-Baumgartner, C. T. Davis, Update: Increase in Human Infections with Novel Asian Lineage Avian Influenza A(H7N9) Viruses During the Fifth Epidemic - China, October 1, 2016-August 7, 2017. MMWR Morb Mortal Wkly Rep **66**, 928-932 (2017).
+28. F. Nimmerjahn, J. V. Ravetch, Divergent immunoglobulin g subclass activity through selective Fc receptor binding. Science **310**, 1510-1512 (2005).
+29. S. J. Gamblin, J. J. Skehel, Influenza hemagglutinin and neuraminidase membrane glycoproteins. J Biol Chem **285**, 28403-28409 (2010).
+30. C. Qiu, Y. Huang, A. Zhang, D. Tian, Y. Wan, X. Zhang, W. Zhang, Z. Zhang, Z. Yuan, Y. Hu, J. Xu, Safe pseudovirus-based assay for neutralization antibodies against influenza A(H7N9) virus. Emerging infectious diseases **19**, 1685-1687 (2013).
+31. D. Corti, J. Voss, S. J. Gamblin, G. Codoni, A. Macagno, D. Jarrossay, S. G. Vachieri, D. Pinna, A. Minola, F. Vanzetta, C. Silacci, B. M. Fernandez-Rodriguez, G. Agatic, S. Bianchi, I. Giacchetto-Sasselli, L. Calder, F. Sallusto, P. Collins, L. F. Haire, N. Temperton, J. P. Langedijk, J. J. Skehel, A. Lanzavecchia, A neutralizing antibody selected from plasma cells that binds to group 1 and group 2 influenza A hemagglutinins. Science **333**, 850-856 (2011).
+32. C. Chothia, A. M. Lesk, Canonical structures for the hypervariable regions of immunoglobulins. J Mol Biol **196**, 901-917 (1987).
+33. C. Chothia, A. M. Lesk, A. Tramontano, M. Levitt, S. J. Smith-Gill, G. Air, S. Sheriff, E. A. Padlan, D. Davies, W. R. Tulip, et al., Conformations of immunoglobulin hypervariable regions. Nature **342**, 877-883 (1989).
+34. B. Al-Lazikani, A. M. Lesk, C. Chothia, Standard conformations for the canonical structures of immunoglobulins. J Mol Biol **273**, 927-948 (1997).
+35. X. Xiong, S. R. Martin, L. F. Haire, S. A. Wharton, R. S. Daniels, M. S. Bennett, J. W. McCauley, P. J. Collins, P. A. Walker, J. J. Skehel, S. J. Gamblin, Receptor binding by an H7N9 influenza virus from humans. Nature **499**, 496-499 (2013).
+36. C. S. Cheung, J. Gorman, S. F. Andrews, R. Rawi, M. Reveiz, C. H. Shen, Y. Wang, D. R. Harris, A. F. Nazzari, A. S. Olia, J. Raab, I. T. Teng, R. Verardi, S. Wang, Y. Yang, G. Y. Chuang, A. B. McDermott, T. Zhou, P. D. Kwong, Structure of an influenza group 2-neutralizing antibody targeting the hemagglutinin stem supersite. Structure **30**, 993-1003 e1006 (2022).
+37. C. Jiao, B. Wang, P. Chen, Y. Jiang, J. Liu, Analysis of the conserved protective epitopes of hemagglutinin on influenza A viruses. Front Immunol **14**, 1086297 (2023).
+38. D. J. DiLillo, G. S. Tan, P. Palese, J. V. Ravetch, Broadly neutralizing hemagglutinin stalk-specific antibodies require FcgammaR interactions for protection against influenza virus in vivo. Nature medicine **20**, 143-151 (2014).
+39. E. Benjamin, W. Wang, J. M. McAuliffe, F. J. Palmer-Hill, N. L. Kallewaard, Z. Chen, J. A. Suzich, W. S. Blair, H. Jin, Q. Zhu, A broadly neutralizing human monoclonal antibody directed against a novel conserved epitope on the influenza virus H3 hemagglutinin globular head. Journal of virology **88**, 6743-6750 (2014).
+40. M. Jia, H. Lu, M. Markowitz, C. Cheng-Mayer, X. Wu, Development of Broadly Neutralizing Antibodies and Their Mapping by Monomeric gp120 in Human Immunodeficiency Virus Type 1-Infected Humans and Simian-Human Immunodeficiency Virus SHIVSF162P3N-Infected Macaques. Journal of virology **90**, 4017-4031 (2016).
+41. T. Tiller, E. Meffre, S. Yurasov, M. Tsuji, M. C. Nussenzweig, H. Wardemann, Efficient generation of monoclonal antibodies from single human B cells by single cell RT-PCR and expression vector cloning. J Immunol Methods **329**, 112-124 (2008).
+42. X. Wu, Z. Y. Yang, Y. Li, C. M. Hogerkorp, W. R. Schief, M. S. Seaman, T. Zhou, S. D. Schmidt, L. Wu, L. Xu, N. S. Longo, K. McKee, S. O'Dell, M. K. Louder, D. L. Wycuff, Y. Feng, M. Nason, N. Doria-Rose, M. Connors, P. D. Kwong, M. Roederer, R. T. Wyatt, G. J. Nabel, J. R. Mascola, Rational design of envelope identifies broadly neutralizing human monoclonal antibodies to HIV-1. Science **329**, 856-861 (2010).
+43. J. F. Scheid, H. Mouquet, B. Ueberheide, R. Diskin, F. Klein, T. Y. Oliveira, J. Pietzsch, D. Fenyo, A. Abadir, K. Velinzon, A. Hurley, S. Myung, F. Boulad, P. Poignard, D. R. Burton, F. Pereyra, D. D. Ho, B. D. Walker, M. S. Seaman, P. J. Bjorkman, B. T. Chait, M. C. Nussenzweig, Sequence and structural convergence of broad and potent HIV antibodies that mimic CD4 binding. Science **333**, 1633-1637 (2011).
+44. K. K. To, A. J. Zhang, I. F. Hung, T. Xu, W. C. Ip, R. T. Wong, J. C. Ng, J. F. Chan, K. H. Chan, K. Y. Yuen, High titer and avidity of nonneutralizing antibodies against influenza vaccine antigen are associated with severe influenza. Clinical and vaccine immunology : CVI **19**, 1012-1018 (2012).
+45. A. Cheng, C. Negro, J. F. Bruhn, W. J. Rice, S. Dallakyan, E. T. Eng, D. G. Waterman, C. S. Potter, B. Carragher, Leginon: New features and applications. Protein Sci **30**, 136-150 (2021).
+46. S. Q. Zheng, E. Palovcak, J. P. Armache, K. A. Verba, Y. Cheng, D. A. Agard, MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat Methods **14**, 331-332 (2017).
+47. A. Punjani, J. L. Rubinstein, D. J. Fleet, M. A. Brubaker, cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat Methods **14**, 290-296 (2017).
+48. J. Dunbar, K. Krawczyk, J. Leem, C. Marks, J. Nowak, C. Regep, G. Georges, S. Kelm, B. Popovic, C. M. Deane, SAbPred: a structure-based antibody prediction server. Nucleic Acids Res **44**, W474-478 (2016).
+49. E. F. Pettersen, T. D. Goddard, C. C. Huang, G. S. Couch, D. M. Greenblatt, E. C. Meng, T. E. Ferrin, UCSF Chimera--a visualization system for exploratory research and analysis. J Comput Chem **25**, 1605-1612 (2004).
+50. A. Casanal, B. Lohkamp, P. Emsley, Current developments in Coot for macromolecular model building of Electron Cryo-microscopy and Crystallographic Data. Protein Sci **29**, 1069-1078 (2020).
+51. P. D. Adams, K. Gopal, R. W. Grosse-Kunstleve, L. W. Hung, T. R. Ioerger, A. J. McCoy, N. W. Moriarty, R. K. Pai, R. J. Read, T. D. Romo, J. C. Sacchettini, N. K. Sauter, L. C. Storoni, T. C. Terwilliger, Recent developments in the PHENIX software for automated crystallographic structure determination. J Synchrotron Radiat **11**, 53-55 (2004).
+52. E. Krissinel, K. Henrick, Inference of macromolecular assemblies from crystalline state. J Mol Biol **372**, 774-797 (2007).
+53. K. K. To, A. J. Zhang, A. S. Chan, C. Li, J. P. Cai, C. C. Lau, C. G. Li, A. S. Jahan, W. L. Wu, L. Li, A. K. Tsang, K. H. Chan, H. Chen, K. Y. Yuen, Recombinant influenza A virus hemagglutinin HA2 subunit protects mice against influenza A(H7N9) virus infection. Archives of virology **160**, 777-786 (2015).
+Acknowledgments
+We thank the patient for donating blood for the study. We thank Reda Rawi and Jeffrey C. Boyington for design of H7 SH13 DS2 6R used for structural analysis. Cryo-EM data were collected at the Columbia University Cryo-Electron Microscopy Center. We thank Shuofeng Yuan and Vincent Poon for assistance with the animal experiments.
+
+Funding
+• U.S. Department of Defense contract No. W911NF-14-C-0001 (DDH and XW)
+• Health@InnoHK, Innovation and Technology Commission of Hong Kong (KY and KKT)
+• Donations from Richard Yu and Carol Yu, Shaw Foundation Hong Kong, Michael Seak-Kan Tong, The Hui Ming, Hui Hoy and Chow Sin Lan Charity Fund Limited, Chan Yin Chuen Memorial Charitable Foundation, Marina Man-Wai Lee, Jessie and George Ho Charitable Foundation, Kai Chong Tong, Tse Kam Ming Laurence, Foo Oi Foundation Limited, Betty Hing-Chu Lee, and Ping Cham So (KY and KKT)
+• Bill and Melinda Gates Foundation grants FNIH SHAP19IUFV (LS) and INV-016167 (LS)
+• National Institutes of Health, National Institute of Allergy and Infectious Disease, Intramural Research Program of the Vaccine Research Center (PDK)
+
+Author contributions
+Conceptualization: XW, DDH, KY
+Methodology: XW, KKT, LS
+Investigation: MJ, HZ, NCM, HL, YL, HD, JEB
+Visualization: XW, NCM
+Funding acquisition: DDH, XW, KY, KKT, LS, PDK
+Project administration: XW, KKT
+Supervision: XW, KKT, KY, PDK, LS
+Writing – original draft: XW, KKT, NCM
+Writing – review & editing: XW, KKT, MJ, HZ, NCM, KY, DDH, PDK, LS
+
+Competing interests
+An U.S. provisional patent titled “Human Protective Neutralizing and Non-neutralizing Antibodies and Their Use against Influenza Viruses” was filed with filing No. 63/578,505 and XW, MJ, NCM, HL, DDH, KY, KKT, PDK, and LS as co-inventors.
+
+Additional information
+Supplementary Figs. 1 to 4
+Table S1
+Fig. 1 Isolation and characterization of human H7N9 mAbs in vitro. (A) FACS depicting the staining and selection of H7-specific B cells from donor H7N9.HK2013 PBMCs 1 year post recovery. SSC-A, side scatter area; FSC-A, forward scatter area. (B) ELISA binding curves of the indicated mAbs to soluble recombinant H7N9 HA and H7N7 HA (upper panels), with or without Endo H treatment, to the matching H7N9 HA1 from 2013 or HA1s from 2016 and 2017 (middle panels), and to 6 other non-H7 HA or HA1 proteins (lower panels). (C) Neutralization curves of H7.HK mAbs against H7N9 2013 (left) and 2016 (right) pseudoviruses infecting MDCK cells. Data shown are mean ± SEM.
+Fig. 2 Structural analysis of H7.HK1 and H7.HK2 in complex with H7 HA trimer. (A) Cryo-EM structures of H7.HK1 and H7.HK2 bound to H7 HA in the head region. (B) Top view of alignment of H7.HK1 and H7.HK2 complex structures. (C) Surface presentation of the H7.HK1 epitope (orange) on H7 HA1, with interacting CDRs shown. (D) H7.HK1 heavy chain forms seven hydrogen bonds and one salt bridge with H7 HA1. (E) H7.HK1 light chain forms one additional hydrogen bond with H7 HA1, and the interactions are stabilized by hydrophobic residues on the periphery of the light chain interface. (F) Modeling published structures of H7 HA1-binding antibodies (PDB: 6I14, 6I18, 6I19, 5V2A) onto the H7.HK1 bound structure, with an escape mutation R47K (green) reported for mAb 07-5F01. (G) Modeling the binding site of human receptor analogue LSTc (red) based on a previous crystal structure (PDB: 4BSE) onto H7 from the H7.HK1 complex, showing that H7.HK1 does not compete with sialic acid on the adjacent protomer (black). (H) Alignment of the H7.HK1 complex with a previous crystal structure of H7 (PDB: 4BSE) shows that the 220-loop (pink) required for sialic acid binding (G209-G219) is disorder in the complex structure and would clash with the H7.HK1 light chain if it were present. Green asterisk symbol denotes the <2 Å clash between the CDR L1 N33 and the predicted location of P212 on HA1.
+Fig. 3 Prophylactic and therapeutic effects of human H7N9 mAbs in mice i.n. challenged with 10 LD50 of A/Anhui/1/2013 H7N9. (A) Mice were i.p. injected with 100 µg (equivalent of 5 mg/kg) or 20 µg (equivalent of 1 mg/kg) of the indicated mAbs (as human IgG1 unless otherwise specified) one day before viral challenge; % survival (less than 20% weight loss) and % body weight of survived mice were plotted over time. (B) Mice were i.p. injected with 100 µg of the indicated mAbs one day after viral challenge; % survival and % body weight of survived mice were plotted over time. Arrows indicate the time when mAbs were administered. Control groups of a non-H7 placebo mAb and PBS were included. Data for each group were combined from 1-2 experiments and shown as mean – SEM. Asterisk symbols denote statistical significance with \( P \) values < 0.05.
+Heavy Chain V-gene
+
+-------------------FR1--------------- CDR H1 ------FR2------ ____CDR H2 ----------------------FR3-------------------
+IGHV4-59 QVQLQESGPGLVKPSETLSLTCTVSGGSIS SYYWSWIRQPPGKLEWIGIYYSGSTN YNPSLKSRTVTISVDTSKNQFSKLSSVTAADTAVYYC
+H7.HK1 QVQLQESGPGLVKPSETLSLTCSVSGGSIN SYYYWTIRQPPGKLEWGYIYHSGSTS YNPSLKSRTITSVAPSKNHFSLELSMTAADDTAVYYCAR
+H7.HK2 QVOLQSGSGPLLRSPSETLSLTCSVSGVSIN SYYYWSWRQPPGKALEWGYIYYSGSTN YNPSLKSRTVTISVDRSKNQFSLKMTSVTAADTARYFCAR
+IGHV7-4-1 QVQLVQSGSEELKKPGASVKSCASGYTFTF SYAMNWVRQAPGQGLEWMGWINTNTGNPTYAQGFTGRFVFSLDTSVSTAYLQICSLKAEDTAVYYC
+H7.HK3 QVQLVQSGSELRRPGASVKVSCASGYTFFT SYTINWVRQAPGQGLEWMGWINTSTGDPTYAQGFTGRFVFSLDTSVSTAYLEISRLLKAEDTAVYYCAR
+IGHV4-61 QVQLQESGPGLVKPSETLSLTCTVSGGSVSSGSYYWSWIRQPPGKLEWIGIYYSGSTN YNPSLKSRTVTISVDTSKNQFSKLSSVTAADTAVYYC
+H7.HK4 QVQLQESGPGLVKPSETLSLTCTVSGGSVRASAYAWSWIRQPPGKLEWIGDIYYSGSTN YNPSLKSRTVTLSVDTAKNRFSLRSVTAADTAVYHCAR
+
+Light Chain V-gene
+
+-------------------FR1--------------- CDR L1 ------FR2------ _CDRL2_ ----------------------FR3-------------------
+IGKV2-28 DIVMTQSPLSLPVTFGEPSASICSRRSQSLHHNGNYLDWLYLQPGQSPQLLIYLGSNRRASGVPDFRSGSGSGSGTDFTLKISRVEAEDEVGYYCC
+H7.HK1 DIVMTQSPLSLPVTFGEPSASICSNSQSLHHNGNYLDWLYLQPGQSPQLMIYLGINRAFVDPDRSGSGSGTDFTLKISRVEAEDEVGYYCC
+H7.HK2 DIVMTQSPLSLPVTFGEPSASICSNSQSLHHNGNYLDWYFQPKGQSPHLILYFLGNRAFVDPDRSGSGSGSGTDFTLKISRVEAEDEVGYYCC
+IGKV1-5 DIQMTQSPSSLASVGDRTVTTCRASQSI SSWLA WYQQKPGKAPKLLIYDASSLESVGSPRSFGSGSGSGSGTDFTLTISLQPDDFATYYC
+H7.HK3 DIQMTQSPSSLASVGDRTVTTCRASQSI SSWLA WYQQKPGKAPKLLIYKASSLESVGSPRSFGSGSGSGSGTDFTLTISLQPDDFATYYC
+IGKV1-16 DIQMTQSPSSLASVGDRTVTTCRASQGI SNYIA WFQQKPGKAPKSLIYAASSLSQSGSPRSFGSGSGSGSGTDFTLTISLQPDDFATYYC
+H7.HK4 DIQMTQSPSSLASVGDRTVTTCRASQGI RNYLA WFQQKPGQAPKSLIFAASSLHTGVPSRFSGSGSGSGTDFTLTISLQPDDFATYYC
+
+CDR3
+
+CDR H3 ------FR4---- __CDR L3 ----FR4---
+H7.HK1 LGGGHDGYGSDY WGQGTLTVTVSS MQALQTPTFFPGPGTRVDIK
+H7.HK2 QGIFGDGYGSDY WGQGTLTVTVSS MQGLQTPTFFPGPGTTVDIK
+H7.HK3 AFGLTVVRGGIVGVWGGGTTVTVSS QQYNSYSQTFGQGTKVEIK
+H7.HK4 ERYYYGSGDFDY WGQGTLTVTVSS QHYNSYPPTFGQGTKLEIK
+
+Supplementary Fig. 1 H7.HK mAb sequences. Protein sequences of the heavy and light chain variable regions of the H7.HK mAbs are aligned to the putative germline V-genes at top, with amino acid substitutions in red, and in magenta for substitutions shared between the clonally related mAbs H7.HK1 and H7.HK2. Spaces are added to maintain alignment; framework regions (FR) and complementarity-determining regions (CDRs) are indicated based on the Chothia nomenclature. Highlighted in yellow are the mAb residues (paratopes of H7.HK1 and H7.HK2) contacting the H7 antigen. The putative N-linked glycosylation sites on the light chain CDR L1 of H7.HK1 and H7.HK2 and the heavy chain CDR H2 of H7.HK3 are underlined.
+A
+H7.HK1        H7.HK2
+
+B
+H7.HK1
+H7.HK2
+
+C
+GSFSC Resolution: 3.62Å
+GSFSC Resolution: 3.69Å
+
+D
+
+E
+H7.HK1
+H7.HK2
+Local Resolution (Å)
+≤3.0
+3.5
+4.0
+4.5
+5.0
+5.5
+≥6.0
+FSC cutoff of 0.5
+
+F
+
+G
+Supplementary Fig. 2 Cryo-EM details of H7.HK1 and H7.HK2 in complex with H7 SH13 DS2 6R HA trimer.
+(A) Representative micrograph of H7.HK1 (left) and H7.HK2 (right). (B) Representative 2D class averages of H7.HK1 and H7.HK2. (C) The gold-standard Fourier Shell Correlation (FSC) resulted in a resolution of 3.62 Å for the overall map of H7.HK1 and 3.69 Å for the overall map of H7.HK2. Non-uniform refinement with C3 symmetry was used for both reconstructions. (D) The orientations of all particles used in the final refinement are shown as a heatmap. (E) The local resolution of the final overall map is shown contoured at 0.0989 for both structures. Resolution estimation was generated through cryoSPARC using an FSC cutoff of 0.5. (F) Representative density is shown for the interface of H7.HK1 heavy chain, light chain, and H7 HA. (G) Representative density is shown for the interface of H7.HK2 heavy chain, light chain, and H7 HA.
+Supplementary Fig. 3 Comparison of H7.HK1 and H7.HK2 binding to H7. (A) Differences in epitopes of H7.HK1 and H7.HK2. Majority of surface contacts are conserved, shown in orange. H7.HK1 specific surfaces are shown in magenta, and H7.HK2 specific surfaces are shown in cyan. (B) Hydrogen bonds and salt bridges formed by H7.HK1 and H7.HK2 with H7. (C) Differences in CDR L2 binding to H7 by H7.HK1 and H7.HK2 as a result of F61S substitution in H7.HK2. S61 forms an additional hydrogen bond with G119 of H7. Additionally, position of Y54 is shifted so that it forms a hydrogen bond with T156 for H7.HK2 instead of Q154 for H7.HK1.
+Supplementary Fig. 4 Antigenic drift of H7 HA1 in 2016-2017. (A) H7 HA1 protein sequences from the indicated viral isolates are aligned to the 2013 Hong Kong H7N9 autologous isolate at top, with identical amino acids shown in dots. Highlighted in yellow are the H7 residues (epitope) forming contacts with both mAbs H7.HK1 and H7.HK2. H7.HK1 specific epitopes are in magenta; H7.HK2 specific epitopes are in cyan. (B) Surface presentation of the H7 HA1 domain highlighting the epitopes (orange) of mAbs H7.HK1 and H7.HK2, with four mutations in red that appeared in the 2016-2017 viral isolates of H7N9. The sticks are interacting CDRs of mAb H7.HK1 heavy and light chains.
+Supplementary Table 1 Cryo-EM data collection, refinement, and validation statistics for H7 SH13 DS2 6R HA in complex with H7.HK1 and H7.HK2 Fabs.
+
+<table>
+  <tr>
+    <th></th>
+    <th>H7 SH13 DS2 6R<br>H7.HK1<br>(EMD-41422)<br>(PDB: 8TNL)</th>
+    <th>H7 SH13 DS2 6R<br>H7.HK2<br>(EMD-41441)<br>(PDB: 8TOA)</th>
+  </tr>
+  <tr>
+    <th colspan="3">Data collection and processing</th>
+  </tr>
+  <tr>
+    <td>Magnification</td>
+    <td>105000</td>
+    <td>105000</td>
+  </tr>
+  <tr>
+    <td>Voltage (kV)</td>
+    <td>300</td>
+    <td>300</td>
+  </tr>
+  <tr>
+    <td>Electron exposure (e--/Ų)</td>
+    <td>58</td>
+    <td>58</td>
+  </tr>
+  <tr>
+    <td>Defocus range (μm)</td>
+    <td>0.8-2</td>
+    <td>0.8-2</td>
+  </tr>
+  <tr>
+    <td>Pixel size (Å)</td>
+    <td>0.83</td>
+    <td>0.83</td>
+  </tr>
+  <tr>
+    <td>Symmetry imposed</td>
+    <td>C3</td>
+    <td>C3</td>
+  </tr>
+  <tr>
+    <td>Initial particle images (no.)</td>
+    <td>5713957</td>
+    <td>2339643</td>
+  </tr>
+  <tr>
+    <td>Final particle images (no.)</td>
+    <td>178347</td>
+    <td>191469</td>
+  </tr>
+  <tr>
+    <td>Map resolution (Å)</td>
+    <td>3.62</td>
+    <td>3.69</td>
+  </tr>
+  <tr>
+    <td>FSC threshold</td>
+    <td>0.143</td>
+    <td>0.143</td>
+  </tr>
+  <tr>
+    <th colspan="3">Refinement</th>
+  </tr>
+  <tr>
+    <td>Initial model used (PDB code)</td>
+    <td>6IDD</td>
+    <td>8TNL</td>
+  </tr>
+  <tr>
+    <td>Model resolution (Å)</td>
+    <td>3.62</td>
+    <td>3.69</td>
+  </tr>
+  <tr>
+    <td>FSC threshold</td>
+    <td>0.143</td>
+    <td>0.143</td>
+  </tr>
+  <tr>
+    <td>Model composition</td>
+    <td colspan="2"></td>
+  </tr>
+  <tr>
+    <td>Non-hydrogen atoms</td>
+    <td>16487</td>
+    <td>15570</td>
+  </tr>
+  <tr>
+    <td>Protein residues</td>
+    <td>2112</td>
+    <td>2109</td>
+  </tr>
+  <tr>
+    <td>Ligands</td>
+    <td>7</td>
+    <td>11</td>
+  </tr>
+  <tr>
+    <td><i>B</i> factors (Ų)</td>
+    <td colspan="2"></td>
+  </tr>
+  <tr>
+    <td>Protein</td>
+    <td>39.71</td>
+    <td>58.34</td>
+  </tr>
+  <tr>
+    <td>Ligand</td>
+    <td>58.78</td>
+    <td>48.38</td>
+  </tr>
+  <tr>
+    <td>R.m.s. deviations</td>
+    <td colspan="2"></td>
+  </tr>
+  <tr>
+    <td>Bond lengths (Å)</td>
+    <td>0.005</td>
+    <td>0.007</td>
+  </tr>
+  <tr>
+    <td>Bond angles (°)</td>
+    <td>1.121</td>
+    <td>1.231</td>
+  </tr>
+  <tr>
+    <th colspan="3">Validation</th>
+  </tr>
+  <tr>
+    <td>MolProbity score</td>
+    <td>1.65</td>
+    <td>2.23</td>
+  </tr>
+  <tr>
+    <td>Clashscore</td>
+    <td>5.45</td>
+    <td>12.08</td>
+  </tr>
+  <tr>
+    <td>Poor rotamers (%)</td>
+    <td>0.06</td>
+    <td>1.62</td>
+  </tr>
+  <tr>
+    <td>Ramachandran plot</td>
+    <td colspan="2"></td>
+  </tr>
+  <tr>
+    <td>Favored (%)</td>
+    <td>94.86</td>
+    <td>92.30</td>
+  </tr>
+  <tr>
+    <td>Allowed (%)</td>
+    <td>5.14</td>
+    <td>7.41</td>
+  </tr>
+  <tr>
+    <td>Disallowed (%)</td>
+    <td>0.0</td>
+    <td>0.29</td>
+  </tr>
+</table>

0353e3e01fab18c4f8a54e1bf5dc7c078c83cc6c97a3dfa961d5b8bce309eaef/peer_review/peer_review.md

@@ -0,0 +1,436 @@
+Peer Review File
+Bacterial N4-methylcytosine as an epigenetic mark in eukaryotic DNA
+
+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/.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The Editor invited me to comment on the data analysis part of the manuscript. The Editor may know that my expertise on algorithm design and machine learning theories are far away from the main topics/supplementary topics of the research.
+
+I looked the subsections in the Methods section. What I can understand is only about the subsections: Genome assembly, Prediction of protein-coding genes and phylogenetic analysis.
+
+From my understanding, the author has taken existing tools for genome assembly, prediction of protein-coding genes and phylogenetic analysis. For example, the genome assembly approach is the commonly used hybrid approach, combining de novo assemblies from Illumina short reads with those from PacBio long reads-based assemblies for joint polishing of the final genome. This approach is very advanced in the field.
+
+Prediction of protein-coding genes is also by an existing tool named BRAKER, a combination of GeneMark-ET and AUGUSTUS. The steps are used correctly, for example the construction of training sets.
+
+Except from this handful of technical comments, I feel that the paper is well organized with clear contribution presented at each section.
+
+Reviewer #2 (Remarks to the Author):
+
+This manuscript reports the important discovery of the acquisition of a bacterial method of DNA modification by a eukaryote via horizontal gene transfer. This discovery is interesting because it sheds light on the process of gains in DNA modification mechanisms, which otherwise happened much deeper in the evolutionary past. The authors have taken great pains to investigate the mechanisms and consequences of this change with detailed molecular investigation. This makes the manuscript quite complex and 'full', but the more general sections do a good job of explaining the context and importance.
+
+One part that is hard to follow for a non-expert and could be improved is the abstract. An initial sentence setting the scene would be useful, i.e. that you are talking about epigenetic modifications to DNA that influence gene expression. Then the statement of the general pattern in eukaryotes and bacteria is in terms of the modified base, but the statement of the enzymes they lack and possess is not easily connected to this earlier sentence for a non-expert, i.e. that C5-methyltransferase is the enzyme that modifies 5-methylcytosine. Line 22 has syntax "histone-read-DNA-write" that is abbreviated and not easily interpretable, and line 23 has a very complex enzyme name and compact description of effects on transposons. I appreciate this is a difficult topic, but I think the abstract could be improved to pull out the general points more clearly. The introduction is much clearer and perhaps a summary of some of this material would help the abstract: "Dna modification is used to regulate expression and in defence against transposable elements. In eukaryotes it normally involves X, in bacteria Y. We find that bdelloid rotifers lack X and have acquired mechanism Y by HGT from bacteria, and deploy this in defence against transposable elements".
+
+Lines 198 to 211 is rather complex to say that 4mC show same pattern as in Av-ref, but that 6mA show a different pattern and are also positively associated with TEs in AvL1. Maybe the detail could move to supplementary and just say what is the same and different.
+Line 230. Why did you use eggs for this part?
+
+Line 350. I lost track of what this entire section is for. Why is it interesting whether they show in vitro activity? What is the take-home message for this whole section?
+
+Curious that both bdelloids and monogononts have lost the C5-MTases, as well as nematodes and tardigrades. What is the hypothesis for why they lost that?
+
+Reviewer #3 (Remarks to the Author):
+
+In this work, Rodriguez and colleagues aim to demonstrate that bdelloid rotifers present N4-methylcytosine in the genome (4mC). To test this, they identify a putative 4mC methyltransferase that has no close homology to eukaryotic sequences. Then perform DIP-seq (antibody-based immunoprecipitation followed by sequencing) and use PacBio reads to identify 4mC and 6mA in the genomes of two very closely related “species” of Adineta vaga rotifers. Then express the putative methyltransferases in vitro and in bacteria to verify the methyltransferase activity. Additionally, the authors perform ChIP-seq of three histone-tail modifications (H3K4me3/H3K9me3/H3K27me3) to understand the possible crosstalk with 4mC (and 6mA). Finally, they try to identify a putative reader of 4mC in the Adineta genome, and perform in vitro experiments to test the capacity of a Methyl Binding Domain protein to perform this function.
+
+Overall, it would an exciting finding to show that 4mC is present in rotifer genomes. This manuscript presents a compelling case for this being the case, but offers quite weak support for the genomic distribution of 4mC in the genomes of these species or its possible functions. Then, many of the other claims raised across the manuscript are also weak. The manuscript is unfocused and too long, difficult to follow in many sections, and should be shortened drastically, with some sections dropped. Also, the methods used to analyse a lot of the data in this manuscript are not aligned with the standards of the field. In sum, I would not recommend the publication of this manuscript in the current form in Nature Communications.
+
+The first criticism in terms of novelty is that Marchantia polymorpha, a land plant, has been reported to contain 4mC recently (https://www.biorxiv.org/content/10.1101/2021.02.12.428880v1). In the Marchantia work, the authors identify the responsible 4mC methyl-transferases and are able to profile 4mC by using whole genome bisulfite sequencing and SMRT-seq, to achieve precise base-resolution identification of 4mC sites (unlike in this manuscript). That work does not imply the findings in rotifers are not meaningful and potentially very novel, but the authors should discuss that work throughout the manuscript and avoiding claims such as this being the first example of 4mC in a eukaryotic genome.
+
+In general, I believe that the methyltransferases identified in this manuscript are possibly active in rotifers, and they are likely originated from bacterial LGT (although the sister group in the tree are cyanophages, which supports a viral origin). But the distribution of 4mC (and 6mA) is way less clear, making the function of these modifications dubious in rotifers. DIP-seq is well known to be prone to artifacts, and the lack of coherence between the DIP-seq datasets between the two genomes used in this manuscript clearly shows that. Also, the analysis of “peak coverage” used in the figures is very unusual, this type of data, like ChIP-seq, is shown as relative enrichment versus input (IgG control), or directly as normalised read coverage. Then, all the SMRT-seq data is also shown as “coverage”, when methylation data across the literature is always shown as a fractional value. Methylation goes from 0 to 1 on a given base depending on the proportion of reads that support methylation, and that is what is informative when showing “metaprofiles” of any methylation mark. This is not done in this manuscript, and renders it very difficult to interpret. Also, the methods used by the authors do not
+show coherence between DIP-seq and SMRT-seq, which makes the whole claim that these marks are enriched on transposable elements dubious.
+
+Including the 6mA data is also quite problematic. The presence of 6mA is highly contested in animal genomes, and the data presented here is not robust. 6mA could be RNA contaminations (as found in mammalian species). The claims regarding this section do not offer much novelty and remain contentious.
+
+Would be good to show the intron-exon structure of the gene with RNA-seq data, and presenting some extra quality checks ensuring these genes are not found in bacterial scaffolds and are indeed encoded as host genes.
+
+Since both genomes are so close, one way to make the claims more robust would be to show orthologous regions across both species with similar profiles in terms of SMRT-seq base modification and DIP-seq.
+
+Figure 1f contains some errors. There is no need to establish “Phylum (Class)”. This is not very useful, plus, the authors do not stick to this classification for many groups. E.g. Fungi are not a Phylum. “Protist” is not a valid category, Amoebozoans are more closely related to Animals and Fungi than to Ciliates. Similarly, Ciliates should be sister group to the plant lineages. “Higher plants” does not have any meaning, there is no “lower” or “higher” plants, please refer to the correct group (Embryophytes?). Early Metazoa -> Non bilaterians. Green algae -> Chlorophyta? Some of the common names are confusing.
+
+Figure 2 is a mess. Why there are 2 letter codes? Some of them are in capital, others in lowercase, very confusing.
+
+Figure 2a. What does “coverage” represent? Peaks? What is “peak” coverage? Number of peaks that intersect with that window? That is not how these metaprofile plots are usually computed. These take the aligned reads from the DIP-seq (or ChIP-seq) and transform it to a read coverage plot, corrected by coverage (e.g. Counts per Million), and, if available, input DNA (IgG only). Those are the values then used to compute “coverage” plots such as these. The authors have used DeepTools in Extended Data Fig 2 c/d, but that figure is rather unclear. First, it shows enrichment in all features for both 4mC/5mC. It also says “Relative Fold Enrichment” (respect to what?). TS should be TSS (Transcriptional Start Site), and TTS should be TTS (Transcriptional Termination Site), these are standards in the field, why change them? DeepTools plots can be modified in Inkscape or Illustrator, that would be highly beneficial, since “Gene distance” is not very intuitive when representing Transposable Elements.
+
+The Figure 2a “coverage” metaprofiles do not fit at all with those shown in Extended Data Fig 2a/b. This highlights how unreliable are these DIP-seq techniques, which show very different results in two very closely related rotifers.
+
+Figure 2d. I believe the SMRT analysis software provides motifs enriched on methylated bases, why did the authors show MEME-ChIP which clearly does not fit this data?
+
+Figure 2g and h. PacBio counts should not be used, methylation levels are always shown as fractional levels since this is a technique with base resolution.
+
+Figure 2i indicates that Adineta has enriched 4mC and 6mA on transposable elements, when other plots in this same figure indicate that 6mA is not enriched on these elements. This is very confusing. Also, showing “dynamite plots” is bad practice, these values should be shown as a distribution (boxplot?). I still do not know why SMRT counts are used instead of fractional values.
+Figure 2c. What is “coverage”? PacBio base modification relies on coverage on positions, but then gives fractional values on any given position (0 to 1 values). In fact, “coverage” is a meaningless feature to determine the level of a base modification in any given position, but the “fraction methylation levels” should be averaged and displayed for a metaprofile like this. Again, I would recommend the authors to clarify what this is and maybe use DeepTools2. Why does it say 2.5 Kb window? The plot shows 3kb. Same for the one below, 500 bp window but there are 600 bp shown in the x axis.
+
+Extended Data Fig 3 is very hard to read, pixelated when zoomed in. There is not need to show these are “Circos” plots, simple Genome Browser representations would be much easier to interpret, zooming out or in depending on the features that need to be highlighted.
+
+The sequences of the conserved motifs shown in Figure 3f do not contain any CG dinucleotide, which in figure 2b was shown to be the preferred substrate of these enzymes according to SMRT-seq data. This is inconsistent and is not discussed.
+How are the + and – in Figure 3g determined? With an immuno-dot blot? Where is it? Not all combinations seem to be found in h and i panels. H and I panels are hard to interpret and poorly explained in the figure legend.
+
+Figure 4. H3K9me3 and H3K27me3 are not found usually on the same sites, one is constitutive chromatin and the other facultative (at least in mammals). This “expectation” is not very realistic. To validate the ChIP-seq, it would be good that the plots in Figure 4a/c would be sorted the same way, e.g. using the one to one orthology relationships between assemblies/species. Also, it would be good to add more information on the figure itself, reading the legend is confusing.
+
+Figure 4e/f. This plot is suspicious, the shape of the “peak” indicates that there is unclear smoothing within that window. Usually peak metaprofiles are centred around the peak summit (or centre), without setting hard borders as if they were genes, which tends to generate artefactual shapes as those shown here. Also, figure legends are barely visible.
+
+Figure 4h. Almost impossible to read, too pixelated. Better to show linear genome browser snapshots highlighting some associations instead of dumping too much data in a circular plot that is difficult to follow.
+
+Figure 4g. This should resemble figure 4f using a metaprofile displaying the fractional level of methylation (4mC/6mA) on those peaks, and not a “dynamite plot” that is difficult to interpret.
+
+The section on the small RNAs feels confusing and unnecessary for this manuscript.
+
+The section about SETDB1 as a possible reader of 4mC is also very weak and unnecessary. Not all animal MBD proteins bind to methylated cytosines, that is well established. MBD1/2/3/4 and MeCP2 are the families that have been shown to recognise methylated CpGs in other animals. In some cases, like Drosophila melanogaster, it harbours a highly divergent MBD2/3, that has been shown not to bind to preferentially to methylated CpGs, which is expected since Drosophila lacks 5mC methylation. The competition experiments shown in Figure 6 are not very strong indication of this domain preferentially binding to 4mC methylation.
+
+Figure 6f. Given the data shown in this manuscript, it feels premature to draw a model on how this system works in tardigrades.
+
+Claims such as “Finally, it demonstrates that horizontally transferred genes, contrary to the established view”. Lateral Gene Transfer is contested as a major source of innovation in eukaryotic lineages by some authors, but this is far from being “the established view”.
+Plots such as those in Extended Data Fig 5 e,f,g,i are basically noise. This is likely due to the non-standard ways of representing this data, and no information can be extracted from these.
+
+Reviewer #4 (Remarks to the Author):
+
+Comments to the author
+Rodriguez et al. performed an impressive research on 4mC occurrence in eukaryotic DNA. The major result underlines the 4mC presence in bdelloid rotifers by combining multiple trials I believe the results reported would be interesting to wide spectra of biologist and other specialists.
+Points raised:
+
+1. In figure 1a,1b, is it more intuitive to mark the direction with the N-terminal or C-terminal instead of using numbers to mark the direction? In addition, using cylinders to roughly represent multiple enzymes seems insufficient. Is there a way to show the structural differences more clearly? I’m not whetted on it, but think the article would be improved, if the authors consider this and take some actions.
+2. The author mentioned that for gene profiles, the modification density is much lower, and it seems the opposite compared to TE profile. Can the authors explain the opposite distribution phenomenon of gene profile and TE profile in a more understandable way?
+3. As far as I know, coverage level is a major effect factor on correction efficiency. In the range of 10x to 50x, the correction result increases as the coverage increases (https://www.uni-wuerzburg.de/fileadmin/07030400/AG_Genomics/Proovread/proovread-preprint.pdf). Therefore, does the author think that after increasing the coverage level, the overlap between SMRT-seq and DIP-seq is higher?
+4. In line 247, the author mentioned "At 4mC sites, CpG and CpA dinucleotides are the most prevalent, making up 74% of modified doublets." Please explain the calculation process.
+5. In line 259, the author reports the different methylation levels of 4mC and 6mA. Whether the conclusion drawn is related to the state of the sample. In other words, the methylation levels show differences at different time nodes.
+6. Please use the definite article accurately and polish other details that I haven't noticed. For example, “the genome-wide”, “We visualized the distribution”, etc.
+7. The ordinate of Fig. 2B is not clear, and the picture resolution needs to be improved.
+8. It may be due to the typesetting that makes Figure 2 look confusing. Please adjust the position and label of the picture to make it easier for readers to read. In addition, m4C, m6A in the 2F should be modified to 4mC, 6mA, please keep the abbreviations consistent.
+9. Some typoes should be revised.
+Reviewer #1 (Remarks to the Author):
+
+The Editor invited me to comment on the data analysis part of the manuscript. The Editor may know that my expertise on algorithm design and machine learning theories are far away from the main topics/supplementary topics of the research.
+
+I looked the subsections in the Methods section. What I can understand is only about the subsections: Genome assembly, Prediction of protein-coding genes and phylogenetic analysis.
+
+From my understanding, the author has taken existing tools for genome assembly, prediction of protein-coding genes and phylogenetic analysis. For example, the genome assembly approach is the commonly used hybrid approach, combining de novo assemblies from Illumina short reads with those from PacBio long reads-based assemblies for joint polishing of the final genome. This approach is very advanced in the field.
+
+Prediction of protein-coding genes is also by an existing tool named BRAKER, a combination of GeneMark-ET and AUGUSTUS. The steps are used correctly, for example the construction of training sets.
+
+Except from this handful of technical comments, I feel that the paper is well organized with clear contribution presented at each section.
+
+We thank the reviewer for the positive assessment of our genome assembly and annotation approaches and of the manuscript in general.
+
+Reviewer #2 (Remarks to the Author):
+
+This manuscript reports the important discovery of the acquisition of a bacterial method of DNA modification by a eukaryote via horizontal gene transfer. This discovery is interesting because it sheds light on the process of gains in DNA modification mechanisms, which otherwise happened much deeper in the evolutionary past. The authors have taken great pains to investigate the mechanisms and consequences of this change with detailed molecular investigation. This makes the manuscript quite complex and ‘full’, but the more general sections do a good job of explaining the context and importance.
+
+One part that is hard to follow for a non-expert and could be improved is the abstract. An initial sentence setting the scene would be useful, i.e. that you are talking about epigenetic modifications to DNA that influence gene expression. Then the statement of the general pattern in eukaryotes and bacteria is in terms of the modified base, but the statement of the enzymes they lack and possess is not easily connected to this earlier sentence for a non-expert, i.e. that C5-methyltransferase is the enzyme that modifies 5-methylcytosine. Line 22 has syntax “histone-read-DNA-write” that is abbreviated and not easily interpretable, and line 23 has a very complex enzyme name and compact description of effects on transposons. I appreciate this is a difficult topic, but I think the abstract could be improved to pull out the general points more clearly. The introduction is much clearer and perhaps a summary of some of this material would help the abstract: “Dna modification is used to regulate expression and in defence against transposable elements. In eukaryotes it normally involves X, in bacteria Y. We find that bdelloid rotifers lack X and have acquired mechanism Y by HGT from bacteria, and deploy this in defence against transposable elements”.
+
+Thank you for the valuable comments. We have reworded the beginning of the abstract using the suggested summary, to make it clearer for non-experts. We removed the extra “eggless” component from the enzyme name, limiting it to the more widely recognized SETDB1. The 150-word limit prevents us from fully deciphering the “read-write” syntax, which we believe is simplistic enough and is often used to facilitate understanding (it is used even in titles, e.g. ref. #70 Park et al. 2019).
+
+Lines 198 to 211 is rather complex to say that 4mC show same pattern as in Av-ref, but that 6mA show a different pattern and are also positively associated with TEs in AvL1. Maybe the detail could move to supplementary and just say what is the same and different.
+
+We have moved the details to the Supplementary Note 1, which helped to shorten the main text.
+
+Line 230. Why did you use eggs for this part?
+The advantage of using eggs to generate PacBio data was twofold: to discern methylation patterns that are most stable throughout all stages of development, and to increase PacBio coverage by minimizing bacterial contamination, which was achieved by washing and Clorox treatment of the collected eggs. We introduced additional clarifications in Methods.
+
+Line 350. I lost track of what this entire section is for. Why is it interesting whether they show in vitro activity? What is the take-home message for this whole section?
+
+Demonstration of in vitro activity was necessary because some N4C and N6A MTases were previously reported to display overlapping target base specificities (Jeltsch et al. 1999; Jeltsch 2001), and amino-MTases in general show polyphyletic origin (Bujnicki 1999). It was reassuring to see only 4mC addition and not 6mA addition by N4CMT, indicating that it acts exclusively as N4C-MTase and is unable to add 6mA marks. The most interesting finding, however, was the identification of preferred substrates, defined by the intrinsic ability of the chromodomain-less version to recognize certain motifs in DNA which confer substrate recognition properties to previously unmethylatable substrates, highlighting the dual recognition mode of the enzyme (the regional mode conferred by the chromodomain, and the sequence-specific mode likely inherited from bacteria via the TRD). We have emphasized these points in the text, while delegating the lengthy description of our search for minimal substrates to the Supplementary Note 2 and Supplementary Fig. 7.
+
+Curious that both bdelloids and monogononts have lost the C5-MTases, as well as nematodes and tardigrades. What is the hypothesis for why they lost that?
+
+The evolutionary forces that underlie loss of DNA methylation remain unclear, and such losses have occurred independently in multiple branches on the tree of life. While Drosophiliids and rhabditid nematodes are the best-studied case of C5-MT loss, recent observations also include myxozoan cnidarians (Kyger et al 2020). Other interesting examples include loss of de novo but not maintenance C5-MTases, as in Cryptococcus (Catania et al. 2020). It is hard to discern a single underlying reason, as it is more likely that lineage-specific circumstances would be involved. In general, DNA modification is secondary to histone modification in the epigenetic hierarchy, and lineage-specific compensatory changes are expected to evolve in each case. We would prefer to leave this discussion out of the scope of the present work, as it would be purely hypothetical.
+
+Reviewer #3 (Remarks to the Author):
+
+In this work, Rodriguez and colleagues aim to demonstrate that bdelloid rotifers present N4-methylcytosine in the genome (4mC). To test this, they identify a putative 4mC methyltransferase that has no close homology to eukaryotic sequences. Then perform DIP-seq (antibody-based immunoprecipitation followed by sequencing) and use PacBio reads to identify 4mC and 6mA in the genomes of two very closely related “species” of Adineta vaga rotifers. Then express the putative methyltransferases in vitro and in bacteria to verify the methyltransferase activity. Additionally, the authors perform ChIP-seq of three histone-tail modifications (H3K4me3/H3K9me3/H3K27me3) to understand the possible crosstalk with 4mC (and 6mA). Finally, they try to identify a putative reader of 4mC in the Adineta genome, and perform in vitro experiments to test the capacity of a Methyl Binding Domain protein to perform this function.
+
+Overall, it would an exciting finding to show that 4mC is present in rotifer genomes. This manuscript presents a compelling case for this being the case, but offers quite weak support for the genomic distribution of 4mC in the genomes of these species or its possible functions. Then, many of the other claims raised across the manuscript are also weak. The manuscript is unfocused and too long, difficult to follow in many sections, and should be shortened drastically, with some sections dropped. Also, the methods used to analyse a lot of the data in this manuscript are not aligned with the standards of the field. In sum, I would not recommend the publication of this manuscript in the current form in Nature Communications.
+
+We thank the reviewer for thorough analysis of our data, which helped significantly in improving the manuscript. As suggested, we have drastically shortened the text, creating four Supplementary Notes; aligned the analysis methods with the prevailing standards; and strengthened the evidence for non-random 4mC genomic distribution, as is expected from the presence of the chromodomain in N4CMT. We hope that our responses to the points below help to strengthen our case.
+
+The first criticism in terms of novelty is that Marchantia polymorpha, a land plant, has been reported to contain 4mC recently (https://www.biorxiv.org/content/10.1101/2021.02.12.428880v1). In the Marchantia work, the authors identify the responsible 4mC methyl-transferases and are able to profile 4mC by using whole genome bisulfite sequencing and SMRT-seq, to achieve precise base-resolution identification of 4mC sites (unlike in this manuscript). That work does not
+imply the findings in rotifers are not meaningful and potentially very novel, but the authors should discuss that work throughout the manuscript and avoiding claims such as this being the first example of 4mC in a eukaryotic genome.
+
+We adjusted the abstract and the text, focusing mainly on the epigenetic aspects of 4mC mark, as already emphasized in the title. We now discuss the Marchantia work, noting that the enzyme lacks additional N- or C-terminal domains and therefore adds 4mC marks indiscriminately across over one-half of all CpG sites in this genome. We removed the “first” claims, however please note with regard to priority that the time stamp for our work in Research Square is March 2021: www.researchgate.net/publication/353941577_Bacterial_N4-methylcytosine_as_an_epigenetic_mark_in_eukaryotic_DNA as we tried to obtain “Under review” status for our preprint.
+
+In general, I believe that the methyltransferases identified in this manuscript are possibly active in rotifers, and they are likely originated from bacterial LGT (although the sister group in the tree are cyanophages, which supports a viral origin). But the distribution of 4mC (and 6mA) is way less clear, making the function of these modifications dubious in rotifers. DIP-seq is well known to be prone to artifacts, and the lack of coherence between the DIP-seq datasets between the two genomes used in this manuscript clearly shows that.
+
+We employed DIP-seq primarily for initial evaluation of genome-wide distribution, being conscious of its disadvantages. As mentioned in the text, DIP-seq is a low-resolution methodology which limits the power of correlation analyses to the length of DNA fragments used for antibody binding (250-450 bp), not to mention residual IgG binding to non-modified fragments inherent to the method (original lines 220 and 488). Further, the DIP-seq datasets may be expected to overlap only partially for biological reasons, as we employed two distinct morphospecies, with 88% genome identity between Av-ref (clonally maintained under laboratory conditions for over 30 years) and the natural isolate AvL1 recently captured in the wild (line 124). We assumed that the lab strain may have gradually lost some of its methylation potential due to the lack of selective pressures experienced by the field populations, and indeed one of Av-ref N4CMT alleles shows lower activity than the other, differing by 6 aa substitutions. Nevertheless, after changing the visual peak representation method to DeepTools2, we are observing better coherence between two DIP-seq datasets (cf. Fig. 2a and Supp. Fig. 6a). The elevated 6mA signal from the gene set in AvL1 can be partially explained by the presence of unknown TE types in the gene set, which are erroneously annotated as genes by automated annotation pipelines but escape detection by repeat-mining tools due to the unusually low TE copy numbers in bdelloids (3-4% overall TE content, with most TE families containing only a few copies). Genometric correlation methods (original line 189, Table S4; added Supplementary Note 1 and Supplementary Fig.4) also detect positive correlation between 4mC and TE annotations in both genomes, while 4mC and gene annotations show negative correlation.
+
+Also, the analysis of “peak coverage” used in the figures is very unusual, this type of data, like ChIP-seq, is shown as relative enrichment versus input (IgG control), or directly as normalised read coverage.
+
+We agree that the term “peak coverage” may confuse the reader and have changed it to “IP occupancy” as in Fu et al. 2015 Fig. 3A (original ref #73), explaining the Y-axis in more detail in the Methods. Initially, DIP-seq profiles were obtained with annotatePeaks.pl from the Homer suite (Heinz et al. 2010), a program for performing peak annotation and associating ChIP-seq with nearby genes (or TEs) or expression data, which is actively maintained by the Benner Lab at UCSD. Unlike deepTools with graphical output, annotatePeaks.pl prints an output table with the “ChIP-Fragment Coverage, which is calculated by extending tags by their estimated ChIP-fragment length” (http://homer.ucsd.edu/homer/ngs/quantification.html) values, i.e. the ChIP-depth per annotation along coordinates, which we then post-processed for graphical output. We have now switched to deepTools suite to improve analysis and graphical representation and are plotting DIP-seq read occupancy instead of peak occupancy.
+
+Then, all the SMRT-seq data is also shown as “coverage”, when methylation data across the literature is always shown as a fractional value. Methylation goes from 0 to 1 on a given base depending on the proportion of reads that support methylation, and that is what is informative when showing “metaprofiles” of any methylation mark. This is not done in this manuscript, and renders it very difficult to interpret.
+
+We opted for presenting the methylation fraction (from 0 to 1) and linear representation (along contigs) in circos plots, which is represented as height (y-axis) in the corresponding methylation layer for 4mC and 6mA along with PacBio read coverage (Fig. 4h and Supplementary Fig. 6), as in Liang et al. 2018 Figure 2B and Figure S4C (former ref #74). In Fig. 2e, we present the methylation fraction distribution at modified sites detected by SMRT-seq, and note that most of the 4mC sites are nearly fully methylated, as is also visible on Circos plots. On “profile” plots with SMRT-seq methylation (Fig. 2c) we employed the same approach as in Beh et al. 2019 (ref #72 with Fig. 3C representing “Number of 6mA sites”), and Liang et al. 2018 (ref #74 with Fig. 3D,F representing “6mA occupancy”). Both methods used the distribution of methylated sites,
+employing different bin sizes for plotting purposes. We have removed the confusing term “coverage” and plotted 4mC and 6mA occupancy near the 5’ TE insertion boundary in the revised Fig. 2c using two distinct windows and bin values, with the y-axis now named “SMRT 4mC/6mA occupancy”.
+
+Also, the methods used by the authors do not show coherence between DIP-seq and SMRT-seq, which makes the whole claim that these marks are enriched on transposable elements dubious.
+
+Indisputably, the DIP-seq methodology is prone to sensitivity issues, nevertheless the overlap that we find between DIP-seq and SMRT-seq modification patterns is considerable, with 36% of 4mC DIP-seq peaks and 32% of 6mA peaks overlapping with 4mC and 6mA identified by SMRT analysis, respectively. Given the modest percentage of modified bases, with 0.0643% of the total cytosines in the assembly (21,016 4mC modifications) and 0.0236% of total adenines (17,886 6mA modifications) defined by SMRT-seq, the overlap between DIP-seq and SMRT-seq is quite substantial, considering the likely sources of natural variability such as the developmental stage (animals vs eggs). We further provide several independent lines of evidence for enrichment at TEs and tandem repeats, collectively supporting our claims. Concentration of modified bases over transcriptionally active TEs is readily visible in the Circos plots of Supplementary Fig. 6c,d. Furthermore, manual inspection of 36 unannotated high-density 4mC regions (originally described in Supplementary Note 3 and shown in Supplementary Fig. 6f and Supplementary Table 12) revealed that at least one-half corresponds to unrecognized TEs, further strengthening our claims. We have now inserted this information from the supplement into the main text.
+
+Including the 6mA data is also quite problematic. The presence of 6mA is highly contested in animal genomes, and the data presented here is not robust. 6mA could be RNA contaminations (as found in mammalian species). The claims regarding this section do not offer much novelty and remain contentious.
+
+We are aware of the controversy surrounding 6mA, especially in mammals, and have exercised caution in interpreting 6mA data, most of which was delegated to the supplement. Notably, the most represented motifs at 6mA addition sites (Fig. 2d) do not match the RRACH motif characteristic for RNA MTases, which is one of the criteria used by Douvlataniotis et al. (2020). Another line of evidence against RNA contaminations as the main source of 6mA, which we also present in the supplement, is the occurrence of the 6mA double peak in the non-transcribed region upstream of the TSS in a set of gene orthologs from two species, which is visible not only in SMRT-seq, but even in the low-resolution DIP-seq data from Av-ref and A. vaga. We suggested METTL4 as the most likely 6mA-adding enzyme, due to its N-terminal domain which has the potential to interacting with DNA and to the presence of a nuclear localization signal, although its activity still needs to be tested biochemically. We cannot rule out that some 6mA sites closely linked to 4mC in SMRT-seq data might be caused by trickle-down of IPD signal from 4mC (similarly to 5mC), although this is unlikely to affect antibody specificity in DIP-seq, and a substantial fraction of 6mA sites is distanced from 4mC (as is visible in circos plots Fig. 4h and Supplementary Fig. 6). Considering multiple publications on 6mA, especially in non-mammals (examples in Supplementary Table 1, to which we added a reference emphasizing low reliability of mammalian data), we believe that, despite its controversial nature which is acknowledged, 6mA may still be of biological relevance in rotifers, which borrowed multiple genomic features from fungi. As our MS is not focused on 6mA, keeping the 6mA data as a supplemental addition to our principal findings, without placing much emphasis on it in the main text, should make it available to other researchers for independent evaluation.
+
+Would be good to show the intron-exon structure of the gene with RNA-seq data, and presenting some extra quality checks ensuring these genes are not found in bacterial scaffolds and are indeed encoded as host genes.
+
+We now present the intron-exon structure in a new Supplementary Fig. 1, which also shows RNA-seq coverage for two allelic variants (a) and the syntheny of genomic environments in the genus Adineta (b). An additional Supplementary Fig. 2 shows intron positions in the amino acid sequence alignment of the N4CMT proteins from ten bdelloid species, as well as the C-terminal fusion to the chromodomain and distinct N- and C-terminal extensions clearly distinguishing these proteins from shorter bacterial counterparts. These four lines of evidence (two alleles, introns, preservation of syntenic environment on long contigs over evolutionary time scales, fusion to a eukaryotic domain) combine to yield irrefutable evidence of N4CMT presence in metazoan hosts. A bacterial N4NG-MTase-containing contig (OESY010524654) was bioinformatically identified as a contaminant in Rotaria magnacalcarata assembly and is mentioned in Methods as an example.
+
+Since both genomes are so close, one way to make the claims more robust would be to show orthologous regions across both species with similar profiles in terms of SMRT-seq base modification and DIP-seq.
+
+The difficulty in showing orthologous regions is the highly variable nature of TE insertion sites, which do not coincide between genomes even if they are closely related. Thus, we chose to focus on treating TE features as a whole or grouped by length, and patterns for representative TE insertions are shown on the circos plots in the supplementary Fig. 6. Ortholog
+analysis was employed for 6mA and is presented in the Supplementary Note 3 and Supplementary Fig. 9 for orthologous gene subsets in both genomes (Av-ref and AvL1) with similar profiles for SMRT-seq, DIP-seq and RNA-seq. The small increase of 6mA initially observed in AvL1 by SMRT-seq near the TSS (Supplementary Fig. 9f, right panel) took us to explore DIP-seq profiles those genes (1212 gene models) and, after blastp search, on their Av-ref homologous genes. Confirming our AvL1 observations, Av-ref homolog gene models not only showed similar DIP-seq peak profiles (Supplementary Fig. 9i), but also significantly higher RNA-seq transcript values than average (Supplementary Fig. 9j).
+
+Figure 1f contains some errors. There is no need to establish “Phylum (Class)”. This is not very useful, plus, the authors do not stick to this classification for many groups. E.g. Fungi are not a Phylum. “Protist” is not a valid category, Amoebozoans are more closely related to Animals and Fungi than to Ciliates. Similarly, Ciliates should be sister group to the plant lineages. “Higher plants” does not have any meaning, there is no “lower” or “higher” plants, please refer to the correct group (Embryophytes?). Early Metazoa -> Non bilaterians. Green algae -> Chlorophyta? Some of the common names are confusing.
+
+The taxonomic issues in Fig. 1f have been corrected. We removed “Phylum (Class)” designation and used more appropriate designations for each group, as suggested. Amoebozoa were moved up with other opisthokonts. Ciliates were designated as SAR instead of protists. Higher plants were replaced with Tracheophyta (vascular plants). Green algae included Chlorophyta and Streptophyta minus Embryophyta, as explained in the legend. Taxon inclusion is necessarily limited by availability of sequenced genomes available for BLAST searches at NCBI.
+
+Figure 2 is a mess. Why there are 2 letter codes? Some of them are in capital, others in lowercase, very confusing.
+
+We apologize for the mishap, which occurred during reformatting. All letters are now in lowercase.
+
+Figure 2a. What does “coverage” represent? Peaks? What is “peak” coverage? Number of peaks that intersect with that window? That is not how these metaprofile plots are usually computed. These take the aligned reads from the DIP-seq (or ChIP-seq) and transform it to a read coverage plot, corrected by coverage (e.g. Counts per Milion), and, if available, input DNA (IgG only). Those are the values then used to compute “coverage” plots such as these.
+
+Please see above for explanation for the initial use of the confusing term “coverage”. Initial DIP-seq plots for peak occupancy have now been replaced with DIP-seq read occupancy plots using deepTools2 (Fig. 2a; Supp. Fig. 5), and “coverage” was replaced with “IP occupancy”, as explained in detail in the legend and Methods.
+
+The authors have used DeepTools in Extended Data Fig 2 c/d, but that figure is rather unclear. First, it shows enrichment in all features for both 4mC/5mC. It also says “Relative Fold Enrichment” (respect to what?). TS should be TSS (Transcriptional Start Site), and TTS should be TTS (Transcriptional Termination Site), these are standards in the field, why change them? DeepTools plots can be modified in Inkscape or Illustrator, that would be highly beneficial, since “Gene distance” is not very intuitive when representing Transposable Elements.
+
+Extended Data Fig 2 c/d showed enrichment mostly on TEs, for both 4mC and 6mA (DIP-seq). We made use of deepTools cluster analysis (where the heatmap matrix is split into clusters using the k-means algorithm) to sort features by DIP-seq coverage and display different profiles contributing to the initial plot. This is how we identify a group of TE annotations (cluster 1 & 2 in Supplementary Fig. 5d, right profiles) with an enrichment of DIP-seq 4mC marks, and no such enrichment for 6mA. To our understanding, deepTools “relative fold enrichment” is calculated as the number of reads overlapping each feature from a given BED/GTF file, after initial normalization by RPGC (reads per genome coverage). We have restored the TSS/TTS designations, which were initially changed to TS/TT to avoid font overlapping with the window size in heatmaps.
+
+The Figure 2a “coverage” metaprofiles do not fit at all with those shown in Extended Data Fig 2a/b. This highlights how unreliable are these DIP-seq techniques, which show very different results in two very closely related rotifers.
+
+As recommended, we switched to deepTools2 in figures, and abandoned other graphical representations of the peak calling outputs, keeping only the number of MACS peaks in the text. DIP-seq data represent one out of several lines of evidence, as we are aware of low reliability of each individual method. Some difference in results from two rotifers is anticipated, because the Av-ref strain has been maintained in the lab for over 30 years and has not experienced the same selection pressures as the AvL1 natural isolate (see above). We achieved further improvement in DIP-seq data by additional clean-up of unannotated TEs, which unavoidably penetrate the outputs of the automated gene annotation pipelines. As a result, we could improve the quality of the figures to look more similar. However, the generally low TE copy number prevents a complete clean-up of unknown TE types, and thus some entries in the gene set may represent yet unidentifiable TEs.
+Figure 2d. I believe the SMRT analysis software provides motifs enriched on methylated bases, why did the authors show MEME-ChIP which clearly does not fit this data?
+
+SMRT analysis software contains the motif analysis tool “MotifMaker” which is a motif finding algorithm displaying similarities and differences with other motif finders such as MEME-ChIP. The reason we did not employ this additional SMRT analysis module was because MotifMaker is described in the PacBio documentation and white paper as “a tool to identify motifs associated with DNA modifications in prokaryotic genomes” (https://github.com/PacificBiosciences/MotifMaker), which discouraged us from using it in a eukaryotic genome. MotifMaker is better suited for finding a single motif, while others like MEME-ChIP are based on aligning motifs and identification of a major variety of motifs. Single motifs could be more representative for prokaryotic genomes, like GATC/CTGCAG in E. coli (Fang et al. 2012, Genome-wide mapping of methylated adenine residues in pathogenic Escherichia coli using single-molecule real-time sequencing, Nat Biotechnol) which are useful for identifying highly specific targets for methylases. In eukaryotes, at least for 6mA, methylation events seem to be less of a “single” motif due to other functional regulations (Fu et al. 2015, ref #73; Wu et al. 2016, ref #84). We followed the steps outlined in Greer et al. 2015 (ref #21) and Liang et al. 2018 (ref #74) for motif identification, by extracting the sequences upstream and downstream from each methylated position, and used MEME-ChIP because it would provide a more representative mixture of motifs, better reflecting the complexity of regulation in eukaryotic genomes.
+
+Figure 2g and h. PacBio counts should not be used, methylation levels are always shown as fractional levels since this is a technique with base resolution.
+
+In Figure 2g and h we intended to represent the distribution of methylated sites (SMRT-seq) across the genome and to make comparisons between the number of observed 4mC/6mA sites in gene bodies as well as in transposons and tandem repeat regions. Similar distribution graphs are shown in Liang et al 2018 (ref #74) in Figure3A/B, which counted the number of 6mA sites (from SMRT-seq) in intergenic region and genes. We were mostly following approaches for 6mA characterization, as we needed to characterize only 6mA and 4mC, but not 5mC.
+
+Figure 2i indicates that Adineta has enriched 4mC and 6mA on transposable elements, when other plots in this same figure indicate that 6mA is not enriched on these elements. This is very confusing. Also, showing “dynamite plots” is bad practice, these values should be shown as a distribution (boxplot?). I still do not know why SMRT counts are used instead of fractional values.
+
+While browsing (Gbrowse or IGV) through the Adineta genome, we noticed the dual enrichment of 4mC and 6mA, which was also notable when looking at SMRT counts per annotation (Fig. 2g). However, we observed that the distribution was slightly different: while 6mA sites are rather randomly distributed over the TE body or upstream/downstream, 4mC shows a more precise distribution towards the 5’ insertion point. This is visible in the plot of Fig. 2c using two different window/bin sizes. We further noted that some TE copies are more prone to carry methylation marks than others. This is why we divided the TE dataset into three categories (full, medium, short) based on the length of a copy compared with the reference TE, i.e. assuming that more recently acquired TEs would have a more intact sequence (full/medium) vs old truncated ones (short), indicating that active TE copies are preferentially targeted. The distribution of the data is strongly skewed towards zero methylation (left-skewed), which is likely due to TE under-annotation (see above) and transcriptional inactivity of many TEs in the Adineta genome. This makes it difficult to represent the data in a boxplot for standard distribution. Our intention was to illustrate the average number of modifications using the graph in Fig 2i, so that we could make comparisons between TE subsets without browsing through every TE annotation or determining transcriptional activity of individual copies, which is not feasible with this type of data. Nevertheless, in a few individual examples which are provided in Supplementary Fig. 6, the over-representation of methyl marks over annotated full-length or nearly full-length TEs is readily visible.
+
+Figure 2c. What is “coverage”? PacBio base modification relies on coverage on positions, but then gives fractional values on any given position (0 to 1 values). In fact, “coverage” is a meaningless feature to determine the level of a base modification in any given position, but the “fraction methylation levels” should be averaged and displayed for a metaprofile like this. Again, I would recommend the authors to clarify what this is and maybe use DeepTools2. Why does it say 2.5 Kb window? The plot shows 3kb. Same for the one below, 500 bp window but there are 600 bp shown in the x axis.
+
+“Coverage” was replaced with “SMRT 4mC/6mA occupancy”. We have edited the text to make it clear. The plots were adjusted to show the correct window size of 2.5 kb (top) and 500 bp (bottom).
+Extended Data Fig 3 is very hard to read, pixelated when zoomed in. There is not need to show these are “Circos” plots, simple Genome Browser representations would be much easier to interpret, zooming out or in depending on the features that need to be highlighted.
+
+We have reorganized this figure, removing non-informative parts from four circles, reducing pixelation, and adding ID’s for each track for easier interpretation. We are using browser snapshots in the new Supp. Fig. 1 to demonstrate intron-exon structure and expression, but Av-ref and AvL1 data poorly fit onto the same browser due to 12% divergence of AvL1 from the reference. Circular plots have an added advantage of combining several genomic regions within the same panel, avoiding the pileup of multiple plots as in Supp. Fig. 1.
+
+The sequences of the conserved motifs shown in Figure 3f do not contain any CG dinucleotide, which in figure 2b was shown to be the preferred substrate of these enzymes according to SMRT-seq data. This is inconsistent and is not discussed.
+
+Our explanation of the conserved motif significance apparently was not clear enough – in vivo methylation sites at CG dinucleotides within the 460-bp repeat were shown with red arrows in Fig. 3g (now Supp. Fig. S7e), however no arrows are marked within the conserved motifs, which led us to hypothesize that these motifs ensure sequence-specific target recognition by the MT moiety independently of the chromodomain, as confirmed by efficient in vitro recognition by the chromodomain-less N4CMT mutant, but do not necessarily coincide with methylated sites in vivo, where chromodomain-based recognition would also play a role. We now make this distinction clearer in the text.
+
+How are the + and – in Figure 3g determined? With an immuno-dot blot? Where is it? Not all combinations seem to be found in h and i panels. H and I panels are hard to interpret and poorly explained in the figure legend.
+
+The presence or absence of 4mC mark after treatment with N4CMT, labeled as “+” or “-”, was determined based on immuno-dot blot experiments with anti-4mC antibody. Each DNA was tested at least two times in independent experiments, and one of the respective dot blots is included in updated Supplementary Fig. 7. In each experiment we included control spots: reaction mix without DNA template and mix with DNA but without N4CMT, to control for background noise and non-specific chemiluminescence signals. We have re-organized Fig. 3 and moved four panels, including h and i, into the Supplementary Fig. 7 as part of the manuscript shortening. The immuno-dot-blots covering all combinations of + and - are now combined in the supplement. Explanations in the figure legends have been improved.
+
+Figure 4. H3K9me3 and H3K27me3 are not found usually on the same sites, one is constitutive chromatin and the other facultative (at least in mammals). This “expectation” is not very realistic. To validate the ChIP-seq, it would be good that the plots in Figure 4a/c would be sorted the same way, e.g. using the one to one orthology relationships between assemblies/species. Also, it would be good to add more information on the figure itself, reading the legend is confusing.
+
+Indeed, for a long time the separation of function between constitutive chromatin marked by H3K9me3 and silencing TEs, and facultative H3K27me3 silencing genes, has been the prevailing paradigm. However, there is a growing body of evidence from less traditional models, such as ciliates and bryophytes (recently reviewed by Délérís et al. 2021, PMID: 34210514), that H3K27me3 can serve as an ancestral TE-silencing mark, with the subdivision into constitutive K9 and facultative K27 occurring later in evolution, in higher plants and mammals. Although we have employed orthology to assess the significance of the genic 6mA marks (supplementary note 3), it would be problematic to employ it for validation of TE marks since TE insertions lack orthology and can be analyzed only as individual insertions, or in some combinations (e.g. full-length vs truncated). The anti-H3K9me3 and anti-H3K27me3 antibodies have been validated for the lack of cross-reactivity, as explained in Methods.
+
+Figure 4e/f. This plot is suspicious, the shape of the “peak” indicates that there is unclear smoothing within that window. Usually peak metaprofiles are centred around the peak summit (or centre), without setting hard borders as if they were genes, which tends to generate artefactual shapes as those shown here. Also, figure legends are barely visible.
+
+Figures 4e/f have been corrected by centering on the peak summit. The font size in the legends has been increased.
+
+Figure 4h. Almost impossible to read, too pixelated. Better to show linear genome browser snapshots highlighting some associations instead of dumping too much data in a circular plot that is difficult to follow.
+
+We removed the least informative regions from the circular plot, which helped to improve resolution and reduce pixelation. As explained above, genome browser snapshots are difficult to use with divergent reference genomes and occupy a lot of
+space, as may be seen from the new Supplementary Fig. S1, while the circular plots accommodate several contigs showing associations between DIP-seq, SMRT-seq, H3Kme, RNA-seq, and small RNA layers.
+
+Figure 4g. This should resemble figure 4f using a metaprofile displaying the fractional level of methylation (4mC/6mA) on those peaks, and not a “dynamite plot” that is difficult to interpret.
+
+Unlike Fig. 2c, which shows an enrichment of 4mC towards the 5’ boundaries of TEs, the H3K9me3 and H3K27me3 histone marks do not show a specific profile with regard to DNA methylation. As visualized in Fig. 4h and Supplementary Fig. 6, we normally see several peaks (both H3K9 and H3K27) on a contig region which shows 4mC and 6mA DNA marks distributed quite broadly. The H3K4 mark, as expected, is commonly found in gene-rich regions. We opted to represent in this plot the average number of modifications for each histone mark, as a proxy to better visualize global patterns across the genome. A large number of annotations with zero counts presents a skewed distribution and is difficult to eliminate.
+
+The section on the small RNAs feels confusing and unnecessary for this manuscript.
+
+We have moved this section to the Supplementary Note 4. It would be undesirable to delete it altogether, as it strengthens the link to TE transcriptional activity, which serves as a prerequisite for small RNA production (as is well known from the literature), as well as for 4mC deposition (as may be inferred from Fig. 2i).
+
+The section about SETDB1 as a possible reader of 4mC is also very weak and unnecessary. Not all animal MBD proteins bind to methylated cytosines, that is well established. MBD1/2/3/4 and MeCP2 are the families that have been shown to recognise methylated CpGs in other animals. In some cases, like Drosophila melanogaster, it harbours a highly divergent MBD2/3, that has been shown not to bind to preferentially to methylated CpGs, which is expected since Drosophila lacks 5mC methylation. The competition experiments shown in Figure 6 are not very strong indication of this domain preferentially binding to 4mC methylation.
+
+We fully agree that MBD is a generic DNA-binding domain, and this was clearly stated in the main text, with a full overview of MBD proteins presented in supplementary Fig. 11. While the difference in binding affinity of the Av314 variant is not drastic, it is nevertheless significant, and we provide the source data file containing the results from replicate experiments. Other observations in support of Av314 role are (i) the lack of detectable differences in DNA binding for the other five paralogs, and (ii) identification of Av314 as the only variant for which the loss is associated with increase in vertically-transmitted LINE-like TEs (Fig. 6b,c). Collectively, these lines of evidence fit well with the “reader” concept, as the highly unusual amplification observed for SETDB1 could provide the raw evolutionary material for developing 4mC preference and serve as the missing link to H3K9me3 as implied by our initial finding of the chromodomain in N4CMT. Although this section could be moved to the supplement, we feel that there is no overstatement in keeping it in the main text with full explanation of the caveats, however removing it altogether would leave a void.
+
+Figure 6f. Given the data shown in this manuscript, it feels premature to draw a model on how this system works in tardigrades.
+
+We explicitly say that the model in rotifers is hypothetical, nevertheless it is fully consistent with the data presented in the manuscript and assists the reader in placing our findings into the plausible chromatin context, in agreement with known functional properties of the CMT and KMT enzymes involved. These are the only connections that were drawn in non-dashed lines in the model. We also left out other known processes that may be involved but are less relevant to the present work, such as deacetylation, ubiquitination, sumoylation etc.
+
+Claims such as “Finally, it demonstrates that horizontally transferred genes, contrary to the established view”. Lateral Gene Transfer is contested as a major source of innovation in eukaryotic lineages by some authors, but this is far from being “the established view”.
+
+We changed the wording in this sentence: “Finally, it demonstrates that horizontal gene transfer, the role of which in eukaryotic regulatory evolution is a subject of intense debate...”. The emphasis here is on regulatory evolution, as opposed to the much more widespread HGT of “operational” genes controlling specific metabolic reactions.
+
+Plots such as those in Extended Data Fig 5 e,f,g,i are basically noise. This is likely due to the non-standard ways of representing this data, and no information can be extracted from these.
+
+We agree that most of the genic 6mA marks in this figure are noise, except for two peaks visible upstream of the TSS, which are the only meaningful locations, as they appear independently in two orthologous gene subsets from Av-ref and AvL1. The relevant information that can be extracted from these peaks is their localization in non-transcribed regions, which argues
+against RNA origin of the signal and justifies keeping it as supplementary material. The Y scale in (e) has been adjusted to that in (f), emphasizing the noisy character of the genic marks which are identifiable only in a subset at the peak location.
+
+Reviewer #4 (Remarks to the Author):
+
+Comments to the author
+Rodriguez et al. performed an impressive research on 4mC occurrence in eukaryotic DNA. The major result underlines the 4mC presence in bdelloid rotifers by combining multiple trials I believe the results reported would be interesting to wide spectra of biologist and other specialists.
+
+We thank the reviewer for the positive evaluation of our work.
+
+Points raised:
+
+1. In figure 1a,1b, is it more intuitive to mark the direction with the N-terminal or C-terminal instead of using numbers to mark the direction? In addition, using cylinders to roughly represent multiple enzymes seems insufficient. Is there a way to show the structural differences more clearly? I’m not whetted on it, but think the article would be improved, if the authors consider this and take some actions.
+
+We have added Supplementary Fig. 2 to show the alignment of full-length ORFs indicating the length of each protein, functional domains and motifs, catalytic residues, secondary structure elements, and the degree of conservation. Since each species shows the same structural organization, the approximate schematic representation in Fig. 1a appears adequate, as it reflects only the basic domain architecture and the cylindrical MTase domains are evolutionarily related. Using N and C to mark the direction does not seem more informative than indicating length.
+
+2. The author mentioned that for gene profiles, the modification density is much lower, and it seems the opposite compared to TE profile. Can the authors explain the opposite distribution phenomenon of gene profile and TE profile in a more understandable way?
+
+We have redrawn the plots for gene profiles and TE profiles for both Av-ref and AvL1 (Fig. 2a; Supplementary Fig. 5a), and the increased density of modified bases over TEs is visible in both, apparently due to preferential targeting of N4CMT to silent H3K9/27me3-marked chromatin via its chromodomain moiety. Modifications in genic regions are much less pronounced and do not show robust association using statistical correlations (Supplementary Table 4). We have explained this in the text and introduced an additional Supplementary Note 1 and Supplementary Fig. 4 to clarify the details of statistical analysis.
+
+3. As far as I know, coverage level is a major effect factor on correction efficiency. In the range of 10x to 50x, the correction result increases as the coverage increases (https://www.uni-wuerzburg.de/fileadmin/07030400/AG_Genomics/Proovread/proovread-preprint.pdf). Therefore, does the author think that after increasing the coverage level, the overlap between SMRT-seq and DIP-seq is higher?
+
+Indeed, as we document in Supplementary Table 6, the number of modified bases increases with coverage, and further increase of PacBio SMRT-seq coverage could shed more light onto regions with undetected methylated sites having lower SMRT coverage profiles, e.g. towards telomeres. Nevertheless, most discrepancies in the overlap between the two techniques may be attributed to DIP-seq, where the methodology is highly prone to sensitivity issues, notwithstanding the difference in developmental stage explained above. However, given the modest percentage of modified bases, with 0.0643% of the total cytosines in the assembly (21,016 4mC modifications) and 0.0236% of total adenines (17,886 6mA modifications) as defined by SMRT-seq, the overlap between DIP-seq and SMRT-seq is quite substantial, with 36% of 4mC DIP-seq peaks and 32% of 6mA peaks overlapping with 4mC and 6mA identified by SMRT analysis, respectively.
+
+4. In line 247, the author mentioned "At 4mC sites, CpG and CpA dinucleotides are the most prevalent, making up 74% of modified doublets." Please explain the calculation process.
+
+We have updated the methods section with an explanation. Briefly, the upstream and downstream 10-bp sequences from 4mC and 6mA modification sites were extracted for motif identification, and the adjacent nucleotide base of the methylated sites was pulled out for counting the proportion of doublets. This function is provided in the SMRT portal.
+
+5. In line 259, the author reports the different methylation levels of 4mC and 6mA. Whether the conclusion drawn is related to the state of the sample. In other words, the methylation levels show differences at different time nodes.
+We consider the differences in methylation fraction to be mostly due to inherent characteristics of each modification, with 4mC being more reproducible and appearing in a much higher fraction of sites, and 6mA more dynamic and subject to developmental stage- and tissue-specific differences. Our current goal was to discern the most stable modification patterns which would be largely preserved across tissues and developmental stages, and while there may be some interesting developmental dynamics of methylation patterns, we considered such studies to be outside the scope of the present work.
+
+6. Please use the definite article accurately and polish other details that I haven't noticed. For example, “the genome-wide”, “We visualized the distribution”, etc.
+
+Corrected.
+
+7. The ordinate of Fig. 2B is not clear, and the picture resolution needs to be improved.
+
+The Y axis has been redrawn and the resolution improved.
+
+8. It may be due to the typesetting that makes Figure 2 look confusing. Please adjust the position and label of the picture to make it easier for readers to read. In addition, m4C, m6A in the 2F should be modified to 4mC, 6mA, please keep the abbreviations consistent.
+
+The figures have been corrected and the labeling has been improved. Abbreviations have been made consistent.
+
+9. Some typoses should be revised.
+
+Corrected.
+REVIEWER COMMENTS
+
+Reviewer #2 (Remarks to the Author):
+
+Thank you for making changes to address my comments.
+
+Reviewer #3 (Remarks to the Author):
+
+Despite the authors have done a great deal to improve the manuscript, several of my complains have not yet been addressed.
+
+First, the Marchantia work is simply dismissed. Adding it as reference 87 of the Supplementary Material does not work. I don’t know why the authors do this, since the Marchantia results increase the value of their own findings and puts it into context way better than trying to ignore it, maybe just to be “the first” to report 4mC in an eukaryote. I really don’t think it is important who posted this first, both works are now available to read by anybody, so ignoring the preprint makes no sense. Also, when it is a case of 4mC usage in a completely different branch of eukaryotes. If the reports of Adineta 4mC are robust, it would be important enough, the first case in animals, with a LGT methyltransferase, etc. In fact, the title could well be “Bacterial N4-methylcytosine as an epigenetic mark in metazoan DNA” or “rotifers”, which is more precise and impactful enough.
+
+Then, the analysis of the data.
+1) The authors seem to have taken just the DIP-seq read coverage with DeepTools and plotted it without correcting it with the IgG control. If using DeepTools, it would be important to use bamCompare (e.g. bamCompare -b1 DIPseq.bam -b2 IgGcontrol.bam -o log2ratio.bw) to create the bigwig file and plotting those corrected values instead of raw coverage. The same way MACS2 uses the IgG control as background to know what a real peak is, as readers we would like to see this data corrected for noise. For instance, the plots showing exactly the same profiles of 6mA and 4mC in genes for AvL1 (Supp Fig S5) could well be that the sequence bias is driving these patterns irrespectively of true signal. Same thing goes for Figure 4e/f.
+2) In terms of showing the fractional data in Circos plots. Those are very difficult to read. For instance, Figure 4h. There are two tracks, called “Illumina 1” and “Illumina 2” (what is that anyway?), but then the 4mC and 6mA values appear on those tracks... I guess this is the result of wanting to put both assemblies on the same figure, but it is very hard to read/interpret, moreover when different assemblies have different data types. Also, none of those plots include a y-axis. What are those fractions? 0 to 1? 0 to 0.02? The authors say that it is usually 100%, but when I ask for a IGV plot is to clearly see the raw data, and this is not so easily read in the Circos plots. After all, their preferred representation coming from Liang et al 2020 also shows a IGV plot in Figure 1f...
+3) A clear way to show the overlap between 4mC and the DIP-seq would be to 1- call peaks using the DIP-seq. 2- generate a bigwig file of 4mC (e.g. scaffold position methylation_fraction), maybe just including the CG sites to increase the signal. 3- show the heatmap on those peaks using 4mC DIP-seq (corrected) values (see point 1) and 4mC fractional values together. It is easy in DeepTools to specify different colours / scales for various marks, so this is quite easy to get. That way we would see the co-localization of both techniques on the same plot genome wide, instead of just showing some vague global overlaps. This could be also done for the 6mA, and in fact, in a combination (e.g. overlap DIP-seq peaks).
+4) The 4mC and 6mA fractional bigwig files should also be used to present metaProfiles as those shown in Figure 2a. The fractional values should also show an average increase of 4mC marks on TEs, and maybe confirm this very subtle enrichment in the TSS.
+5) The authors say that there are not many TEs in Adineta. The heatmaps seem quite data rich. So how many are shown in each plot? That’s important to interpret those maps, and to draw potential differences between both strains.
+Minor remarks. Abstract – DNA modification is used -> “DNA modifications are used” feels more natural. Then, the second sentence with an “it” is confusing, what is the “it” referring to?
+
+This sentence: “N4CMT adds 4mC to DNA, and its chromodomain shapes the ”histone-read-DNA-write” architecture, together with a ”DNA-read-histone-write” SETDB1 H3K9me3 histone methyltransferase variant preferentially binding 4mC-DNA, to maintain 4mC and silent chromatin at active transposons and tandem repeats.” Is very long and hard to follow. Furthermore, the connection of SETDB1 with 4mC is very premature to be highlighted like this in the abstract. This would be, at most, a suggestion with the current evidence.
+
+“Our results bring the third base modification into the eukaryotic repertoire” this is simply wrong, there are many other base modifications found in eukaryotes. E.g. J base in Trypanosomatids, hydroxymethylcytosine (and other derivatives), 5-Hydroxymethyluracil in dinoflagellates, or this one (https://pubmed.ncbi.nlm.nih.gov/31043749/).
+
+“epigenetic systems to suppress transposon proliferation”. Well, one thing is that it is marking TEs, another one is that is suppressing their proliferation, which this work does not demonstrate.
+
+If 4mC is marking TEs, then it does not have anything to do with “regulatory networks”. Silencing a TE is not a “regulatory” network, but a silencing mechanism, a definition of “regulatory network” according to the Nature publishing group can be found here: https://www.nature.com/subjects/regulatory-networks#:~:text=Definition,genes in a given genome.
+
+The authors should define in the introduction what do they mean by “epigenetic”. There’s a lot of controversy in the field, therefore a clear definition of what do they mean is mandatory (since they find this a unique feature of this system).
+
+Reviewer #4 (Remarks to the Author):
+
+Authors have improved the paper. I think the paper can be accepted.
+RESPONSE TO REVIEWER COMMENTS
+
+Reviewer #2 (Remarks to the Author):
+
+Thank you for making changes to address my comments.
+
+We thank the reviewer for re-assessing the revised version.
+
+Reviewer #3 (Remarks to the Author):
+
+Despite the authors have done a great deal to improve the manuscript, several of my complains have not yet been addressed.
+
+As detailed below, the current revision addresses each of the remaining comments, which were very helpful for improving data analysis and presentation.
+
+First, the Marchantia work is simply dismissed. Adding it as reference 87 of the Supplementary Material does not work. I don’t know why the authors do this, since the Marchantia results increase the value of their own findings and puts it into context way better than trying to ignore it, maybe just to be “the first” to report 4mC in an eukaryote. I really don’t think it is important who posted this first, both works are now available to read by anybody, so ignoring the preprint makes no sense. Also, when it is a case of 4mC usage in a completely different branch of eukaryotes. If the reports of Adineta 4mC are robust, it would be important enough, the first case in animals, with a LGT methyltransferase, etc. In fact, the title could well be “Bacterial N4-methylcytosine as an epigenetic mark in metazoan DNA” or “rotifers”, which is more precise and impactful enough.
+
+The reference was added in the most appropriate place in the Discussion, which happened to be part of the Supplementary Discussion. We have now moved the corresponding part of the Discussion into the main text (p. 17), with the accompanying reference. It would be ideal if the two papers could be published in parallel to increase the value of each other, however this would be beyond our control.
+
+Then, the analysis of the data.
+1) The authors seem to have taken just the DIP-seq read coverage with DeepTools and plotted it without correcting it with the IgG control. If using DeepTools, it would be important to use bamCompare (e.g. bamCompare -b1 DIPseq.bam -b2 IgGcontrol.bam -o log2ratio.bw) to create the bigwig file and plotting those corrected values instead of raw coverage. The same way MACS2 uses the IgG control as background to know what a real peak is, as readers we would like to see this data corrected for noise. For instance, the plots showing exactly the same profiles of 6mA and 4mC in genes for AvL1 (Supp Fig S5) could well be that the sequence bias is driving these patterns irrespectively of true signal. Same thing goes for Figure 4e/f.
+
+We thank the reviewer for valuable advice and detailed instructions. For peak calling in ChIP-seq data with MACS2, we used input DNA, rather than IgG, as chromatin control in both strains for background correction. Following reviewer’s suggestions, we added the correction step before plotting with deepTools, using bamCompare with both the treatment and the input control, to obtain the log2 ratio between the two. Figure 4e/f has now been updated using these corrected profiles, which satisfyingly led to elimination of the artifactual peak previously appearing in the H3K4 plot. Accordingly, we have removed descriptions of this artifactual peak from the text and legend to Figure 4. In 4mC and 6mA DIP-seq experiments, we used MACS v. 1.4.2, rather than MACS2, for peak calling in DIP-seq data, since it is considered effective in capturing the local genomic sequence biases from a ChIP-seq sample alone, in lieu of the control sample (Zhang et al. 2008). Clustering in Supp. Fig. 5c-d then helps to reveal the minor fraction of the data that is likely to represent noise, displaying no difference between genes and TEs.
+
+2) In terms of showing the fractional data in Circos plots. Those are very difficult to read. For instance, Figure 4h. There are two tracks, called “Illumina 1” and “Illumina 2” (what is that anyway?), but then the 4mC and 6mA values appear on those tracks… I guess this is the result of wanting to put both assemblies on the same figure, but it is very hard to read/interpret, moreover when different assemblies have different data types. Also, none of those plots include a y-axis. What are those fractions? 0 to 1? 0 to 0.02? The authors
+say that it is usually 100%, but when I ask for a IGV plot is to clearly see the raw data, and this is not so easily read in the Circos plots. After all, their preferred representation coming from Liang et al 2020 also shows a IGV plot in Figure 1f…
+
+We thank the reviewer for pointing the items which may be difficult to read. We have updated Figure 4h and the legend, and have renamed the confusing “Illumina 1” and “Illumina 2” tracks (representing two different DNA Illumina libraries for Av-ref) to “Illumina 862 bp” and “Illumina 450 bp”, in agreement with the actual insert sizes. The two assemblies (strains) were included not only to illustrate the same correlations with methylation for different datasets, but to show similarities/differences of data types (tracks in circos plot) between the two strains. To improve clarity, the strain name (Av-ref or AvL1) is now highlighted at the contigs base in the circos plot. The y-axis scale was added for the PacBio SMRT-seq methylation fraction values (from 0 to 1). Overall, we opted for displaying selected genome sections of the assemblies with circos because, among other things, it can represent large portions of the genome in a single sub-figure within a manuscript, in this case, Figure 4h which represents ~485 Kb of genome data. Although IGV representation and functionality is very useful for routine visualization, depiction of a similarly sized genome fraction as in Figure 4h using multiple tracks for different contigs would require much more space in the main figure. We therefore chose to present selected genomic regions as IGV plots in the supplement (Supplementary Fig. 8).
+
+3) A clear way to show the overlap between 4mC and the DIP-seq would be to 1- call peaks using the DIP-seq. 2- generate a bigwig file of 4mC (e.g. scaffold position methylation_fraction), maybe just including the CG sites to increase the signal. 3- show the heatmap on those peaks using 4mC DIP-seq (corrected) values (see point 1) and 4mC fractional values together. It is easy in DeepTools to specify different colours / scales for various marks, so this is quite easy to get. That way we would see the co-localization of both techniques on the same plot genome wide, instead of just showing some vague global overlaps. This could be also done for the 6mA, and in fact, in a combination (e.g. overlap DIP-seq peaks).
+
+Initially, we did not use the bigwig (.bw) format for PacBio SMRT-seq methylation, since it is meant for dense and continuous datasets. Following the reviewer’s suggestion, we explored the overlap between DIP-seq peaks and SMRT-seq methylation by using the bedgraph—>bigwig conversion format, with fractional value attached to each methylation position. We used deepTools for plotting DIP-seq peak profiles, centered around the peak summit, for each methylation mark. Indeed, we can clearly see the co-localization of both techniques for 4mC and 6mA (see the added Supplementary Fig. 6a), where regions harboring a called peak from DIP-seq have an increase of methylated sites/fractional values called after SMRT-seq analysis. As requested, we also plotted the corresponding profile using only CpG sites within SMRT-seq 4mC methyl sites (Supplementary Fig. 6a, middle), which similarly shows an increase towards 4mC peaks.
+
+4) The 4mC and 6mA fractional bigwig files should also be used to present metaProfiles as those shown in Figure 2a. The fractional values should also show an average increase of 4mC marks on TEs, and maybe confirm this very subtle enrichment in the TSS.
+
+The 4mC and 6mA fractional bigwig files generated from SMRT-seq data (from point #4) were taken to represent their values across annotated TEs to confirm this enrichment. As was previously done in Figure 2c, we plotted TE metaprofiles with deepTools using different upstream/downstream regions and bin sizes. Once again, the TE insertion point (5’ end) shows an overlying increase of 4mC marks, which have now been factored by fraction values (see the added Supplementary Fig. 6b).
+
+5) The authors say that there are not many TEs in Adineta. The heatmaps seem quite data rich. So how many are shown in each plot? That’s important to interpret those maps, and to draw potential differences between both strains.
+
+We agree that the heatmaps should display information on the data density, so that the highly dense datasets (genes) could be compared with less abundant annotations (in this case, TEs). While we chose to keep the same heatmap height, we have now added the total number of annotations (genes, TEs) on the side of each heatmap in Fig. 4a-d, to provide a context for comparison and interpretation.
+
+Minor remarks. Abstract – DNA modification is used -> “DNA modifications are used” feels more natural. Then, the second sentence with an “it” is confusing, what is the ���it” referring to?
+Both sentences were corrected as requested: “it” was replaced with “modifications”.
+
+This sentence: “N4CMT adds 4mC to DNA, and its chromodomain shapes the “histone-read-DNA-write” architecture, together with a “DNA-read-histone-write” SETDB1 H3K9me3 histone methyltransferase variant preferentially binding 4mC-DNA, to maintain 4mC and silent chromatin at active transposons and tandem repeats.” Is very long and hard to follow. Furthermore, the connection of SETDB1 with 4mC is very premature to be highlighted like this in the abstract. This would be, at most, a suggestion with the current evidence.
+
+This long sentence was split in two, with the second sentence reflecting the suggestive nature of SETDB1 findings (lines 24-28).
+
+“Our results bring the third base modification into the eukaryotic repertoire” this is simply wrong, there are many other base modifications found in eukaryotes. E.g. J base in Trypanosomatids, hydroxymethylcytosine (and other derivates), 5-Hydroxymethyluracil in dinoflagellates, or this one (https://pubmed.ncbi.nlm.nih.gov/31043749/).
+
+Indeed, our intention was to focus on bacterial modifications rather than the numerous existing eukaryotic derivatives. We have replaced “the third base modification” with “the third bacterial modification”.
+
+“epigenetic systems to suppress transposon proliferation”. Well, one thing is that it is marking TEs, another one is that is suppressing their proliferation, which this work does not demonstrate.
+
+Replaced with “epigenetic systems to silence transposons”. Our work emphasizes strong correlations with transposon proliferation based on comparative analysis of multiple rotifer species, especially the TE-rich bdelloid D. carousus without N4CMT which lost the 4mC-prefering SETDB1 variant, as shown in Figure 6. However, observing the actual reduction in genomic copy numbers over the evolutionary time scales would be a long-term process, which is shaped by several opposing forces and extends far beyond what is possible to achieve in the lab during an experimental study in metazoan species.
+
+If 4mC is marking TEs, then it does not have anything to do with “regulatory networks”. Silencing a TE is not a “regulatory” network, but a silencing mechanism, a definition of “regulatory network” according to the Nature publishing group can be found here: https://www.nature.com/subjects/regulatory-networks#:~:text=Definition,genes in a given genome.
+
+Although the involvement of transcription factors in regulation of expression, as required by the above definition, is strongly suggested by the concentration of epigenetic marks near the TSS (as mentioned in the discussion), it does not constitute the principal focus of the current study. We have therefore replaced “gene regulatory networks” with “gene silencing systems”, although this concept is broader and includes not only TGS, but also PTGS and nuclear organization (https://www.nature.com/subjects/gene-silencing).
+
+The authors should define in the introduction what do they mean by “epigenetic”. There’s a lot of controversy in the field, therefore a clear definition of what do they mean is mandatory (since they find this a unique feature of this system).
+
+This is a very useful suggestion, given the possible contradictory definitions. For consistency with the previous item, for the purpose of this study we keep following the definitions on the same website (https://www.nature.com/subjects/epigenetics), and explain in the Introduction (lines 57-59): “We focus our attention on epigenetic silencing phenomena that involve DNA and histone modifications, without expanding into broader areas involving nuclear organization or post-transcriptional silencing.”
+
+Reviewer #4 (Remarks to the Author):
+
+Authors have improved the paper. I think the paper can be accepted.
+
+We are pleased that the reviewer finds our responses satisfactory.
+REVIEWERS’ COMMENTS
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have addressed most of my points, here are my minor remaining points:
+
+The authors mention that they have used “input DNA” to correct DIP-seq tracks. That’s fine, but is the new Figure 2a showing the background corrected version? “IP occupancy” is a bit cryptic and I cannot guess from the Figure legend. In Figure 4e/f it says “log2 ratio” which is very evident, so it is a bit strange that two plots showing the same data type have different y-axis labels/measures.
+
+Supplementary Figure 8 is really useful, I thank the authors for this. The first example is really clear, with a very nice cluster of 4mC and 6mA sites. The two other examples are a bit more sparse on 4mC, were they chosen for a particular reason? Since there are not expectations on how 4mC might look like in any eukaryotic genome, I guess this is fine, but for the readers it would be nice to know the rationale (first example being the clearest in the genome and the other two being the 2nd and 3rd? or chosen for other reasons?).
+
+The bigwig tracks made for 4mC and 6mA would be important to upload to a public repository (e.g. the GEO? Or Figshare/github/etc), since the analysis of PacBio is not straightforward and these data might be useful for other researchers interested in the findings described in this manuscript.
+RESPONSE TO REVIEWERS' COMMENTS
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have addressed most of my points, here are my minor remaining points:
+
+The authors mention that they have used “input DNA” to correct DIP-seq tracks. That’s fine, but is the new Figure 2a showing the background corrected version? “IP occupancy” is a bit cryptic and I cannot guess from the Figure legend. In Figure 4e/f it says “log2 ratio” which is very evident, so it is a bit strange that two plots showing the same data type have different y-axis labels/measures.
+
+For ChIP-seq data, we used input DNA as chromatin control in both strains for background correction using MACS2. Using bamCompare with both the treatment and the input control, we obtained the log2 ratio between the two (Figure 4e/f). In 4mC and 6mA DIP-seq experiments, we used MACS v. 1.4.2, rather than MACS2, for peak calling in DIP-seq data, since it is considered effective in capturing the local genomic sequence biases from a ChIP-seq sample alone, in lieu of the control sample (Zhang et al. 2008). Thus, Figure 2a is showing IP occupancy, as there is no input to derive a log2 ratio from DIP-seq data.
+
+Supplementary Figure 8 is really useful, I thank the authors for this. The first example is really clear, with a very nice cluster of 4mC and 6mA sites. The two other examples are a bit more sparse on 4mC, were they chosen for a particular reason? Since there are not expectations on how 4mC might look like in any eukaryotic genome, I guess this is fine, but for the readers it would be nice to know the rationale (first example being the clearest in the genome and the other two being the 2nd and 3rd? or chosen for other reasons?).
+
+Contig As785 is by far one of the highest in methylation density which is observed; it covers a tandem repeat (shown in Supplementary Fig. 7a) located between Athena elements. Other contigs with 4mC and 6mA clusters normally show lower density of methylated sites, as illustrated in two other examples showing DNA TEs, which were not presented in earlier circos plots. Contig 1073 represents an under-annotated transposon region (Supplementary Table 11), which contains a potentially active, polymorphic insertion of Chapaev DNA TE present only in PacBio, but not in Illumina data. Contig1534 is another example of association between 4mC methylation marks and TE insertion (while the 4mC marks are indeed more sparse, the 6mA marks are entirely lacking). This information has been included into the figure legend.
+
+The bigwig tracks made for 4mC and 6mA would be important to upload to a public repository (e.g. the GEO? Or Figshare/github/etc), since the analysis of PacBio is not straightforward and these data might be useful for other researchers interested in the findings described in this manuscript.
+
+BigWig (BW) tracks for 4mC and 6mA have been added to the GEO dataset GSE140050 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE140050], which has been made publicly available.

035e55afc9c40b2a32b10bb61cbaf9c417c4c43287f20e12b4733b13052ac290/peer_review/peer_review.md

@@ -0,0 +1,559 @@
+Peer Review File
+
+Probing charge redistribution at the interface of self-assembled cyclo-P5 pentamers on Ag(111)
+
+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/.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+The authors present here a combined experimental and theoretical study of cyclo-P5 pentamers on a Ag(111) surface. Using non-contact AFM with a functionalized tip and DFT calculations, they characterize the atomic structure of cyclo-P5 pentamers assembled on the Ag(111) surface and reveal that a strong charge transfer occurs from the silver surface to the phosphorous structure. Consequently, they deduce the change in the surface work function.
+In the first part of the manuscript, they have characterized and modelled phosphorous chains on the Ag(111) surface with the nc-AFM, and recover the same results as obtained in Ref. [10]. The DFT simulations neither provide anything new, as the model is the same as in Ref. [10]. Therefore, I do not see the interest of this part of the manuscript. In particular, the “excellent agreement” between experiment and theory here was already the topic of the work in Ref. [10].
+In the second part of the manuscript, the authors focus on the synthesis of cyclo-P5 pentamers on the Ag(111) surface, by repeating the procedure described in Ref. [11]. They obtain the same structure and simulate the images according to the model developed in Ref. [12], yielding a good agreement.
+Then, and this is probably the most innovative part of the manuscript, they measure the surface electrostatic potential on the cyclo-P5 structure and on the bare silver surface, leading to a potential difference representative of an interface dipole and a strong charge transfer from silver to phosphorous. These features are supported by DFT calculations, enabling to determine the value of the surface dipole as well as the corresponding charge transfer. Furthermore, they have also characterized the Image Potential States at the interface.
+To my opinion, this work is a very good re-characterization of the already known cyclo-P5 system (and its precursor’s chains) with nc-AFM with functionalized tips images instead of STM images, and some additional spectroscopic measurements to characterize the interface states and charge transfer. DFT calculations have already been published together with experimental results for the P-chains (Ref. [10]) and in a single article for the pentamers (Ref. [12]), published by some co-authors of the present paper, and as such, do not bring
+anything new to the present manuscript.
+
+To summarize, while the experimental characterization of this known system is well executed, it does not bring any real innovative feature on this system, and therefore, it does not meet the criteria for publication in Nature Communications. Hence, this work should be published, but in a more standard journal, as an educated repetition of previous works.
+
+Reviewer #2 (Remarks to the Author):
+The article titled 'Probing charge redistribution at the interface of self-assembled cyclo-P5 pentamers on Ag(111),' authored by Outhmane Chahib, Yulin Yin, Jung-Ching Liu, Chao Li, Thilo Glatzel, Feng Ding, Qinghong Yuan, Ernst Meyer, Rémy Pawlak, describes the synthesis and detailed characterization of the cyclical P5 phosphorus allotrope on the Ag(111) surface. Utilizing advanced scanning probe microscopy, the team presents high-resolution nc-AFM images, offering unparalleled structural insights for P5. Although the synthesis technique is previously documented (in Ref. 11), the novel contribution lies in the detailed examination of P5 molecule interactions with the Ag(111) surface. While this topic might appeal more to specialized circles, the depth of technical discussion presented may surpass the general scope preferred by the broad readership of Nature Communications. This nuanced exploration, though valuable for experts, could potentially challenge the wider audience's engagement due to its specialized technical complexity. The manuscript is well structured and the data in most cases supports well the claimed results.
+
+Furthermore, I must emphasize my deep-seated concerns on several major issues:
+1. In the sentence: “Thus, we conclude that the cyclo-P5 does not have a pure anionic character for the P5 molecule when adsorbed on Ag(111) (i.e. cyclo-P–5 ), as expected by theory for its gas-phase counterpart.” the authors compare P5 molecules adsorbed on a surface with gas phase molecules, which is very confusing. How could the isolated P5 molecule have an anionic character in the gas phase? This needs clarification.
+2. The authors show dl/dV curve for P5 on Fig. 4a and discuss it in comparison with analogous measurement from Ref. 11. However, these curves have significant differences. For example, in the region of +2 V the curve drops down in Fig. 4a but it rises in Ref. 11. Additionally, the measured bandgaps are also different: 0.9 eV vs. 1.2 eV. This needs further discussion and the differences must be explained.
+Additionally, I have a few minor points to discuss:
+
+1. The manuscript fails to define the abbreviation "IPS." It's crucial to introduce all abbreviations to maintain clarity for all readers.
+
+2. The description of the binding energy of P5 pentamers on Ag(111) as "strong" (Page 6) lacks specificity. Quantitative data or a more detailed qualitative assessment should replace this vague term to convey the findings accurately.
+
+3. The statement on Page 8 about cyclo-P5 having an "excess of charges" should clarify the charge type, presumably negative, to avoid ambiguity.
+
+4. The reference to an increase in the LWF at the P5/Ag interface on Page 8 needs specification. Clarifying whether the LWF is a calculated value would add to the reader's understanding.
+
+5. Misreferencing in the discussion of Figure 4 on Page 10 is very confusing and needs correction: I think "4d" should be updated to "4f," and "4e" should be "4g" to accurately guide the reader through the document.
+
+6. In the Data Availability section the authors refer to the Supporting Information, which, I guess, does not exist.
+
+7. Typographical and grammatical errors require comprehensive review. For instance:
+   a. "by the AFM image of Fig. 2c" should be corrected to "by the AFM image in Fig. 2c" (Page 5).
+   b. “an hexagonal lattice” should be corrected to “a hexagonal lattice” (Page 5)
+   c. "charge analysis show" should be revised to "charge analysis shows" (Page 8).
+
+In summary, this study sheds significant light on cyclo-P5, presenting findings that could captivate a broader audience. However, the highly technical nature of the discussion may exceed the typical expertise of Nature Communications readers. To enhance its accessibility and fit for publication in Nature Communications, I recommend simplifying the technical discussion while addressing the previously mentioned concerns.
+Reviewer #3 (Remarks to the Author):
+The authors report the on-surface synthesis of cyclo-P5 pentamers on Ag(111), which is an interesting new system that has only scarcely been studied. Via low-temperature scanning probe microscopy/spectroscopy and DFT calculations a comprehensive analysis of the structural and electronic properties of this system is performed. This gives important new insights into the charge transfer between the Ag substrate and the pentamers as well as the change of local work function that is caused by adsorption of the pentamers. This is particularly interesting for future applications of such new materials and should be appealing to a broad audience. The results are also well presented and structured. Therefore, I strongly recommend publishing the article after some minor revisions.
+
+Comments:
+- Please ensure that all abbreviations are explained (I guess “IPS” is missing)
+
+- Is the observed hexagonal lattice of cyclo-P5 pentamers commensurate with the substrate or completely incommensurate. In case of rather strong interactions with the surface atoms one would assume a commensurate structure.
+
+- Regarding the measured P-P bond length: Due to the bending of the CO-tip the apparent bond lengths which are extracted from the AFM images are usually somewhat higher or lower than the expected values (see e.g. Science 365, 142 – 145, 2019). In case of a ringlike molecule presumably one would expect an overestimation of the bond length (see e.g. Angew. Chem.Int. Ed.2023,62, e202310121). Can you comment on this. Does the measured size e.g. depend on the tip surface distance?
+
+- Fig2c: scale bar is missing.
+- Fig4b and c: Vs = -1.25 V or -0.5 V? The values in the caption and the images are different.
+- Page 9 bottom paragraph: “while measuring dl/dV” should read “dZ/dV”
+Reviewer #1 (Remarks to the Author):
+
+The authors present here a combined experimental and theoretical study of cyclo-P5 pentamers on a Ag(111) surface. Using non-contact AFM with a functionalized tip and DFT calculations, they characterize the atomic structure of cyclo-P5 pentamers assembled on the Ag(111) surface and reveal that a strong charge transfer occurs from the silver surface to the phosphorous structure. Consequently, they deduce the change in the surface work function.
+
+In the first part of the manuscript, they have characterized and modelled phosphorous chains on the Ag(111) surface with the nc-AFM, and recover the same results as obtained in Ref. [10]. The DFT simulations neither provide anything new, as the model is the same as in Ref. [10]. Therefore, I do not see the interest of this part of the manuscript. In particular, the “excellent agreement” between experiment and theory here was already the topic of the work in Ref. [10].
+
+In the second part of the manuscript, the authors focus on the synthesis of cyclo-P5 pentamers on the Ag(111) surface, by repeating the procedure described in Ref. [11]. They obtain the same structure and simulate the images according to the model developed in Ref. [12], yielding a good agreement.
+
+Then, and this is probably the most innovative part of the manuscript, they measure the surface electrostatic potential on the cyclo-P5 structure and on the bare silver surface, leading to a potential difference representative of an interface dipole and a strong charge transfer from silver to phosphorous. These features are supported by DFT calculations, enabling to determine the value of the surface dipole as well as the corresponding charge transfer. Furthermore, they have also characterized the Image Potential States at the interface.
+
+To my opinion, this work is a very good re-characterization of the already known cyclo-P5 system (and its precursor’s chains) with nc-AFM with functionalized tips images instead of STM images, and some additional spectroscopic measurements to characterize the interface states and charge transfer. DFT calculations have already been published together with experimental results for the Pchains (Ref. [10]) and in a single article for the pentamers (Ref. [12]), published by some co-authors of the present paper, and as such, do not bring anything new to the present manuscript.
+
+To summarize, while the experimental characterization of this known system is well executed, it does not bring any real innovative feature on this system, and therefore, it does not meet the criteria for
+publication in Nature Communications. Hence, this work should be published, but in a more standard journal, as an educated repetition of previous works.
+
+We thank the referee for her/his time to assess our manuscript. Despite a positive opinion on the quality of
+
+our experimental data, the referee’s decision to reject our manuscript is based on the criticism of the structural “re-characterization” of the phosphorus structures on Ag(111) that, according to her/him “does not
+
+bring any real innovative feature on this system”. While we agree that the synthesis of the cyclo-P5 pentamers and phosphorus chains have already been the subject of two experimental publications (Refs. 10-12) as well as one theoretical paper (Ref. 11), we do not share the same opinion regarding the lack of novelty. In comparison to these references, we provide a series of new experimental data by means of atomic force microscopy (AFM), force spectroscopy and tunneling spectroscopy performed at cryogenic temperature (4.5 K), whereas these works focused on STM conducted at room temperature or 77 K supported by DFT to disentangle the P structures. We reproduced the synthesis of the P structures as previously reported, which we acknowledge throughout the manuscript. However, our work goes a step beyond not only by confirming these P structures on Ag(111) using atomic force microscopy (AFM) data but
+
+also by providing an in-depth characterization of their interfacial properties at low temperature.
+
+To convince the referee on the novelty of our work, we must recall the context of our study.
+
+1- First AFM images on phosphorous chains and pentamers
+
+We agree that the references [10] and [11] are excellent scientific works, which give already good insight into
+
+the structure of the phosphorus chains and pentamers. However, the used STM method can be influenced
+
+by local variations of the local density states and structural information has to be derived with some caution.
+
+In our case, we use nc-AFM imaging with CO-termination, which has become an incontrovertible tool to determine structures with submolecular resolution. Since 2009 and the pioneered work by Gross et al on the
+
+pentacene molecule (Reference 19), atomic force microscopy (AFM) operated at low temperature with
+functionalized tips has provided unprecedented real-space resolution of atomic and molecule structures at
+
+the atomic level. This tool has become an undeniable asset for surface characterization and has notably boosted the field of on-surface chemistry. While lots of groups (including our) focus on imaging nanographene structures with such unprecedented resolution, we have recently targeted the exploration of
+
+two dimensional Xenes, starting with silicene on Ag(111) (Reference 23). Within this context, we used in this
+
+study our long-standing experience in imaging with the non-contact AFM technique to accurately investigate
+
+phosphorous pentamers and chains at low temperature. We also provide site-dependent force spectroscopic
+
+measurements to confirm the absence of buckling within the cyclo-P5 structure, in analogy to that observed
+
+for carbon atoms in nanographene. Therefore, we unambiguously determined the planar character of cycloP5 pentamers on Ag(111), which is for instance in stark contrast with the blue phosphorene on Au(111). In
+
+addition, we now extracted the P-P bond lengths per pentamer and found to three short and two long apparent bonds, which is consistent with the DFT calculations. This arises from the small deformations of the
+
+structure from its interaction with the substrate.
+
+2- DFT calculations are mandatory for AFM simulations
+
+One of the main criticism of the referee is that we reproduce the DFT calculations of the P5 pentamers and
+
+phosphorus chains, which have been extensively studied in Reference 12. However, for the sake of completeness, our work systematically provide alongside the experimental AFM images their simulated counterparts. These AFM simulations, obtained using the probe particle model (Reference 48) requires the
+
+DFT coordinates as inputs. More precisely, these simulations can be seen as the analogue of simulated STM
+
+images using the Tersoff-Hamman approximation for STM imaging and are essential for confirming the AFM
+
+contrast. Whereas our results confirms the previously published structures as pointed out by referee,
+understanding this contrast with functionalized tips still remains of prime importance for the AFM community,
+
+particularly when probing other elements beyond carbon in graphene. With this method, our investigation
+
+sets a new experimental approach based on AFM imaging and force spectroscopy. Therefore, removing the
+
+structural characterization of the cyclo-P5 pentamers on Ag(111) from our manuscript would severely impact
+
+the readability of our manuscript. The direct comparison of experimental AFM images and simulated AFM
+
+images based on DFT coordinates shows that the small height variations are due to different adsorption sites
+
+of the pentamer. If the center of the pentamer is slightly offset from the Ag-atom below, the heights of phosphorous are different by about 20 to 30pm. Only in the case, when the center of the pentamer is exactly
+
+on top of the Ag atom, the pentamer is perfectly flat-lying.
+
+3-Cryogenic measurements to unveil quantum states
+
+Our manuscript targets a complete description of the phosphorus system on Ag(111) from their structures to
+
+their electronic properties. By restricting our contribution to mainly a “re-characterization” of structures published elsewhere, we are afraid that the referee eludes the main message of our manuscript which deals
+
+with the charge redistribution at the P/Ag interface. Because none of the previous papers mentioned by referee have addressed this experimentally nor theoretically, we are convinced of the high novelty of our work. We hope that the referee shares our opinion on this particular point.
+
+Furthermore, although we do observe a substantial charge transfer towards the Ag substrate, our experiments clearly show that the pentamers are not hosting one unpaired electron per pentamer (absence
+
+of Kondo resonance), which is in clear contrast to its gas-phase counterpart (i.e. cyclo-P5-anion). Both DFT
+
+calculations with Bader charge analysis and AFM/IPS-spectra show that the remaining charge on the pentamer is about -0.115e on the P side and 0.057e on the Ag layer. This analysis is novel and was never
+addressed in the previous works.
+
+In light of the referee’s report, it appears to us that we have not been able to stress out the importance of the
+
+AFM characterization as compared to the previous literature. We thus decided to modify the main text as
+
+follows:
+
+“Low-temperature scanning probe microscopy is an incontrovertible tool for assessing atomic structures in
+
+contact to metals and characterizing their electronic properties with high spectral resolution in UHV. Atomic
+
+force microscopy (AFM) with functionalized tips19,20 has opened new avenues into the real-space imaging
+
+with improved lateral and vertical resolution of aromatic molecules and cyclo-carbons21 Recently, AFM imaging and spectroscopy have also tackled monoelemental 2D materials demonstrating a precise quantification of the atomic buckling in these structures.22,23 Not only restricted to structural characterization,
+
+charge distributions and work function changes are also accessible at the nanometer scale using Kelvin probe force microscopy (KPFM).24–27 In addition, the investigation of the local density of states (LDOS) of 2D
+
+materials near the Fermi level is readily achieved by means of scanning tunneling microscopy and spectroscopy (STM/STS). Tunneling spectroscopy can also probe the image potential states (IPS) of 2D synthetic materials as demonstrated in the case of graphene,
+
+28 germanene29 or borophene.
+
+30,31 Quantifying
+
+these Stark-shifted unoccupied states lying below the vacuum level give not only access to the fundamental
+
+physical processes involved in charge carrier dynamics but also to quantify local modulations of the work function at the interface between 2D materials and metals.”
+
+We also insist that we clearly acknowledge the previous references:
+
+“Based on the DFT coordinates we simulated the AFM image (see Methods, Fig.1f). The excellent agreement with the experimental image of Fig. 1d confirms the armchair structure of the P chains on Ag(111), similar to reference.
+"We thus confirm that the cyclo-P5 assembly adopts a (V7x6V7)R19° unit cell with respect to the Ag(111) surface lattice as previously reported by Zhang et al.11"
+
+Reviewer #2 (Remarks to the Author):
+
+The article titled 'Probing charge redistribution at the interface of self-assembled cyclo-P5 pentamers on Ag(111),' authored by Outhmane Chahib, Yulin Yin, Jung-Ching Liu, Chao Li, Thilo Glatzel, Feng Ding, Qinghong Yuan, Ernst Meyer, Rémy Pawlak, describes the synthesis and detailed characterization of the cyclical P5 phosphorus allotrope on the Ag(111) surface. Utilizing advanced scanning probe microscopy, the team presents high-resolution nc-AFM images, offering unparalleled structural insights for P5. Although the synthesis technique is previously documented (in Ref. 11), the novel contribution lies in the detailed examination of P5 molecule interactions with the Ag(111) surface. While this topic might appeal more to specialized circles, the depth of technical discussion presented may surpass the general scope preferred by the broad readership of Nature Communications. This nuanced exploration, though valuable for experts, could potentially challenge the wider audience's engagement due to its specialized technical complexity. The manuscript is well structured and the data in most cases supports well the claimed results.
+
+We thank the referee for the positive comments. We have now revised the main text to improve its readability.
+
+Furthermore, I must emphasize my deep-seated concerns on several major issues:
+
+1. In the sentence: "Thus, we conclude that the cyclo-P5 does not have a pure anionic character for the P5 molecule when adsorbed on Ag(111) (i.e. cyclo-P–5 ), as expected by theory for its gas-phase counterpart." the authors compare P5 molecules adsorbed on a surface with gas phase molecules, which is very confusing. How could the isolated P5 molecule have an anionic character in the gas phase? This needs clarification.
+
+Figure 1. The hexaphosphabenzenne (P6) and the cyclo-P5- anion
+
+We thank the referee for correcting this incomplete discussion. Each phosphorus atom has 5 valence
+electrons and can form up to five bonds. In the case of a planar hexaphosphabenzenne (P6), all electrons are
+
+paired forming each one single and on double bonds with neighboring P atoms (see Figure 1). For the cycloP5, one electron is unpaired leading to the anionic state of the cyclo-P5- ring. The cyclo-P5-
+
+anion has been
+
+identified using nuclear magnetic resonance (NMR) as a ring of unsubstituted twofold coordinated P atoms in
+
+gas phase (Baudler et al. Z. Naturforsch. B 1989 44, 381-382, now reference 41) or as a ligand (O. J. Scherer et al. Angew. Chem. Int. Ed. 1986, 25 363-364, now reference 42).
+
+We synthesized cyclo-P5 pentamers on Ag(111) by sublimating phosphorus atoms from a black phosphorus
+
+crystal thoroughly degassed in ultra high vacuum conditions. Therefore, one could also envisage that the cyclo-P5 ring keeps such anionic state in vacuum and upon adsorption on Ag(111). However, the cyclo-P5-
+
+ion can also coordinate with the Ag metal underneath, which induces a charge redistribution at the P5/Ag
+
+interface and the increase of local work function as described in our manuscript. Our dI/dV measurements
+
+show no Kondo resonance nor spin excitations near the Fermi level, which could have been the fingerprint of
+
+an unpaired electron of the cyclo-P5-
+
+anion. Note that such Kondo resonance has been recently observed by
+
+our group in the case of an anionic molecule adsorbed on Ag(111) (C. Li et al. Nat. Comm., 2023 14, 5956,
+
+now reference 43). To avoid any ambiguity, we thus think that it is important to discuss this point and clearly
+
+state that the cyclo-P5 ring on Ag(111) is not in such anionic state.
+
+We thus modified the main text as follows:
+
+“Finally, the planar cyclo-P5 structure has in principle an unpaired electron leading to an anionic state (i.e.
+
+cyclo-P5-
+
+), which has been identified by nuclear magnetic resonance (NMR) in gas phase or as a ligand. 38-39
+
+Upon adsorption on Ag(111), the cyclo-P5- anion can coordinate with the Ag atoms below it, leading to a
+charge redistribution at the interface and the formation of an interfacial state. This charge transfer modifies
+
+the amount of charge of the pentamer away from an integer, as confirmed by the Bader charge analysis showing an accumulation of electrons on P atoms (-0.115 e) and an electron depletion (+0.057 e) of the depleted Ag layer. We therefore conclude that cyclo-P5 molecules does not retain their anionic character on Ag(111). This conclusion is further corroborated by the absence of a Kondo resonance or spin excitations in dl/dV spectra acquired near the Fermi level.40
+
+2. The authors show dl/dV curve for P5 on Fig. 4a and discuss it in comparison with analogous measurement from Ref. 11. However, these curves have significant differences. For example, in the region of +2 V the curve drops down in Fig. 4a but it rises in Ref. 11. Additionally, the measured bandgaps are also different: 0.9 eV vs. 1.2 eV. This needs further discussion and the differences must be explained.
+
+Figure 2. Zoom of dl/dV spectra of Fig. 1.used to assign the valence band edge maximum (VBE) and the conduction band edge minimum (CBE). SS marks the position of the surface state of Ag(111).
+
+We thank the referee for pointing out errors of our manuscript. In reference 11, dl/dV spectra measured at 77 K of phosphorus pentamers on Ag(111) are shown to estimate a band gap of 1,2 eV. They attributed the valence band edge (VBE) at -1V and the conduction band edge (CBE) at about +0,2 eV. We note also that the spectra acquired on the pristine Ag(111) do not show a clear signature of a silver surface state. The referee noticed that we assign in the main text a gap of 0,9-1,0 eV. However, the arrow of Figure 4a shows a band gap of 1,2 eV by using the same CBE and VBE positions of Reference 11. This is an erroneous assignment, and we apologize for it.
+
+To address this, Figure 2 shows the rescaled dl/dV spectra of Figure 4a acquired on the bare metal (orange)
+
+and the phosphorus pentamers on Ag(111) (blue), respectively. It is important for us to first observe the
+surface state (SS) of silver in the STS spectra to confirm the cleanness of the tip. We observe the SS at about (-45 meV), which is slightly up-shifted by the presence of surrounding P5 domains as compared a clean Ag(111) (-68 meV). Looking at the orange spectra, we now assign with dashed lines the VBE at -0.45 eV and the CBE at 0.3 eV giving a band gap of about 0.75 eV.
+
+We now corrected the arrow of the Figure 4a. We also modified the main text as follows:
+
+“We marked with dashed lines the valence band edge maximum (VBE) at -0.45 eV and the conduction band edge minimum (CBE) to 0.3 eV, providing a gap Eg of the P5 assembly to about 0.75 eV.“
+
+Regarding the positive voltage region discussed by the referee, we must admit that it is difficult to judge experiments we have not carried out. We usually crosschecked the quality of the tip for STS measurements
+
+by observing the surface state position.
+
+Additionally, I have a few minor points to discuss:
+
+1. The manuscript fails to define the abbreviation "IPS." It's crucial to introduce all abbreviations to maintain clarity for all readers.
+
+We would like to thank the referee for noticing this error and apologize for the misunderstanding. We now
+
+revised the introduction and added the definition of the IPS acronym as follows:
+
+“The investigation of the local density of states (LDOS) of 2D materials near the Fermi level is readily achieved by means of scanning tunneling microscopy and spectroscopy (STM/STS). Tunneling spectroscopy can also probe the image potential states (IPS) of 2D synthetic materials such as graphene,28 germanene29 or borophene.30,31”
+
+2. The description of the binding energy of P5 pentamers on Ag(111) as "strong" (Page 6) lacks specificity. Quantitative data or a more detailed qualitative assessment should replace this vague term to convey the findings accurately.
+
+We regret the lack of clarity in the previous manuscript. In our revised document, we aim to definitively demonstrate the strong affinity between the pentamer P5 and the Ag(111) surface. To this end, we have
+provided the calculated binding energy of the P5 pentamer on the Ag(111) surface. Our calculations reveal
+
+that the binding energy of the pentamer on the Ag(111) substrate is -0.90 eV per atom (Figure 3, now Figure S1), highlighting the strong interaction between the pentamer and the Ag(111) surface.
+
+Figure 3. The configurations and corresponding binding energies of pentamer and hexagon on Ag(111) substrate.
+
+To enhance clarity and accuracy, we have accordingly revised the text:
+
+"Through Density Functional Theory (DFT) calculations, it has been determined that the cyclic P5 pentamer exhibits a higher binding energy on the Ag(111) surface, amounting to -0.90 eV per atom. This larger energy value is primarily facilitated through a charge transfer mechanism, promoting the stability of the cyclic P5 structure."
+
+This revision is aimed at providing a clearer understanding of the strong binding characteristics of the P5 pentamer on the Ag(111) substrate, highlighting the crucial role of charge transfer in stabilizing these structures.”
+
+3. The statement on Page 8 about cyclo-P5 having an "excess of charges" should clarify the charge type, presumably negative, to avoid ambiguity.
+
+We thank the referee for the comments. We corrected the sentence as follows:
+
+“Considering that the last Ag layer is depleted (red) while each cyclo-P5 has an excess of negative charges (blue), the P5 assembly can be approximated to a lattice of surface dipole moments of D = 1.42 Debye pointing towards the substrate.”
+
+4. The reference to an increase in the LWF at the P5/Ag interface on Page 8 needs specification.
+
+Clarifying whether the LWF is a calculated value would add to the reader's understanding.
+
+We thank the referee for its comment helping us to clarify our discussion. The increase of local work function (LWF) is experimentally observed by Kelvin probe force microscope (KPFM) as a shift of the local contact potential difference (LCPD, i.e. V* in Fig. 3b of the main manuscript) between tip and sample. The shift of the
+top of the parabola towards more positive (negative) voltage values provides a good indication of the increase (decrease) of the LWF. However, quantifying the exact change of work function is tedious because
+
+the work function of the tip is not precisely known and such measurements is strongly affected by averaging
+
+effects of the electrostatic forces between tip and sample. In the diagram of Fig. 3b, the increase of work function \( \Delta \Phi \) is thus set to an arbitrary value. We now clearly state about that in the main text as follows :
+
+“The observation of such surface dipole moments is consistent with an increase of the LWF at the P5/Ag interface, fixed to an arbitrary value \( \Delta \Phi \) in the diagram of Fig. 3d. This observation is consistent with an increase of the LWF at the $P_{(5)}$/Ag interface induced by an inwards dipole moment denoted \( \Delta \Phi \) in Fig.
+
+3d, which is in agreement with the increase of the LCPD in force spectroscopy. While the LWF increase agrees with the LCPD shift to more positive values in force spectroscopy, it is important to note that the LCPD variations are a qualitative indicate of the LWF changes at the atomic scale since it can have have a strong distance-dependence on metal substrate.39,40 Indeed, the \( \Delta V^* \) cannot directly account for the difference of work function \( \Delta \Phi = \Delta \Phi \) (P5/Ag) - \( \Delta \Phi \)(Ag) shown in Fig. 3d, due to averaging effects of the electrostatic interactions between tip and sample and the uncertainty in the work function of the tip. As we will be discuss later, a quantitative experimental value of LWF (0.46 eV) can be obtained by the analysis of IPS spectra.”
+
+5. Misreferencing in the discussion of Figure 4 on Page 10 is very confusing and needs correction: I think "4d" should be updated to "4f," and "4e" should be "4g" to accurately guide the reader through the document.
+
+We thank the referee for noticing these mistakes. The referencing of Figure 4 in the main text has been corrected accordingly.
+
+6. In the Data Availability section the authors refer to the Supporting Information, which, I guess, does not exist.
+
+The referee is right. Following the referee’s comments, we have now added a supporting information file.
+
+7. Typographical and grammatical errors require comprehensive review.
+For instance:
+
+a. "by the AFM image of Fig. 2c" should be corrected to "by the AFM image in Fig. 2c" (Page 5).
+
+b. “an hexagonal lattice” should be corrected to “a hexagonal lattice” (Page 5)
+
+c. "charge analysis show" should be revised to "charge analysis shows" (Page 8).0
+
+We thank the referee for the corrections, which we revised accordingly. We have also crosschecked the main
+
+text to improve its readability.
+
+In summary, this study sheds significant light on cyclo-P5, presenting findings that could captivate a broader audience. However, the highly technical nature of the discussion may exceed the typical expertise of Nature Communications readers. To enhance its accessibility and fit for publication in Nature Communications, I recommend simplifying the technical discussion while addressing the previously mentioned concerns.
+
+We thank the reviewer for it comments helping us to improve the manuscript.
+
+Reviewer #3 (Remarks to the Author):
+
+The authors report the on-surface synthesis of cyclo-P5 pentamers on Ag(111), which is an interesting new system that has only scarcely been studied. Via low-temperature scanning probe microscopy/spectroscopy and DFT calculations a comprehensive analysis of the structural and electronic properties of this system is performed. This gives important new insights into the charge transfer between the Ag substrate and the pentamers as well as the change of local work function that is caused by adsorption of the pentamers. This is particularly interesting for future applications of such new materials and should be appealing to a broad audience. The results are also well presented and structured. Therefore, I strongly recommend publishing the article after some minor revisions.
+
+Comments:
+
+- Please ensure that all abbreviations are explained (I guess “IPS” is missing)
+
+We would like to thank the referee for noticing this error and apologize. We now revised the introduction as
+
+follows:
+
+“The investigation of the local density of states (LDOS) of 2D materials near the Fermi level is readily
+achieved by means of scanning tunneling microscopy and spectroscopy (STM/STS). Tunneling spectroscopy can also probe the image potential states (IPS) of 2D synthetic materials such as graphene,28 germanene29 or borophene.30,31
+
+- Is the observed hexagonal lattice of cyclo-P5 pentamers commensurate with the substrate or completely incommensurate. In case of rather strong interactions with the surface atoms one would assume a commensurate structure.
+
+We thank the referee for the valuable comment. The assembly of cyclo-P5 is indeed in registry with the Ag substrate. This commensurability has been measured in reference 11, where the authors estimated the lattice parameter of the hexagonal assembly of 7.6 Å, which perfectly matches our experimental value extracted by STM and AFM. The stripes are distant by 4.56 nm which corresponds to six P5 rows. The pentamer assembly is rotated by 19° with respect to the Ag lattice, which is also confirmed by our DFT calculations. Therefore, we conclude as in reference 11 that the pentamer assembly presents a (V7×6V7) R19° unit cell with respect to Ag(111).
+
+We thus modified the text as follows:
+
+“In Fig. 2b, the close-up STM image reveals the structure of the self-assembled domains consisting of a hexagonal lattice with parameters a1 = b1 = 7.6 Å. Each bright protrusion corresponds to one cyclo-P5 molecule as schematized by the black dashed pentagons. Domains of cyclo-P5 rings exhibit a superstructure characterized by stripes separated by ~ 4.56 nm (i.e. six P5 rows) as shown by black dotted lines in Fig. 2a. These lines are rotated by 19° as compared to the [1-10] directions of the Ag(111) substrate,
+
+which agrees with previous experimental works11 as well as the relaxed structure obtained by DFT calculations of Fig. 2d. We thus confirm that the cyclo-P5 assembly adopts a (V7×6V7)R19° unit cell with respect to the Ag(111) surface lattice as previously reported by Zhang et al.11”
+
+- Regarding the measured P-P bond length: Due to the bending of the CO-tip the apparent bond lengths which are extracted from the AFM images are usually somewhat higher or lower than the
+expected values (see e.g. Science 365, 142 – 145, 2019). In case of a ringlike molecule presumably one would expect an overestimation of the bond length (see e.g. Angew. Chem.Int. Ed.2023,62, e202310121). Can you comment on this. Does the measured size e.g. depend on the tip surface distance?
+
+We thank the referee for the valuable comment since we overlooked this aspect of the system in our manuscript. The referee is right that the P-P bond lengths extracted from AFM data are always larger than
+
+the theoretical values. DFT calculations estimate a P-P bond length of about 2.16-2.22 Å for the pentamer
+
+adsorbed on Ag(111), which is in relative with the P-P bond length of 2.145 Å for the pentamer in gas-phase
+
+(Q. Jin et al. J. Mol. Struct. 2005 713, 113-117, now reference 33). The apparent bond lengths measured by
+
+AFM are in the range of 2.5 up to 2.7 Å. More precisely, each cyclo-P5 ring shows three short bonds of 2.5 Å
+
+(see red bonds of Figure 3) and two longer ones of 2.7 Å (see blue bonds of Figure 4). A similar distribution
+
+of bond lengths is also observed by DFT as shown in Figure 3c, with three short bonds of 2.17 and two long
+
+ones of 2.2.
+
+Regarding the measured P-P distance versus tip sample distance, we have not explored in details the impact
+
+of the tip-sample on the AFM contrast and thus cannot provide such series of AFM image versus tip-sample
+
+height. We however scanned at constant-height for moderated distances where no artifacts from CO tilting is
+
+expected.
+
+As pointed out by referee, it is well-established that apparent bond lengths extracted from AFM images are
+
+systematically larger than the theoretical ones. This has been for instance demonstrated in polycyclic aromatic hydrocarbons by Gross et al. (Science 2012 337,1236-1329 now reference 34) or in the case of P3N3 molecules by Zhong et al. (Angew. Chem. Int. Ed. 2023, 62, e202310121, now reference 35). This
+overestimation of the apparent bond lengths are due to the tilting of the CO attached to the AFM tip. Our data
+
+yield the same conclusion with a systematic increase of apparent bond lengths of about 8% as compared to
+
+DFT values.
+
+Figure 4. a, Representative AFM image of the cyclo-P5 assembly on Ag(111). Each pentamer consists of three short bonds of about 2.5 Å (depicted in blue) and two long ones of 2.7 Å (red). b, In the cycloP5-anion (top), P atoms have a valency 3 thus forming three single bonds (red) and two double bonds (blue). Alternative bond order (bottom) of the phosphorus pentamer considering four P atoms with valency 3 and one P atom with valency 5. c, Relaxed structure of P5 pentamers on Ag(111) by DFT, with the measured bond lengths.
+
+Let’s now discuss the bond order of the cyclo-P5 pentamer on Ag(111). Reference 34 has demonstrated that the apparent bond lengths of C-C single bonds in carbon rings are always longer than C=C double bonds, opening the accurate determination of bond order in aromatic molecules. Phosphorus atom can bond in two equivalent manner thanks to to the coexistence of the valency 3 (such as PH3) and 5 (such as PCI5). The cyclo-P5-anion in gas phase contains five P atoms, all having a valency 3 (see Fig. 4b). According to this
+reference, AFM images should show two short (blue) and three long (red) apparent bonds in the AFM images
+
+for the structure shown in Fig. 4b (top). However, we observe an opposite distribution by AFM, e.g. three short and two long apparent bonds as depicted by red and blue bonds in Fig. 4a, respectively. Following Ref.
+
+34, an alternative explanation of such bond order would be to consider a mixed valency per P5 pentamer with four P atoms of valency 3 and one of valency 5 as described in Fig. 4b (bottom). We however think that
+
+this configuration is unlikely. Instead, this particular bond order is induced by the small structural relaxation in
+
+Z direction of the cyclo-P5 ring induced by the Ag lattice, that we estimate of about 20 pm.
+
+To confirm this, we compared our experimental observations to bond lengths extracted from DFT calculations. In Figure 4c, we measured the P-P bond length for few pentamers on Ag(111) and found out a
+
+very similar distribution, namely two long bonds of 2.22 Å and three short ones of 2.18 Å.
+
+We thank the referee for the valuable comment. We modified the Figure 2 and add a supplementary figure to
+
+show the bond distribution. We also commented in the manuscript as follows:
+
+“The P-P bond length within the cyclo-P5 pentamer extracted by AFM varies from 2.5 to 2.7 Å, which is always larger by about 8% than that of DFT calculations for the pentamer on Ag(111) (Fig. 2d and and Supplementary Fig. 1) or in gas phase (2.185 Å).33 It is well-established that this overestimation of apparent
+
+bond lengths is induced by the tilting of the CO molecule attached to the AFM tip upon scanning, as shown
+
+for planar polyaromatic hydrocarbons or P3N3 molecules.34,35 Moreover, each cyclo-P5 ring is composed of
+
+three short apparent bonds (2.5 Å, blue bonds in Fig. 2b) and two longer ones (2.7 Å, red bonds). This particular bond order, also confirmed in the relaxed structure calculated by DFT (Fig. 2d), is likely induced by
+
+the small buckling of the structure when adsorbed on Ag(111).”
+
+- Fig2c: scale bar is missing.
+
+We thank the referee for noticing. We added the missing scale bar.
+- Fig4b and c: Vs = -1.25 V or -0.5 V? The values in the caption and the images are different.
+
+We thank the referee for correcting this mistake. The correct value is the one written in the caption. We thus
+
+modified the Figure 4 accordingly, by replacing -0.5V by -1.25V.
+
+- Page 9 bottom paragraph: “while measuring dI/dV” should read “dZ/dV”
+
+We thank the referee for the correction, we have corrected the text accordingly
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+The authors agree in their response that the synthesis and characterization of cyclo-P5 pentamers and phosphorous chains have already been the subject of at least three published articles (Ref. 10-11-12). They also claim to confirm P structures on Ag(111) using atomic force microscopy (AFM) data. The added value of their work lies in the characterization by atomic force microscopy, force spectroscopy at 4.5k instead of 77K which only confirms previous studies.
+
+Based on these statements and my previous review, I think this article is suitable for other journals such as PRB or JPCC but not for a prestigious journal like Nature Communications.
+
+Reviewer #2 (Remarks to the Author):
+The authors have addressed my previous comments to a large extent. However, as also noted by Reviewer 1, the manuscript lacks sufficient novelty to justify publication in Nature Communications. Additionally, it remains very technical, which might pose challenges for the broader audience of Nature Communications.
+Some additional minor comments:
+1. Authors use both K and °C units of temperature. I think one should be used throughout the manuscript.
+2. On page 5, authors write: “Each chain configuration has a relative STM height of 1.6 A, and a width of 11 °A, 17 °A, and 26 °A for the single, double and triple chains, respectively.”. Why the width per chain is decreasing nonlinearly? Concerning that there is space between chains one would expect increase in the width per chain in double and triple chains.
+3. On page 5 authors write: “a hexagonal lattice with parameters”. This is very confusing because pentagons make a pentagonal pattern rather than hexagonal.
+4. On Page 6 there is LaTeX syntax/typo mistake: “textitcyclo-P5”
+
+In summary, the manuscript is solid and shows good peace of work that is well described.
+However, I think it is more suitable to more specific journal.
+
+Reviewer #3 (Remarks to the Author):
+The authors have addressed all my earlier points satisfactorily. Therefore, I recommend to accept the manuscript.
+Title: Probing charge redistribution at the interface of self-assembled cyclo-P5 pentamers on Ag(111)
+
+We thank you the reviewers for their time and commitment in assessing our manuscript. We are delighted by their comments. Replies to individual comments can be found below, together with a list of changes/additions to the main text. The document also includes an annotated copy of the manuscript with track changes highlighted in blue.
+
+We do hope that you will find our revised manuscript suitable for further consideration in Nature Communication.
+
+Yours sincerely,
+
+Rémy Pawlak, on behalf of co-authors.
+
+Reviewer #1 (Remarks to the Author):
+
+The authors agree in their response that the synthesis and characterization of cyclo-P5 pentamers and phosphorous chains have already been the subject of at least three published articles (Ref. 1011-12). They also claim to confirm P structures on Ag(111) using atomic force microscopy (AFM) data. The added value of their work lies in the characterization by atomic force microscopy, force spectroscopy at 4.5k instead of 77K which only confirms previous studies. Based on these statements and my previous review, I think this article is suitable for other journals such as PRB or JPCC but not for a prestigious journal like Nature Communications.
+
+We cannot accept the referee’s decision as it stands since we think that she/he has overlooked the main advances of our work throughout the reports.
+
+Our manuscript demonstrates for the first time the experimental observation of a charge redistribution at the P5/Ag(111) interface. These new electronic properties represents an undeniable step forward in our understanding of the future implementation of phosphorus structures (cyclo-P5 and others) close to the electrodes in a device, a finding that goes well beyond its simple structural characterization. These results have never been published in the previous studies (Refs. 10-12), to which she/he refers to justify rejecting our work for lack of novelty. In comparison to these three references focusing on the synthesis of P structures on Ag(111), we measure for the first time the band structure of the P5/Ag system, an interface state and the change of local work function at the nanometer scale. Our experimental results thus surpass these previous works.
+
+The statement “the added value of their work lies in the characterization by atomic force microscopy, force spectroscopy at 4.5k instead of 77K which only confirms previous studies” is not justified. Atomic force microscopy (AFM) is used here for the first time to characterize the P structures on Ag(111) and its local work function. These data are new and deserve publication as they represent a fundamental breakthrough in the characterization of phosphorus structures at surfaces using the AFM technique. “Cyclo-P5 pentamers and phosphorous chains have already been the subject of at least three published articles (Ref. 10-11-12)”: this is an overstatement. To our knowledge, there are only three publications, all of which related to synthesis and structural characterization by scanning tunneling microscopy (STM). Reference [11] is purely theoretical about the synthesis and references [10] is only about the chains. Reference [12] deals with the pentamers (called “nanoflowers”) but are restricted to only STM experiments. No solid structural insights can be extracted from these data, which are comparable to our AFM images.
+
+We find regrettable that the discussion with Referee #1 has biased the reviewing process on the basis of questionable assessments. We are deeply convinced that, by trying to restrict the debate to the comparison with the structural characterization of previous papers, the referee deliberately sidestepped our results in order to conclude to a lack of novelty. Obviously, the content of our manuscript is much richer than the structural characterization of the system. In response to this criticism which we think to be unjustified, we list here the advances made by our work in comparison to the literature:
+
+1- Synthesis of P5/Ag(111) - Not novel but it is not the subject of our work
+As Referee #1 pointed out in her/his reports, all previous works deal with the synthesis of the P chains and P5 on Ag(111). They are clearly acknowledged in our manuscript, starting from the first sentence of the abstract. Our work focuses on the “charge redistribution at the interface of self-assembled cyclo-P5 pentamers on Ag(111)” as expressed in its title, abstract and main content.
+
+2- Structural characterization of P5/Ag(111) by AFM – Novel
+
+STM cannot resolve atomic structure of the P5 structure without the help of DFT calculations as it is primarily sensitive to electronic density, while AFM can by using force-distance curves and AFM imaging (see Pawlak et al. PNAS 2020, 117, 228-237). We applied such new experimental method based on AFM to accurately determinate atomic buckling and bond lengths in the cyclo-P5 structures. To date, this question has never been addressed experimentally.
+
+3- Electronic band structure – Novel
+
+Two STS spectra are shown in Reference 11. However, the absence of the interface state in these data should question the referee. In our work, we provide a series of dl/dV measurements that unambiguously resolve the valence band edge, conduction band edge and the interface state (IS) of the P5 system on Ag(111). These substantial advances, which were never published elsewhere, cannot be simply overlooked by the referee. Furthermore, our STS data cannot just be considered as a repetition since they go well beyond the previous publications.
+
+4- Quantifying local change of work function – Novel
+
+We estimate the direction of surface dipole moments created at the P5/Ag using force-voltage spectroscopic measurements, which is corroborated by DFT calculations as a result of a charge transfer from the substrate to the P structures. We quantify the local increase of work function of 0.46 eV using field-effect resonant tunneling (FERT). These experimental results have never been addressed in previous publications, which demonstrates once more the novelty of our work.
+
+Finally, we shall comment on why we think our work is suitable for Nature Communications as is. Nature Communications offers publication of significant and cutting-edge research in all areas of nanomaterials and nanotechnology with no limit on word length and number of figures. This allows authors to publish detailed and comprehensive studies, which we have taken advantage of to provide a complete analysis of the P structures on Ag system that goes well beyond the previous literature in terms of applied experimental techniques and experimental results. Therefore, our work not only provides new insights into the P5/Ag(111) system by unveiling its electronic property, it also set an experimental methodology combining both STM and AFM techniques to scrutinize these system in great details. Additionally, we demonstrate that the P5/Ag interface (see Figure 1) has the prototypical behavior of a p-type semiconductor-metal Schottky junction, opening up prospects for its use in field-effect transistor, diodes or solar cells.
+
+Reviewer #2 (Remarks to the Author):
+
+The authors have addressed my previous comments to a large extent. However, as also noted by Reviewer 1, the manuscript lacks sufficient novelty to justify publication in Nature Communications. Additionally, it remains very technical, which might pose challenges for the broader audience of Nature Communications.
+
+We invite the reviewer #2 to consult our response to referee #1 regarding the discussion on the lack of novelty claimed by the referee #1. As we discussed there, our manuscript demonstrates for the first time the experimental observation of a charge redistribution at the P5/Ag(111) interface. These new electronic properties represents an undeniable step forward in our understanding of the future implementation of phosphorus structures (cyclo-P5 and others) close to the electrodes in a device, a finding that goes well beyond its simple structural characterization.
+
+We shall also discuss the concerns of referee #2 regarding the technical aspect of our work. We utilized in our work a large number of experimental imaging and spectroscopic techniques. Each of them needs to be described in order to understand how the physical quantities are extracted from the measurement. In addition, we believe that our work sets a complete methodology to scrutinize the structure, band structure and work function changes of the 2D materials on Ag(111). This methodology is used here to interpret the
+energy level alignment of P5 assembly on Ag, but could be applied in general to 2D semiconductors on metals.
+
+To better underline the impact of our experimental approach, we summarize in Figure 1 the energy level diagram of the P5/Ag interface built only using the experimental estimates presented in our manuscript.
+
+![Energy level alignment at the P5/Ag interface extracted from the experimental data.](page_184_370_627_496.png)
+
+Figure 1. Energy level alignment at the P5/Ag interface extracted from the experimental data.
+The positions of the conduction band edge (CBE) and valence band edge (VBE) are set to 0.3 eV and - 0.5 eV in agreement with the dI/dV measurements. ΔΦ = 0.46 eV is the change of work function measured by field effect resonance tunneling (FERT). This value also approximates the band bending of the P5 conduction and valence band. The interface state position at 2.5 eV (red) is due to the confinement of holes from the conduction band. The cyclo-P5 assembly on Ag(111) behaves as a prototypical p-type Schottky semiconductor-metal junction.
+
+First, the positions of the conduction band edge (CBE) and valence band edge (VBE) are set to 0.3 eV and - 0.5 eV in agreement with the dI/dV measurements. The exact energy positions of the vacuum level \( E_{vac} \) for both Ag(111) and P5 are determined by FERT spectroscopy as well as the local variation of work function \( \Delta \Phi = 0.46 \) eV between them. The \( \Delta \Phi \) value also allows to approximate the band bending of the valence and conduction bands induced at the interface between the P5 assembly and the Ag(111).
+
+As described in our manuscript, the adsorption of P5 on Ag(111) leads to charge redistribution at the P5/Ag interface with a charge transfer from the metal to the P structures. The Ag metal is depleted (+ in Figure 1) while the P5 assembly is negatively charged (-). This indicates that the cyclo-P5 assembly on Ag(111) behaves as a prototypical p-type semiconductor-metal junction with a Schottky barrier at the interface. The interface state of the P5 assembly at +2.5 eV is then related to hole confinement at the barrier formed near the conduction band maximum. The width of the conduction band ?
+
+We now decided to add the Figure 1 to the supplementary information as Figure S3 and comment on it in the outlook of the main manuscript as follows:
+
+"Based on our experimental estimates, we summarize in Supplementary Fig. 3 the energy level alignment at the cyclo-P5 /Ag interface. Considering that the Ag metal is depleted while the P5 assembly is negatively charged, we conclude that the P5 /Ag system behaves as a prototypical p-type semiconductor-metal junction with a Schottky barrier built at the interface. It further shows that this system could have potential applications in field-effect transistors, diodes or solar cells. Finally, by exploring the fundamental characteristics of the prototypical cyclo-P5 /metal interface, our methodology (applicable to other emerging 2D materials and related quantum materials) not only showcase the importance of scanning probe microscopy as powerful techniques to study structural and electronic properties at the atomic scale, but also provides new insights for improved performances of phosphorus-based devices."
+
+We modified the conclusion as follows:
+
+"Our experimental approach suggest that the cyclo-P5 /Ag interface has the characteristic ingredients of a p-type semiconductor-metal Schottky junction with potential applications in field-effect transistors, diodes or
+solar cells. Our results suggest that the high quality of the cyclo P5 /Ag interface might serve as a prototypical system for electric contacts in phosphorus based semiconductor devices."
+
+Some additional minor comments:
+1. Authors use both K and °C units of temperature. I think one should be used throughout the manuscript.
+
+We thank the referee for noticing that. We now write all temperatures in Kelvin unit.
+
+2. On page 5, authors write: "Each chain configuration has a relative STM height of 1.6 Å, and a width of 11 * Å, 17 * Å, and 26 *Å for the single, double and triple chains, respectively.". Why the width per chain is decreasing non-linearly? Concerning that there is space between chains one would expect increase in the width per chain in double and triple chains.
+
+We thank the referee for the comments. We are afraid that estimating STM apparent heights and distances from topographic image can be impacted by the convolution with the tip apex. Since the AFM images provide a better estimate of these distances, we decided to delete this sentence from the manuscript. Instead, we provide the width of single, double and triple chains extracted from the AFM images (Figs. 1 b-d) as follows:
+
+"In each configuration, the chain has always an apparent width by AFM of 6.0 Å with a minimum distance between them of 0.25 Å. Thus, double and triple chains have apparent AFM widths of 14.5 Å and 23 Å, respectively. "
+
+3. On page 5 authors write: "a hexagonal lattice with parameters". This is very confusing because pentagons make a pentagonal pattern rather than hexagonal.
+
+We thank the referee for the correction and revised the text accordingly. We deleted the word hexagonal from the sentence, which now reads as follows:
+
+"The close-up STM image reveals the structure of the self-assembled domains consisting of a hexagonal lattice with parameters \( a_1 = b_1 = 7.6 \) Å."
+
+4. On Page 6 there is LaTeX syntax/typo mistake: “textitcyclo-P5”
+
+We thank the referee for the correction and revised the text accordingly.
+
+In summary, the manuscript is solid and shows good peace of work that is well described. However, I think it is more suitable to more specific journal.
+
+We thank the referee for the positive comments helping us to improve the readability and quality of our manuscript.
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have addressed all my earlier points satisfactorily. Therefore, I recommend to accept the manuscript.
+
+We thank the referee for its positive assessment.
+REVIEWERS' COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+As I already mentioned in my previous reports, these results have already been the subject of at least three published articles. The new aspect of the data is the characterization by atomic force microscopy, force spectroscopy at 4.5k instead of 77K which confirms the previous published results.
+Hence, I think this article is suitable for other journals such as PRB or JPCC but not for Nature Communications.
+
+Reviewer #2 (Remarks to the Author):
+The response to Reviewer 1 effectively argues for the novelty of this work, making a reasonable case for its publication in Nature Communications. Additionally, the authors have satisfactorily addressed my other comments in the latest version of the manuscript.

035e55afc9c40b2a32b10bb61cbaf9c417c4c43287f20e12b4733b13052ac290/preprint/preprint.md

@@ -0,0 +1,250 @@
+Probing charge redistribution at the interface of self-assembled cyclo-P5 pentamers on Ag(111)
+
+Rémy Pawlak
+remy.pawlak@unibas.ch
+
+University of Basel   https://orcid.org/0000-0001-8295-7241
+Outhmane Chahib
+University of Basel
+Yulin Yin
+Chinese Academy of Sciences
+Jung-Ching Liu
+University of Basel   https://orcid.org/0000-0002-9472-3343
+Chao Li
+Department of Physics, University of Basel   https://orcid.org/0000-0003-2125-9989
+Thilo Glatzel
+University of Basel   https://orcid.org/0000-0002-3533-4217
+Feng Ding
+Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences   https://orcid.org/0000-0001-9153-9279
+Qinghong Yuan
+East China Normal University   https://orcid.org/0000-0003-4683-2112
+Ernst Meyer
+https://orcid.org/0000-0001-6385-3412
+
+Article
+
+Keywords: cyclo-P–5 pentamer, work function, atomic force microscopy, scanning tunneling microscopy, field-emission resonance spectroscopy, density functional theory
+
+Posted Date: January 30th, 2024
+
+DOI: https://doi.org/10.21203/rs.3.rs-3777510/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on August 2nd, 2024. See the published version at https://doi.org/10.1038/s41467-024-50862-4.
+Probing charge redistribution at the interface of self-assembled cyclo-\(P_5\) pentamers on Ag(111)
+
+Outhmane Chahib,\(^1\) Yulin Yin,\(^2\) Jung-Ching Liu,\(^1\) Chao Li,\(^1\) Thilo Glatzel,\(^1\) Feng Ding,\(^2\) Qinghong Yuan,\(^3\) Ernst Meyer\(^{1,*}\) & Rémy Pawlak\(^{1,*}\)
+
+\(^1\)Department of Physics, University of Basel, Klingelbergstrasse 82, 4056 Basel, Switzerland
+\(^2\)Faculty of Materials Science and Engineering/Institute of Technology for Carbon Neutrality, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China
+\(^3\)State Key Laboratory of Precision Spectroscopy School of Physics and Electronic Science, East China Normal University, 500 Dongchuan Road, Shanghai 200241, China
+
+Abstract
+
+Phosphorus pentamer (\(cyclo-P_5^-\)) ions are unstable in nature but can be synthesized at the Ag(111) surface. Unlike monolayer black phosphorous, little is known about their electronic properties when in contact with metal electrodes, although this is crucial for future applications. Here we characterize the atomic structure of \(cyclo-P_5\) assembled on Ag(111) using atomic force microscopy with functionalized tips and density functional theory. Combining force and tunneling spectroscopy, we find that a strong charge transfer induces an inward dipole moment at the \(cyclo-P_5/\)Ag interface as well as the formation of an interface state. We probe the image potential states by field-effect resonant tunneling
+and quantify the increase of the local change of work function of 0.46 eV at the cyclo-\(P_5\) assembly. Our results suggest that the high-quality of the cyclo-\(P_5\)/Ag interface might serve as a prototypical system for electric contacts in phosphorus-based semiconductor devices.
+
+Keywords:  cyclo-\(P_5^-\) pentamer, work function, atomic force microscopy, scanning tunneling microscopy, field-emission resonance spectroscopy, density functional theory
+
+Introduction
+
+Elemental phosphorus (P) is not only ubiquitous in human life, it is also one of the most fascinating areas of chemistry as it can exist in a large diversity of allotropes,\(^{1,3}\) in various cluster configurations\(^{4,5}\) or in organic compounds.\(^{6}\) The phosphorous polymorphism is even multiplied on the atomic-scale when using a surface to constrain the reaction in two dimensions (2D). As in the field of on-surface chemistry producing complex nanographene structures in ultra-high vacuum (UHV),\(^{2,8}\) surface-assisted phosphorus reactions on metals have synthesized blue phosphorus,\(^{9}\) P chains,\(^{10}\) or even planar cyclo-\(P_5\) rings.\(^{11,12}\) Since then, phosphorus allotropes have emerged as a promising one-atom thick 2D material beyond graphene, due to its moderate direct band gap (0.3 to 2.0 eV)\(^{13}\) suitable for nanoelectronics and nanophotonics applications.\(^{14,15}\) However, allotropic configurations, their atomic buckling, defects or potential alloy formation can be detrimental for the semiconducting character. In addition, the interaction of 2D materials with delocalized electrons of a
+metal, as well as the dynamical charge transfer between the two media, are key factors in fostering new gate-tunable functionalities such as superconductivity.[16][17] Experimental study of these aspects at the fundamental level is therefore essential for future quantum applications where metallic electrical contacts are required.[18]
+
+Low-temperature scanning probe microscopy is an incontrovertible tool for assessing atomic structures in contact to metals and characterizing their electronic properties with high spectral resolution in UHV. Atomic force microscopy (AFM) with functionalized tips[9][20] has opened new avenues into the real-space imaging with improved lateral and vertical resolution of aromatic molecules, cyclo-carbon[21] and monoelemental 2D materials,[22][23] while charge distributions and work function changes at the nanometer scale are also accessible using Kelvin probe force microscopy (KPFM).[24][27] The investigation of the local density of states (LDOS) of 2D materials near the Fermi level is readily achieved by means of scanning tunneling microscopy and spectroscopy (STM/STS). Tunneling spectroscopy can also probe the IPS of 2D synthetic materials such as graphene,[25] germanene[29] or borophene.[30][31] Quantifying these Stark-shifted unoccupied states lying below the vacuum level give not only access to the fundamental physical processes involved in charge carrier dynamics but also to quantify local modulations of the work function at the interface between 2D materials and metals.
+
+By applying this *in-situ* methodology, we determine here the structure of phosphorus
+chains and self-assembled *cyclo*-*P$_5$* pentamers on Ag(111) using low-temperature (4.5 K) AFM imaging with CO-terminated tips. KPFM spectroscopic measurements indicates the formation of an inwards dipole moment at the *P$_5$*/Ag interface, which results from the charge transfer from the Ag substrate to the network as confirmed by DFT calculations. This charge transfer leads to a complex charge redistribution and the formation of an interfacial hybridized state (IS). Through field-emission resonance tunneling (FERT) and STS spectroscopy, we determined the energy position of the IS state and the series of IPS at the *cyclo*-*P$_5$* assembly as compared to pristine Ag, confirming an increase of the local work function of \( \sim 0.46 \) eV. Given the strong interest in tailoring the physical characteristics of monoelemental 2D materials contacted to a metal, we think that the *cyclo*-*P$_5$*/Ag interface might serve as a model system for future devices involving electric contacts.
+
+**Atomic-scale imaging of phosphorus chains and *cyclo*-*P$_5$* pentamers**
+
+Phosphorus atoms were sublimed in UHV onto the Ag(111) substrate kept at about 150 °C (see Methods). Figure[1h] shows an STM overview image of the resulting structures for a relative coverage of less than 0.3 monolayer (ML). Extended 1D chains aligned along the [1\(\bar{1}\)0] directions of Ag(111) (marked 1 in Fig. 1h) coexist with domains of *P$_5$* molecules (2), as recently reported in references.[10][11] The inset of Fig. 1a further shows a STM image of the double and triple chains, that depends on the P deposition rate.[11] Each chain configuration has a relative STM height of 1.6 Å, and a width of 11 Å, 17 Å, and 26 Å for the single, double and triple chains, respectively.
+To precisely determine their atomic configurations, we employed AFM imaging with CO-terminated tips (see Methods, Figs.1c-d).19 A common AFM contrast is observed for all configurations assigned to an armchair structure of the chain, which resembles that of hydrocarbon chains.32 The relaxed structure of the triple chain configuration calculated from DFT is shown in Fig. 1e. Phosphorus atoms colored in orange sit on bridge sites of the Ag lattice (gray) and are aligned along one [1\bar{1}0] direction in accordance with the experimental data. Based on the DFT coordinates we simulated the AFM image (see Methods, Fig1). The excellent agreement with the experimental image of Fig. 1d confirms the armchair structure of the P chains on Ag(111).
+
+Increasing the P coverage to about 0.4 ML while keeping the substrate at 150°C leads to the formation of large islands of *cyclo*-P$_5$ pentamers relative to the chains (Fig. 2a).12 In Fig. 2b, the close-up STM image reveals the structure of the self-assembled domains consisting of an hexagonal lattice with parameters $a_1 = b_1 = 7.3$ Å. Each bright protrusion corresponds to one *cyclo*-P$_5$ molecule as schematized by the black dashed pentagons. Domains of *cyclo*-P$_5$ rings also exhibit a superstructure characterized by stripes separated by \( \approx 3.8 \) nm (i.e. 6 $P_5$ rows) as shown by black dotted lines in Fig. 2a. These lines are rotated by 19° as compared to the [1\bar{1}0] directions of the Ag(111) substrate, which agrees with previous experimental works11 as well as the relaxed structure obtained by DFT calculations (Fig. 2d).12 A deeper insights into the chemical structure of the *cyclo*-P$_5$ molecules is provided by the AFM image of Fig. 2c. The P-P bond length within the pentagon extracted by AFM is about
+2.2 Å, which is comparable to the value of 2.185 Å obtained by DFT. For comparison, we also simulated the AFM image based on the DFT coordinates, allowing us to confirm the exact position and structure of the \( P_5 \) molecules in their self-assembly in registry with the Ag(111).
+
+To accurately quantify the atomic corrugation within the *cyclo*-\( P_5 \) structure, we acquired a series of site-dependent \( \Delta f(Z) \) spectroscopic curves (Fig. 2f), at the locations marked in the inset AFM image. The black and gray curves were obtained on Ag and between two pentamers, respectively. On top of neighboring atoms of a *cyclo*-\( P_5 \) (orange and brown curve), the spectra exhibits a characteristic dip arising from the interaction between the front-end oxygen atom of the CO-terminated tip with the phosphorus atom. The dashed vertical lines indicate the Z position of their bottoms and is the signature of the relative atomic Z height.\(^{23}\) The difference \( \Delta Z \) of \( \approx 20\text{-}30 \) pm thus represents the intrinsic atomic corrugation within the *cyclo*-\( P_5 \) pentagonal structure,\(^{5}\) which is comparable with atomic corrugations in graphene\(^{33}\) or planar molecules.\(^{34}\) Thus, this confirms the planarity of the *cyclo*-\( P_5 \) structure,\(^{5}\) as reflected in the constant-height AFM image of Fig. 2e.
+
+**Charge distribution at the *cyclo*-**\( P_5 \)**/**Ag interface**
+
+The binding energy of \( P_5 \) pentamers has been calculated by DFT to be strong on Ag(111), allowing the stabilization of the *cyclo*-\( P_5 \) structure through a charge transfer.\(^{12}\) To provide
+insights into the charge distribution at the cyclo-P5/Ag interface, we performed force *versus* voltage spectroscopic measurements (see Methods). Experimentally, the frequency shift \( \Delta f \) as a function of the sample bias \( V_s \) is measured at a constant tip height \( Z \), providing in the \( \Delta f(V) \) curve a parabola due to the electric force acting between tip and sample. The voltage \( V^* \) at top of the parabola represents the local contact potential difference (LCPD) between tip and sample, which allows one to image charge distributions and work function changes with nanoscale resolution.[24][27] Figure 3a shows a \( \Delta f(V) \) cross-section acquired across a \( P_5 \) domain (see STM inset of Fig. 3a). Single \( \Delta f(V) \) point-spectra on top of the \( P_5 \) network (orange) and on Ag(111) (black ) are plotted in Fig. 3b, respectively. Dashed lines in both figures refers to the \( V^* \) position. The LCPD value systematically shifts towards positive values (\( \Delta V^* \approx 0.22 \) V) for the pentamer assembly as compared to the pristine Ag substrate. This indicates the accumulation of charges at the \( P_5 \) network as compared to the Ag substrate.
+
+To better rationalize this, we calculated the charge redistribution at the cyclo-P5/Ag interface (see Methods), whose top and side views of isosurfaces of electron accumulation (blue, +\( 13 \times 10^{-3} \) e/\(\text{\AA}^3\)) and depletion (red, -\( 13 \times 10^{-3} \) e/\(\text{\AA}^3\)) are displayed in Fig. 3c. An electron transfer from the Ag(111) substrate to the P atoms of the pentamers (is observed as a charge accumulation located at the cyclo-P5 ring (red). In the \( P_5 \)/Ag gap (marked by white and black dashed lines in the side view of Fig. 3f), charge accumulation/depletion layers emerges below each cyclo-P5 structure, which supports the formation of an hybridized state.[30][31] We emphasize that such an interface state is not restricted to 2D Xenes on metals
+since it is well-established in organic/metal systems.\(^{35}\)
+
+Between *cyclo*-*P*\(_5\) rings, we note the absence of in-plane charge redistribution. Considering that the last Ag layer is depleted (red) while each *cyclo*-*P*\(_5\) has an excess of charges (blue), the *P*\(_5\) assembly can be approximated to a lattice of surface dipole moments of \(D = 1.42\) Debye pointing towards the substrate (see arrow in Fig. 3c). This observation is consistent with an increase of the LWF at the *P*\(_5\)/Ag interface induced by an inwards dipole moment as schematized in Fig. 3d, which is in agreement with the increase of the LCPD in force spectroscopy. It is important to mention here that the LCPD value has a strong distance-dependence on metal substrate, which is good indicate of the local work function changes at the atomic scale, but prevents quantitative determination.\(^{36,37}\) Indeed, the \(\Delta V^*\) cannot directly account for the difference of work function \(\Delta \phi = \phi_{P_5/Ag} - \phi_{Ag}\) shown in Fig. 3d, due to averaging effects of the electrostatic interactions between tip and sample. Last, the presence of an interfacial state can alter the amount of charge transfer away from an integer number compared to those with weaker interactions or adsorbed on insulating layers. Indeed, the Bader charge analysis show an accumulation of electrons on P atoms (-0.115 e) and an electron depletion (+0.057 e) of the depleted Ag layer. Thus, we conclude that the *cyclo*-*P*\(_5\) does not have a pure anionic character for the *P*\(_5\) molecule when adsorbed on Ag(111) (i.e. *cyclo*-*P*\(_5^-\)), as expected by theory for its gas-phase counterpart.
+Interface state and work function of the cyclo-\( P_5 \) assembly
+
+To shed more lights into the electronic properties at the \( P_5 \)/Ag interface, we next performed differential conductance measurements (dI/dV) across one \( P_5 \) domain (see Methods). Figure 4h shows the typical dI/dV point-spectra spectra of the network (orange) as compared to Ag(111) (black). We assign the gap of the \( P_5 \) assembly to about 0.9-1.0 eV (dashed lines) similar to Ref.\(^{11}\). The spectra also shows a strong resonance at 2.5 V, which we attribute to tunneling into the interface state (\( IS \)), respectively. dI/dV maps (Fig. 4b) further reveal the density of states at the valence band at \( V_s = -0.5 \) V. This atomic feature evolves to a stripe pattern at \( V_s = +2.5 \) V (Fig. 4c), revealing the spatial modulation of the IS state (Fig. 4c) similar to the superstructure shown in STM topographic image of Fig. 2a. This state, which derived from the occupied Shockley state of the clean Ag(111) surface, is upshifted by more than 2 eV and becomes unoccupied by the presence of the \( P_5 \) assembly.\(^{53}\)
+
+To quantify the local change of work function (LWF), we acquired field-effect resonant tunneling (FERT) spectra in order to probe IPS between the cyclo-\( P_5 \) assembly and silver.\(^{28,31}\) Experimentally, FERT spectra (also called dZ/dV spectroscopy) are obtained by sweeping the sample voltage \( V_s \) while measuring dI/dV at constant-current by adjusting the tip height Z using the STM feedback loop (see Methods). From a quasi-classical approximation (Fig. 4d), tunneling resonances spectrum occurs when the Fermi level of the tip aligns with the Stark-shifted IPS states and follow the equation :
+
+\[
+eV_n = \phi + \left( \frac{3n\pi\hbar eE}{\sqrt{2m}} \right)
+\]
+
+(1)
+where \( V_n \) is the sample voltage for the n\(^{th}\) IPS, \( \phi \) is the work function of the sample, \( m \) is the free electron mass and \( E \) is the electric field.
+
+Figure 4d shows the series of IPS states obtained above the cyclo-\( P_5 \) self-assembly as compared to Ag(111), respectively. The resonance at \( V_s = 2.2 \) eV of the orange spectrum, which is absent for the Ag one (black), corresponds to the IS state. The peaks noted \( n = 1 \) to 7 of the black spectra are the IPS states of the pristine Ag substrate. On the \( P_5 \) assembly, IPS states are clearly shifted to higher voltage when increasing the electric field (i.e. \( V_s \)), which is the signature of the change of LWF.\(^{30,31}\)
+
+A quantitative estimation of the LWF can be obtained from Eq. 1. In Fig. 4e, we plot the voltage position \( V_n \) of the IPS states as a function of \( n^{2/3} \) for both the cyclo-\( P_5 \) network (orange squares) and the Ag(111) substrate (black triangles). By fitting the linear progression of each datasets, we extract the LWF value corresponding to the y-intercepts to \( \phi_{Ag} = 4.49 \) eV and \( \phi_{P_5} = 4.95 \) eV. Considering that our experimental estimate of \( \phi_{Ag} \) is in excellent agreement with that obtained by ultraviolet photoelectron spectroscopy (UPS),\(^{35}\) we confirm the strong increase of LWF of \( \Delta \phi = 0.46 \) eV induced by the cyclo-\( P_5 \) assembly adsorbed on Ag(111). Altogether, the observation of an IS and the shift of the IPS resonances in tunneling spectroscopy point to a charge transfer from the Ag substrate to the cyclo-\( P_5 \) network and the creation of a strong inwards electric dipole at the interface.
+Summary and outlook
+
+In summary, we synthesized phosphorus chains and cyclo-\(P_5\) pentamers by depositing phosphorus atoms on atomically flat Ag(111) in ultra-high vacuum. Using low-temperature AFM with CO-terminated tips, armchair \(P\) chains and the planar cyclo-\(P_5\) rings are resolved with atomic precision. Flat-lying cyclo-\(P_5\) pentamers self-assemble into an extended hexagonal assembly in registry with the Ag substrate. DFT calculations support a substantial charge transfer from the Ag substrate to \(P_5\) pentamers, which results in a complex charge redistribution at the \(P_5/\text{Ag}\) interface and the emergence of an interface state. Using force spectroscopic measurements, the inward surface dipole moment induced by this charge transfer is confirmed as an increase of the LCPD value at the \(P_5\) assembly in comparison to the pristine Ag. This corresponds to an increase of LWF at the \(P_5\) network as compared to the bare metal substrate. We corroborated these measurement with FERT spectroscopy allowing us to quantify the LWF increase of 0.46 eV at the \(P_5/\text{Ag}\) interface. By exploring the fundamental characteristics of the prototypical cyclo-\(P_5\)/metal interface, our results not only underline the importance of scanning probe microscopy (applicable to other emerging 2D materials and related quantum materials) to study structural and electronic properties at the atomic scale, but also provides new insights for improved performances of phosphorus-based devices.
+Methods
+
+Sample preparation
+The Ag(111) substrate purchased from Mateck GmbH was sputtered by Ar\(^+\) ions and annealed at 500 \(^\circ\)C to eliminate any surface contaminations. Phosphorus atoms were sublimed by heating up a black phosphorus crystal contained in a Knudsen cell in ultra high vacuum (UHV). The P flux was estimated using a quartz microbalance. To obtain the phosphorus chains and \(P_5\) domains, we annealed the Ag(111) substrate during deposition at temperatures described in Ref.\(^9\).
+
+STM experiments
+The STM experiments were conducted at a temperature of 4.8 K using an Omicron GmbH low-temperature STM/AFM system operated with Nanonis RC5 electronics. Differential conductance spectroscopy dI/dV(V) spectra were acquired with the lock-in amplifier technique using a modulation of 610 Hz and a modulation amplitude of 10 meV. All voltages refer to the sample bias \(V_s\) with respect to the tip. For field-emission resonance tunneling spectroscopy (FERT), the lock-in amplifier generates a 15-30 mV (RMS) bias modulation at 650 Hz. The FERT spectra is obtained by recording the differential conductance data while the sample bias is swept with a closed feedback loop (setpoints: \(I_t = 1\) pA, \(V_s = 500\) mV).
+
+AFM experiments
+AFM measurements were performed with commercially available tuning-fork sensors in the qPlus configuration\(^{39}\) equipped with a tungsten tip (\(f_0 = 26\) kHz, Q = 10000 to 25 000,
+nominal spring constant \( k = 1800 \), N.m\(^{-1}\), oscillation amplitude A \( \approx 50 \) pm. Constant-height AFM images were obtained using tips terminated with a single carbon monoxide (CO) in the non-contact mode (frequency-modulated AFM–FMAFM) at zero voltage.\(^{19,40}\) CO molecules were adsorbed on the sample maintained at low temperature below 20 K. Before its functionalization, the apex was sharpened by gentle indentations into the silver surface. A single CO molecule was carefully attached to the tip following the procedure of reference.\(^{41}\) Simulations of the AFM images based on the DFT coordinates were carried out using the probe-particle model.\(^{42}\) Site-dependent \( \Delta f(Z) \) spectroscopic measurements to determine the atomic buckling of phosphorus pentamers were obtained with CO-terminated tips. The \( \Delta f(V) \) cross-section of \( 1 \times 85 \) pixels\(^2\) was acquired with Ag-coated metallic tips (tunneling setpoints: \( I_t = 1 \) pA, \( V_s = 800 \) mV, \( Z_{offset} = +80 \) pm).
+
+DFT calculations
+
+All density functional theory calculations were carried out in the Vienna ab initio simulation package (VASP)\(^{43}\) with projector augmented wave (PAW)\(^{44,45}\) method. The generalized gradient approximation (GGA) in the framework of Perdew-Burke-Ernzerhof (PBE)\(^{46}\) was chosen with the plane-wave cutoff energy set at 400 eV for all calculations. The DFT-D3\(^{47}\) method of Grimme was employed to describe the van der Waals (vdW) interactions. The geometries of the structures were relaxed until the force on each atom was less than 0.02 eV Å\(^{-1}\), and the energy convergence criterion of \( 1 \times 10^{-4} \) eV was met. The Brillouin zone was sampled using Gamma k-mesh with a separation criterion of 0.03. Metal slabs with 3 atomic layers was adopted as the substrate and the bottom layer was fixed to simulate
+the bulk. The vacuum spacing between neighboring images was set at least 15 Å along the non-periodic directions to avoid a periodic interaction.
+
+Data availability
+
+The data that supports the findings of this study are available within the paper or its Supplementary Information. All STM/AFM images are raw data. The raw data of spectroscopic measurements are available from the repository ZENODO (link will be updated).
+
+References
+
+1. Liu, H., Du, Y., Deng, Y. & Ye, P. D. Semiconducting black phosphorus: synthesis, transport properties and electronic applications. Chem. Soc. Rev. **44**, 2732–2743 (2015).
+
+2. Carvalho, A. *et al.* Phosphorene: from theory to applications. *Nat. Rev. Mater.* **1**, 1–16 (2016).
+
+3. Batmunkh, M., Bat-Erdene, M. & Shapter, J. G. Phosphorene and phosphorene-based materials – prospects for future applications. *Adv. Mater.* **28**, 8586–8617 (2016).
+
+4. Jones, R. O. & Hohl, D. Structure of phosphorus clusters using simulated annealing - P2 to P8. *J. Chem. Phys.* **92**, 6710–6721 (1990).
+
+5. Chen, M. D., Huang, R. B., Zheng, L. S., Zhang, Q. E. & Au, C. T. A theoretical study for the isomers of neutral, cationic and anionic phosphorus clusters P5, P7, P9. *Chem.
+Phys. Lett. 325, 22–28 (2000).
+
+6. Giusti, L. et al. Coordination chemistry of elemental phosphorus. Coord. Chem. Rev. 441, 213927 (2021).
+
+7. Cai, J. et al. Atomically precise bottom-up fabrication of graphene nanoribbons. Nature 466, 470–473 (2010).
+
+8. Clair, S. & de Oteyza, D. G. Controlling a chemical coupling reaction on a surface: Tools and strategies for on-surface synthesis. Chem. Rev. 119, 4717–4776 (2019).
+
+9. Zhang, J. L. et al. Epitaxial growth of single layer blue phosphorus: a new phase of two-dimensional phosphorus. Nano Lett. 16, 4903–4908 (2016).
+
+10. Zhang, W. et al. Flat epitaxial quasi-1D phosphorene chains. Nat. Comm. 12, 5160 (2021).
+
+11. Zhang, W. et al. Phosphorus pentamers: floating nanoflowers form a 2D network. Adv. Funct. Mater. 30, 2004531 (2020).
+
+12. Yin, Y., Gladkikh, V., Yuan, Q. & Ding, F. Phosphorus chains and pentamers: The precursors of blue phosphorene on the Ag(111) substrate. Chem. Mater. 34, 8230–8236 (2022).
+
+13. Li, L. et al. Direct observation of the layer-dependent electronic structure in phosphorene. Nat. Nanotechnol. 12, 21–25 (2017).
+14. Liu, H. et al. Phosphorene: an unexplored 2D semiconductor with a high hole mobility. ACS Nano **8**, 4033–4041 (2014).
+
+15. Wang, X. et al. Highly anisotropic and robust excitons in monolayer black phosphorus. Nat. Nanotechnol. **10**, 517–521 (2015).
+
+16. Shao, D. F., Lu, W. J., Lv, H. Y. & Sun, Y. P. Electron-doped phosphorene: A potential monolayer superconductor. Europhys. Lett. **108**, 67004 (2014).
+
+17. Zhang, R., Waters, J., Geim, A. K. & Grigorieva, I. V. Intercalant-independent transition temperature in superconducting black phosphorus. Nat. Comm. **8**, 15036 (2017).
+
+18. Li, Y., Yang, S. & Li, J. Modulation of the electronic properties of ultrathin black phosphorus by strain and electrical field. J. Phys. Chem. C **118**, 23970–23976 (2014).
+
+19. Gross, L., Mohn, F., Moll, N., Liljeroth, P. & Meyer, G. The chemical structure of a molecule resolved by atomic force microscopy. Science **325**, 1110–1114 (2009).
+
+20. Mönig, H. et al. Quantitative assessment of intermolecular interactions by atomic force microscopy imaging using copper oxide tips. Nat. Nanotechnol. **13**, 371–375 (2018).
+
+21. Kaiser, K. et al. An sp-hybridized molecular carbon allotrope, cyclo[18]carbon. Science **365**, 1299–1301 (2019). URL https://doi.org/10.1126/science.aay1914.
+
+22. Liu, X. et al. Geometric imaging of borophene polymorphs with functionalized probes. Nat. Comm. **10**, 1642 (2019).
+23. Pawlak, R. et al. Quantitative determination of atomic buckling of silicene by atomic force microscopy. Proc. Nat. Acad. Sci. **117**, 228–237 (2020).
+
+24. Mohn, F., Gross, L., Moll, N. & Meyer, G. Imaging the charge distribution within a single molecule. Nat. Nanotechnol. **7**, 227–231 (2012).
+
+25. Schuler, B. et al. Contrast formation in Kelvin probe force microscopy of single \( \pi \)-conjugated molecules. Nano Lett. **14**, 3342–3346 (2014).
+
+26. Meier, T. et al. Donor–acceptor properties of a single-molecule altered by on-surface complex formation. ACS Nano **11**, 8413–8420 (2017).
+
+27. Pawlak, R. et al. Hydroxyl-induced partial charge states of single porphyrins on titania rutile. J. Phys. Chem. C **121**, 3607–3614 (2017).
+
+28. Borca, B. et al. Potential energy landscape for hot electrons in periodically nanostructured graphene. Phys. Rev. Lett. **105**, 036804 (2010).
+
+29. Borca, B. et al. Image potential states of germanene. 2D Materials **7**, 035021 (2020).
+
+30. Liu, X., Wang, L., Yakobson, B. I. & Hersam, M. C. Nanoscale probing of image-potential states and electron transfer doping in borophene polymorphs. Nano Lett. **21**, 1169–1174 (2021).
+
+31. Liu, X. et al. Borophene synthesis beyond the single-atomic-layer limit. Nat. Mater. **21**, 35–40 (2022).
+32. Giovanelli, L. et al. On-surface synthesis of unsaturated hydrocarbon chains through C-S activation. Chem. A Eur. Jour. **28**, e202200809 (2022).
+
+33. Boneschanscher, M. P., Hämäläinen, S. K., Liljeroth, P. & Swart, I. Sample corrugation affects the apparent bond lengths in atomic force microscopy. *ACS Nano* **8**, 3006–3014 (2014).
+
+34. Kawai, S. *et al.* Diacetylene linked anthracene oligomers synthesized by one-shot homocoupling of trimethylsilyl on Cu(111). *ACS Nano* **12**, 8791–8797 (2018).
+
+35. Gonzalez-Lakunza, N. *et al.* Formation of dispersive hybrid bands at an organic-metal interface. *Phys. Rev. Lett.* **100**, 156805 (2008).
+
+36. Nony, L., Foster, A. S., Bocquet, F. & Loppacher, C. Understanding the atomic-scale contrast in Kelvin probe force microscopy. *Phys. Rev. Lett.* **103**, 036802 (2009).
+
+37. Sadeghi, A. *et al.* Multiscale approach for simulations of Kelvin probe force microscopy with atomic resolution. *Phys. Rev. B* **86**, 075407 (2012).
+
+38. Hofmann, O. T. *et al.* Orientation-dependent work-function modification using substituted pyrene-based acceptors. *J. Phys. Chem. C* **121**, 24657–24668 (2017).
+
+39. Giessibl, F. J. The qPlus sensor, a powerful core for the atomic force microscope. *Rev. Sci. Instrum.* **90**, 011101 (2019).
+40. Pawlak, R., Kawai, S., Fremy, S., Glatzel, T. & Meyer, E. Atomic-scale mechanical properties of orientated C_{60} molecules revealed by noncontact atomic force microscopy. ACS Nano **5**, 6349–6354 (2011).
+
+41. Bartels, L. *et al.* Dynamics of electron-induced manipulation of individual CO molecules on Cu (111). *Phys. Rev. Lett.* **80**, 2004 (1998).
+
+42. Hapala, P. *et al.* Mechanism of high-resolution STM/AFM imaging with functionalized tips. *Phys. Rev. B* **90**, 085421 (2014).
+
+43. Kresse, G. & Furthmüller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. *Phys. Rev. B* **54**, 11169–11186 (1996).
+
+44. Blöchl, P. E. Projector augmented-wave method. *Phys. Rev. B* **50**, 17953–17979 (1994).
+
+45. Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector augmented-wave method. *Phys. Rev. B* **59**, 1758–1775 (1999).
+
+46. Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. *Phys. Rev. Lett.* **77**, 3865–3868 (1996).
+
+47. Grimme, S., Antony, J., Ehrlich, S. & Krieg, H. A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu. *J. Chem. Phys.* **132**, 154104 (2010).
+Acknowledgments
+
+E.M. and R.P. acknowledge funding from the Swiss Nanoscience Institute (SNI), the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (ULTRADISS grant agreement No 834402 and supports as a part of NCCR SPIN, a National Centre of Competence (or Excellence) in Research, funded by the SNF (grant number 51NF40-180604). E.M., T.G. and S.-X.L. acknowledge the Sinergia Project funded by the SNF (CRSII5_213533). E.M., T.G. and R.P. acknowledge the SNF grant (200020_188445). T.G. acknowledges the FET-Open program (Q-AFM grant agreement No 828966) of the European Commission. J.-C.L. acknowledges funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement number 847471. C.L. and E.M. acknowledges the Georg H. Endress Foundation.
+
+Author information
+
+Authors and Affiliations
+
+Department of Physics, University of Basel, Klingelbergstrasse 82, 4056 Basel, Switzerland
+
+Outhmane Chahib, Chao Li, Jung-Ching Liu, Thilo Glatzel, Ernst Meyer & Rémy Pawlak
+
+State Key Laboratory of Precision Spectroscopy School of Physics and Electronic Science,
+East China Normal University, 500 Dongchuan Road, Shanghai 200241, China
+
+Yulin Yin & Qinghong Yuan,
+
+Faculty of Materials Science and Engineering/Institute of Technology for Carbon Neutrality, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China
+
+Feng Ding
+
+Contributions
+
+R.P. and E.M. conceived the experiments. O.C. and R.P. performed the STM/AFM measurements. Y.Y. F.D. and Q.Y. performed DFT calculations. O.C. and R.P. analyzed the data. R.P. wrote the manuscript. All authors discussed on the results and revised the manuscript.
+
+Corresponding authors
+
+Correspondence to ernst.meyer@unibas.ch or remy.pawlak@unibas.ch
+
+Ethics declarations
+
+Competing interests
+
+The authors declare no competing interests.
+Fig. 1: Atomic structure of phosphorus chains on Ag(111). a, STM topographic image after sublimation of phosphorus atoms on Ag(111) leading to P chains (1) and cyclo-P_5 domains (2), (I_T = 1 pA, V = 0.15 mV). The inset shows a STM image of the single, double and triple chains, respectively. b-d, Series of AFM images with CO-terminated tip revealing the armchair structure of single, double and triple P chains, (f_0 = 26 kHz, A = 50 pm). Scale bars are 1 nm. e, Atomic configurations of the triple armchair chains obtained by DFT calculations. Phosphorus and silver atoms are shown in orange and gray, respectively. f, Corresponding AFM simulation using the DFT coordinates.
+Fig. 2: Atomic structure of the self-assembled cyclo-\( P_5 \) molecules. **a**, STM image of the self-assembled pentamers on Ag(111), (\( I_T = 1 \) pA, \( V = 0.15 \) mV). Islands systematically shows a superlattice of bright lines rotated by 19° with respect to the [1\(\overline{1}0\)] directions of Ag(111). **b**, Close-up STM topography showing the \( P_5 \) molecules depicted by dashed pentagons. **c**, Corresponding AFM image revealing the \( P_5 \) chemical structure, (\( f_0 = 26 \) kHz, \( A = 50 \) pm). **d**, Atomic configurations of the pentamer assembly on Ag(111) obtained by DFT. Phosphorus and silver atoms are shown in orange and gray, respectively. **e**, Corresponding AFM simulation using the DFT coordinates. **f**, Site-dependent \( \Delta f(Z) \) spectroscopic curves acquired at one P atoms of a \( P_5 \) molecule (orange), between two \( P_5 \) molecules (brown) and on Ag(111) (black), respectively. The local minima of the \( \Delta f(Z) \) curves indicate the relative height of the phosphorus atoms.
+Fig. 3: Charge redistribution at the cyclo-P5/Ag(111) interface. a, Frequency shift \( \Delta f \) as a function of sample bias voltage \( V_s \), measured across a pentamer domain shown in the STM image (top), (parameters : \( f_0 = 26 \) kHz, \( A = 80 \) pm). b, Single \( \Delta f(V) \) curves at the pentamer assembly (orange) as compared to the Ag(111) (black). Dashed lines mark the top of the parabola allowing to extract a LCPD shift \( \Delta V^* = 0.22 \) V. c, Top and side views of the charge redistribution between pentamers and Ag(111). Blue areas show electron accumulation, red areas electron depletion. The isosurface level of the plot is set to \( \pm 13 \times 10^{-3} \) e/\AA\(^3\). d, Schematic illustration of the charge redistribution at the \( P_5 / \mathrm{Ag}(111) \) interface leading to an inward surface dipole (\( D \)) moment and a local work function change (\( \phi_{P_5/\mathrm{Ag}} \)). The cyclo-P5 layer is colored in orange. \( \Delta V^* \) refers to the LCPD change.
+Fig. 4: Tunneling spectroscopy of the \( P_5 \)/Ag interface. **a**, dI/dV point-spectra acquired above the \( P_5 \) assembly (orange) and on Ag(111) (black), where precise locations are shown in the STM inset. (parameters: \( I_t = 1 \) pA, \( V_s = 500 \) mV, \( A_{mod} = 10 \) mV, \( f = 511 \) Hz). **b**, dI/dV maps at \( V_s = -1.25 \) and 2.5 V corresponding to the valence band energy and the IS interface state , respectively. **c**, STM topographic image of three \( P_5 \) domains and the corresponding dI/dV maps of the IS modulation. **d**, Scheme of the band alignment and the formation of Stark-shifted IPS (orange lines). **e**, Field-effect resonance tunneling (FERT cross-section acquired across the \( P_5 \) assembly along the dashed line in **a**, (Set-points: \( I_t = 1 \) pA, \( V_s = 500 \) mV, \( A_{mod} = 35 \) mV, \( f = 511 \) Hz). **f**, Single FERT spectra of the \( P_5 \) assembly and the Ag(111) substrate, showing the series of n\(^{th}\) IPS. **g**, Extracted IPS peak voltages as a function of \( n^{2/3} \).

039b62c7612baea1877b62435751c729d645b0b0e721961591a8e7584ee5230b/preprint/preprint.md

@@ -0,0 +1,337 @@
+Rapid incidence estimation from SARS-CoV-2 genomes reveals decreased case detection in Europe during summer 2020
+
+Maureen Smith
+Robert Koch Institute
+Maria Trofimova
+Robert Koch Institute
+Ariane Weber
+Max-Planck Institute
+Yannick Duport
+Robert Koch Institute
+Denise Kühnert
+Department of Archaeogenetics, Max Planck Institute for the Science of Human History, 07745 Jena, Germany   https://orcid.org/0000-0002-5657-018X
+Max von Kleist (kleistm@rki.de)
+MF1 Bioinformatics, Robert Koch-Institute   https://orcid.org/0000-0001-6587-6394
+
+Article
+
+Keywords: SARS-CoV-2, epidemiology, genomes
+
+Posted Date: May 27th, 2021
+
+DOI: https://doi.org/10.21203/rs.3.rs-558667/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+Version of Record: A version of this preprint was published at Nature Communications on October 14th, 2021. See the published version at https://doi.org/10.1038/s41467-021-26267-y.
+Rapid incidence estimation from SARS-CoV-2 genomes reveals decreased case detection in Europe during summer 2020
+
+Maureen Rebecca Smith1,2,*,+, Maria Trofimova1,2,*, Ariane Weber3, Yannick Duport1,2, Denise Kühnert3,4, and Max von Kleist1,2,4,+
+
+1Systems Medicine of Infectious Disease (P5), Robert Koch Institute, Berlin, Germany
+2Bioinformatics (MF1), Robert Koch Institute, Berlin, Germany
+3Transmission, Infection, Diversification and Evolution Group, Max-Planck Institute for the Science of Human History, Jena, Germany
+4German COVID Omics Initiative (deCOI)
+*these authors contributed equally to this work
++smithm@rki.de
++kleistm@rki.de
+
+ABSTRACT
+
+By May 2021, over 160 million SARS-CoV-2 diagnoses have been reported worldwide. Yet, the true number of infections is unknown and believed to exceed the reported numbers by several fold. National testing policies, in particular, can strongly affect the proportion of undetected cases.
+
+Here, we propose a novel method (GInPipe) that reconstructs SARS-CoV-2 incidence profiles within minutes, solely from publicly available, time-stamped viral genomes. We validated GInPipe against in silico generated outbreak data and elaborate phylodynamic analyses. We apply the method to reconstruct incidence histories from sequence data for Denmark, Scotland, Switzerland, and Victoria (Australia). GInPipe reconstructs the different pandemic waves robustly and remarkably accurate. We demonstrate how the method can be used to investigate the effects of changing testing policies on the probability to diagnose and report infected individuals. Specifically, we find that under-reporting was highest in mid 2020 in parts of Europe, coinciding with changes towards more liberal testing policies at times of low testing capacities.
+
+Due to the increased use of real-time sequencing, it is envisaged that GInPipe can complement established surveillance tools to monitor the SARS-CoV-2 pandemic. We anticipate that the method is particularly useful in settings where diagnostic and reporting infrastructures are insufficient. In ‘post-pandemic’ times, when diagnostic efforts are decreased, GInPipe may facilitate the detection of hidden infection dynamics.
+
+Introduction
+
+As of May 2021, the global SARS-CoV-2 pandemic is still ongoing in most parts of the world, with 160 million reported cases worldwide. Novel vaccines of high efficacy have been developed within a year of the outbreak [2, 46]. At the time of writing, approximately 8.2% of the worlds population had already received at least one vaccination. However, distribution of vaccines is uneven and achieving global herd immunity may pose an extremely difficult, long-term task [63, 36]. At the same time, novel variants of concern (VOC) have emerged in high prevalence regions [6, 34], which may be able to reinfect individuals [21, 37] and escape vaccine elicited immune responses [33, 66, 45]. For example, Manaus, Brazil, witnessed a massive second wave of infections [51], despite the fact that approx. 80% had already experienced an infection at the onset of the second wave [6]. Because of the evolutionary versatility of SARS-CoV-2 and difficulties in global vaccine distribution, some experts expect that the virus may not be eliminated globally [44]. Even without adaptation to vaccines in the future, it has been postulated that SARS-CoV-2 may resurge [24, 50] and surveillance may have to be maintained into the mid 2020s to monitor virus spread and evolution [24].
+
+Currently, the gold standard of SARS-CoV-2 surveillance is diagnostic testing via polymerase chain reaction (PCR) or antigen-based rapid diagnostic testing (RDT). Diagnostic test results currently define infection case reports, which are used to survey
+epidemiological dynamics and to define thresholds for travel bans and non-pharmaceutical measures. Inevitably, case reporting data is affected by test coverage, which changes when testing policies are adapted. While RDT enables point-of-care diagnosis and is less costly than PCR testing [13, 12], gathering and reporting of test results still requires a sophisticated infrastructure, which is difficult to establish and maintain in many developing countries [35]. Independent and complimentary sources of information, such as social media reports [31, 53] or waste water analysis [9, 43] have been used early on to complement our knowledge of the pandemic dynamics. In addition, many regions of the world sequence SARS-CoV-2 genomes to track virus evolution and the emergence of variants of concern. The gathered viral sequences are regularly provided to public databases, such as GISAID [14, 54]. We hypothesize that the genetic data alone holds information about the pandemic trajectory. More specifically, we presume that the speed at which SARS-CoV-2 evolves on the population level contains information about the number of individuals who are actively infected.
+In the vast majority of cases, SARS-CoV-2 is transmitted within a very short period, only days after infection [30, 17]. The consequence is a well-defined duration of intra-patient evolutionary time before transmission. Thus, the number of infected individuals is correlated to the rate of divergence of the viral population, implicating an ‘evolutionary signal’.
+In this article, we introduce the computational pipeline GInPipe, which only uses time-stamped sequencing data, extracts the ‘evolutionary signal’ and reconstructs SARS-CoV-2 incidence histories. The approach builds on recent work by Khatri and Burt [23], who derived a simple function that relates the mean number of mutant origins to the current allele frequency and the mutational input, which is proportional to the effective population size. Herein, due to the short window of transmission, we anticipate that the effective population size may strongly correlate with the incidence of SARS-CoV-2. We adapt the function derived in [23] and embed it into an automatic computational pipeline (GInPipe) that reconstructs the time course of an incidence correlate \( \phi \) merely from SARS-CoV-2 genetic data. GInPipe is validated threefold and performs robustly: (i) against *in silico* generated outbreak data, (ii) against phylodynamic analysis and (iii) in comparison with case reporting data. We applied the method to SARS-CoV-2 sequencing data from Denmark, Scotland, Switzerland, and the Australian state Victoria to reconstruct their respective incidence histories. Lastly, we utilize the inferred epidemic trajectories to compute changes in the probability that an infected individual is reported and highlight how this probability is affected by changes in testing policies.
+
+Results
+
+Incidence reconstruction
+An outline of GInPipe for SARS-CoV-2 incidence reconstruction is shown in Figure 1A-C. After compiling a set of time-stamped, full-length SARS-CoV-2 genomes, the sequences are placed into temporal bins \( b \) (Fig. 1A). For each bin, we compute the number of mutant sequences \( m_b \), as well as the number of haplotypes \( h_b \). These two inputs are used to infer the incidence correlate \( \phi_b \) (Fig. 1B). We then smooth over all \( \phi_b \) point estimates and derive a reconstructed incidence history along the time axis (Fig. 1C). The reconstructed incidence histories can then be used as a basis to estimate the effective reproduction number \( R_e \), as well as the relative case detection rate as outlined below.
+
+Method validation: *in silico* experiment
+To test whether GInPipe correctly reconstructs incidence histories, we first performed an *in silico* experiment. We considered a population of \( N(t) \) infected individuals at time \( t \) that stochastically generate \( N(t+1) \) infected individuals in the next time step \( t+1 \). Each individual is associated with a virus sequence, which can mutate randomly. Individuals can be removed (the associated sequence is removed), or they transmit their virus (the associated virus is copied over). We record the number of infected individuals per generation, as well as all sequences of the currently circulating viruses. We then use the simulated viral sequences to infer \( \phi(t) \) and reconstruct the incidence history, as presented in Figure 1D-E.
+In Figure 1D, we compare one trajectory of simulated population sizes with the reconstructed incidence histories. The simulated outbreak (red line, right axis) consists of two waves of increasing magnitude. GInPipe reconstructs these dynamics (blue lines and dots, left axis) quite accurately, although the incidence correlate \( \phi(t) \) is on a different scale, implying a linear correlation to the number of infected individuals. To assess this correlation, we performed 10 stochastic simulations and compared the \( \phi(t) \) point estimates with the corresponding number of infected individuals (Fig. 1E). We observed a strong (\( r = 0.96 \)) and highly significant (\( p < 10^{-16} \)) linear relationship between the number of infected individuals \( N(t) \) and the method’s incidence correlate \( \phi(t) \).
+While these simulations represent idealized scenarios, we evaluated the robustness of GInPipe with regards to incomplete, and sparse data sets, thoroughly elaborated in Supplementary Note 1.
+Our analyses showed, that the method can still accurately reconstruct incidence histories over time, when data is missing or when data sampling is unbalanced. In scenarios of extreme under-sampling, the \( \phi \) point estimates are prone to slight underestimation. However, through the smoothing step the reconstructed incidence trajectories still follow the overall population dynamics (Suppl. Note 1, section SN.1.7). Finally, we evaluated whether introductions of foreign sequences affect the reconstruction of incidence histories. Even for extreme and unrealistic cases, a stable reconstruction of the underlying dynamic is possible, but
+Figure 1. Reconstruction of incidence histories using the proposed method. A–C Schematic of the incidence reconstruction method. A The sequences are chronologically ordered by collection date. The line shows the cumulative sum of sequences over time. The sequences are allocated into temporal bins, spanning either the same time frame \( \Delta d_b \) (yellow and purple bins) or containing the same amount of sequences (green bins). B For each bin, the number of distinct variants \( h_b \), as well as the total amount of mutant sequences \( m_b \), are used to infer the incidence correlate \( \phi_b \). C The point estimates for all bins \( \phi_b \) (dots) are smoothed with a convolution filter. For uncertainty estimation, the point estimates are sub-sampled and interpolated. D–E Reconstruction of a simulated outbreak with GInPipe. D \( \phi \) estimates resemble the underlying population dynamics over time. The blue line shows the smoothed median of the sub-sampled \( \phi \) estimates (dots) for a simulated outbreak. The red line indicates true incidence per generation. E. Dotplot showing the true outbreak size from the simulation \( N_{true} \) versus the \( \phi_b \) point estimates for 10 stochastic simulations. The red line depicts the linear fit.
+
+we do observe a slight tendency of overestimation in these extreme cases (Suppl. Note 1, section SN.1.8).
+
+Method validation: phylodynamics
+Phylodynamic methods combine phylogeny reconstruction with epidemic models. For example, the piecewise constant birth-death sampling process [55] implemented in BEAST2 [5], allows the reconstruction of the effective reproduction numbers \( R_e(\tau) \) for given time periods \( \tau \). However, these methods are computationally expensive, so that only moderately sized sequence sets can be used, and advanced knowledge is required to apply them properly to larger data sets.
+We conducted phylodynamic analyses of SARS-CoV-2 sequence data from Denmark, Scotland, Switzerland, and the Australian state Victoria. In analyzing the data we assumed that \( R_e^{\text{BEAST}}(\tau) \) was piecewise constant in between major changes in SARS-CoV-2 non-pharmaceutical interventions (intervals stated in Supplementary Note 2). We then used BEAST2 to estimate
+\( R_e^{\text{BEAST}}(\tau) \) alongside the tree reconstructions.
+
+In parallel, we estimated corresponding effective reproduction numbers \( R_e^\phi(t) \) by applying the Wallinga-Teunis method [61] to incidence correlates \( \phi \) derived by GInPipe. For both methods, we used publicly available full length SARS-CoV-2 sequencing data from GISAID [14, 54](Supplementary Note 4).
+
+Results of both methods are shown in Figure 2. Overall, both methods show congruent trends for the analyzed countries, when comparing the piecewise constant \( R_e^{\text{BEAST}}(\tau) \) from phylodynamic analysis with the median daily \( R_e^\phi(t) \) for the same interval. Noteworthy, GInPipe allows for a much finer time-resolution (daily \( R_e \) estimates) compared to the piecewise constant \( R_e \) estimates on pre-defined intervals, obtained from the phylodynamic analysis.
+
+For Denmark, the first interval spans the decline in the number of infections after the first wave (end of April to mid June). Consequently, we observe \( R_e(\tau) < 1 \) using both methods. For the next intervals, the median or *piece-wise constant* \( R_e(\tau) \) is predicted to be around, or slightly larger than one. However, GInPipe reconstructs a number of peaks in the daily \( R_e^\phi(t) \) estimates, most pronounced in August, coinciding with the summer holidays in Europe. In the interval from November to mid December the estimates deviate slightly, with a larger median estimate from BEAST2, however, both interval estimates are predicted to be \( R_e(t) > 1 \) and the confidence intervals overlap entirely.
+
+The \( R_e(\tau) \) estimates for Scotland agree almost exactly, where GInPipe again allows for a much finer time-resolution. Once again, we see a peak in the summer (August–September 2020), coinciding with the summer holidays in Europe. For the last interval (from December 2020) both methods show a median \( R_e(t) > 1 \), again with a slightly higher median BEAST2 estimate, coinciding with the second wave of infections.
+
+For Switzerland, the estimates disagree slightly, particularly in the first interval (mid March to mid May), which spans both sides of the peak number of infections during the first wave. Although both methods predict a median \( R_e(\tau) < 1 \), the absolute value differs in magnitude between the two methods, with BEAST2 estimating a much lower value. The lower estimate from the BEAST2-analysis in the first interval may be explained by the approximation of transmission clusters, which results in the reconstruction of a relatively high number of transmission events many of which may have occurred outside Switzerland (Supplementary Note 2, Figure SN.12 therein, tree B.1). In the daily estimates, we see a transition from \( R_e^\phi(t) > 1 \) to \( R_e^\phi(t) < 1 \), which may explain why the *median* prediction with GInPipe is close to one for the entire interval. The estimates are qualitatively different for the second interval (mid May – mid June), where GInPipe estimates \( R_e^\phi(\tau) < 1 \), while BEAST2 estimates \( R_e^{\text{BEAST}}(\tau) \approx 1 \). Again, GInPipe estimates a peak in summer (mid June–mid August \( R_e\phi(\tau) > 1 \)). While BEAST2 predicts the onset of transmission in the second wave to already start in mid August (\( R_e(\tau) > 1 \)), GInPipe estimates the first major rise in infections at the end of September.
+
+For Victoria we observe an \( R_e^\phi(t) > 1 \) until mid March in the daily estimates. Overall, \( R_e \) is less than 1 for the first interval between mid March and May, versus \( R_e > 1 \) between June and August. Again, we see various peaks around June and July in the daily \( R_e \) estimates with the proposed method. For the final interval, both methods slightly disagree, with \( R_e^{\text{BEAST}} < 1 \) and \( R_e^\phi(\tau) > 1 \), though the daily \( R_e^\phi(t) \) are decreasing towards the end of the final interval.
+
+In terms of computational time, the entire GInPipe analysis pipeline runs in 20 minutes on the full Denmark data set (n = 40,575 sequences) and in 7 minutes on the Victoria data set (n = 10,710 sequences) on a single notebook (2.3 Ghz, 2 cores). Furthermore, GInPipe does not require to pre-assign any intervals, to exclude particular strains, construct a phylogenetic tree, or cluster sequences based on a their phylogenetic relationship. The BEAST2 analysis alone required about 15 hours on an Intel Xeon E5-2687W (3.1 Ghz, 2 x 12 cores) on a sub-sampled data set (\( n \approx 2500 \) sequences) with additional computation time needed to construct a multiple sequence alignment and approximate transmission clusters.
+
+Reconstructed incidence histories
+We used GInPipe to reconstruct complete incidence histories for Denmark, Scotland, Switzerland, and Victoria (Australia) from publicly available full length SARS-CoV-2 sequencing data provided through GISAID [14, 54] (Supplementary Note 4).
+In Figure 3, we compare the reconstructed incidence histories (blue lines and dots, left axis) to the 7-day rolling average of officially reported new cases (red line, right axis). Overall, the reconstructed incidence estimates reflect the different pandemic waves deduced from the reporting data, although there are quantitative differences between the reconstructed and reported incidence trajectories over time. In particular, during the first wave in Scotland, and Victoria (Fig. 3B,D) our method estimates higher incidences than reported, whereas the curves align at later points for the second and third wave. It is worth mentioning that testing capacities were particularly low in Scotland in April (during the first wave), suggesting extensive under-reporting in the initial phase of the pandemic. This is also supported by test positive rates of almost 40% during April 2020 in Scotland (Supplementary Fig. 1). In Victoria, sufficient testing capacities were not available until May, but test positive rates were already declining from April to May (Supplementary Fig. 1). This indicates that the first wave may have been under-reported in magnitude, but had vanished by May.
+Interestingly, the proposed incidence reconstruction method predicts small summer waves in August in the three European countries (Fig. 3A–C) that are not visible in the reporting data. In the incidence reconstruction method these ‘summer waves’
+Figure 2. Effective reproduction number \( R_e \) estimates using the proposed method (\( \phi \)) and phylodynamics (BEAST2). Piecewise constant \( R_e^{\text{BEAST}}(\tau) \) estimates (green solid lines) where calculated using the BDSKY model for the indicated intervals, as described in the Methods section. Daily estimates \( R_e^\phi(t) \) (blue dots) were directly calculated from the incidence correlates \( \phi \) using the Wallinga-Teunis method [61]. The median of these values for the indicated intervals \( R_e^\phi(\tau) \) is shown as solid blue lines. The 95% confidence interval is specified by the shaded areas. Justifications of the intervals are found in Supplementary Note 2.
+
+are immediately followed by the second SARS-CoV-2 wave. For the second wave, reconstructed incidence histories correspond to the reported cases, particularly in Denmark, Scotland, and Victoria. (Fig. 3A-B & D). For Scotland, our method predicts a more long-lasting third wave with rising incidence rates until February 2021 and a moderate decline with several smaller peaks until May, whereas the reporting data indicates a peak in January 2021 with a subsequent fast regression. The argument, that ongoing vaccination in Great Britain could explain the immediate decline of reported infected cases, can be objected with the fact, that by March 2021 (end of the prediction horizon) only about 2% of the Scottish population were fully vaccinated. For Switzerland, we predict a larger wave around January-February 2021 (third wave) that is not reflected in the reporting data. Towards the end of the prediction horizon, from March 2021 onwards, the reported cases and the incidence estimation both indicate a rise in numbers (fourth wave).
+
+Relative case detection rate
+We investigated whether the proposed incidence reconstruction method may be used to learn about the proportion of infected cases that are actually tested, detected and reported, \( P_t \) (tested|infected).
+The proportion of SARS-CoV-2 infected who are actually reported can be calculated using Bayes’ formula (see Methods section). In order to perform the calculation, the proportion of actively infected individuals in the population \( P_t \) (infected) needs to be known. We have shown that the incidence correlates \( \phi \) from our method are proportional to the number of infected individuals, \( c \cdot \phi_t = N_{\text{eff}} \) (Fig. 1D–E, Fig. 3), and hence to the probability of being infected \( P_t \) (infected). Consequently, we may use the reconstructed incidence profiles, together with the test sensitivity and specificity, the respective information about the proportion of positive tests, as well as the testing capacities for each country or region to calculate changes in the case detection rate, scaled by unknown factor \( c \).
+In Figure 4, we show the \( \log_2 \) scaled detection probabilities for Denmark, Scotland, Switzerland, and Victoria (Australia).
+Figure 3. Incidence reconstruction based on sequencing data. The graphic depicts the genome-based incidence reconstruction (in blue) using the proposed method (left axis) vs. the 7 days rolling average of newly reported cases in red (right axis). Blue dots depict \( \phi_b \) point estimates of the incidence correlate, where the size of the dot is related to the number of sequences used to infer \( \phi_b \). The solid and dashed blue lines denote the median smoothed trajectories and their 5th and 95th percentiles. The black markers on the x-axis depict the collected sequences at the given dates. **A** Denmark (n = 40,575 sequences) **B** Scotland (n = 30,258 sequences) **C** Switzerland (n = 25,779 sequences) **D** Victoria (n = 10,710 sequences)
+
+The log scaling allows us to easily gauge the relative change in (under-)detection of the infected population over time (e.g. 2-fold, 4-fold increase or decrease in case detection rate). The dashed vertical lines in the graphics indicate major changes in testing policies in the respective countries. Individual parameters used in the inference procedure, \( P(\text{tested}), P(\text{inf|tested}), \) and \( c \cdot P(\text{infected}) \) are shown in Supplementary Figure 1.
+
+For Denmark, we observe an initial period of massive SARS-CoV-2 under-detection in the beginning of March 2020, Fig. 4A (upper panel), which coincides with very low testing capacities at the beginning of the pandemic (Fig. 4A, lower panel). From mid March, case detection stabilizes at a 6-fold higher level, compared to the first week of March. The second interval begins around mid May with an important policy change, allowing every citizen to get tested without medical referral. Interestingly, compared to the fairly stable case detection levels from mid March to mid May, this policy change leads to a 2-3 fold drop in case detection in the summer months from July-September. Of note, while everybody is granted the possibility to test for SARS-CoV-2, testing capacities remained fairly unchanged (Fig. 4A, lower panel). According to our calculations, the largest proportion of infections remained undetected in July. From end of August, testing capacities were steadily increased in Denmark (Fig. 4A, lower panel), particularly in Copenhagen and at the airports, followed by prioritized testing. From September on, this leads to a nearly 8-fold increase of the case detection rate, with a peak in December. From end of December the detection rate drops more than 4-fold, despite continuous testing.
+
+For Scotland (Fig. 4B), the earliest test data is available only from the end of March. Therefore, the data captures only the second part of the first wave, compare Fig. 3B. In the beginning of May, testing capacities were more than doubled (Fig. 3B, lower panel) and outbreak investigation intensified. This led to a doubling of the relative case detection rate from May, compared to the first phase. On 18 May, SARS-CoV-2 testing was opened for everyone with symptoms. However, only in July testing capacities were increased. This may have led to a drop in case detection from mid May to July, after which case detection increased and remained during August at roughly the levels achieved in May. After 25 August, testing capacities
+Figure 4. Relative case detection rate. Black line in upper graphics: Estimated and scaled probability of detecting SARS-CoV-2 infected individuals \( c \cdot P(\text{tested}|inf) \). Blue line in the lower graphics: Number of conducted tests per calendar week. Dashed vertical lines indicate major changes in the testing strategies in the respective location. The sources for testing data and strategies are given in Supplementary Note 3. **A** Denmark. Policy changes: 18 May 20: testing for everyone; 9 September 20: increasing testing available **B** Scotland. Policy changes: 1 May 20: expanded testing strategy including enhanced outbreak investigation; 18 May 20: testing for everyone with symptoms; 22 July 20: including young children for testing; 25 August 20: increasing capacity and accessibility of testing; 25 November 20: expansion of testing in health care; 15 December 20: increase of testing capacity; 1 January 21: community testing in areas with high coronavirus prevalence. **C** Switzerland. Policy changes: 18 May 20: priority testing; 2 November 20: rapid antigen tests are included in the testing strategy; 27 February 21: recommended preventative and repeated testing as part of precautionary measures. **D** Victoria (Australia). Policy changes: 14 April 20: anyone having symptoms can be tested; 30 April 20: start of 2 weeks 'testing blitz'; 11 May 20: increased surveillance with testing of sewerage; 1 July 20: expanded 'testing blitzes' in outbreak regions; 30 December 21: urging to be tested after re-emergence of positive cases.
+
+and accessibility of testing steadily increased. Accordingly, case detection increased about 6-fold until winter 20/21. From 25 November, testing capacities were further expanded, especially in the health sector, including hospital patients, health and social care staff, with fairly stable case detection rates. Further increase of testing capacities in the end of December allowed to double the probability to detect infected individuals. From beginning of the year 2021, the Scottish government pushed community testing in areas with high SARS-CoV-2 prevalence. At the same time, the proportion of positive tests start to decline (Suppl. Fig. 1), and consequently the case detection rate collapses until April by 9-fold.
+Similar to Denmark, Switzerland shows an initial period of massive SARS-CoV-2 under-detection in the beginning of March 2020 (Fig. 4C, upper panel), which coincides with very low testing capacities at the beginning of the pandemic (Fig. 4C, lower panel). When testing capacities increase by mid March, case detection rates grow 8-fold. However, from beginning of April, we observe drop in the probability to detect infections that lasts until mid May (overall 10-fold drop). This trend coincides with a drop of positivity rates (Supplementary Figure 1), as well as the extension of testing criteria on 22nd April: From this date, anybody with symptoms was allowed to get tested, despite the fact that the availability of tests was not increased (Fig. 4C, lower panel). From 18th May, tests were partly prioritized for hospitalized and vulnerable individuals. At the same time, testing capacities steadily increased and incidences dropped. As a net effect, the probability of detecting infected people increases steadily to a maximum at the end of October with a relative difference of nearly 20-fold compared to the low point in mid May. On 2 November, Switzerland begins to supply antigen-based rapid diagnostic tests (RDT) for self-testing as part of their COVID containment strategy. Interestingly, our model predicts that this led to a sharp decline in case detection, again corresponding with the decline in positivity rates (Supplementary Figure 1). From 21st February 2021, further precautionary actions were taken, and the government recommended repeated testing. This is associated with a stable, but relatively low detection rate for infected people until end of April 2021.
+For the Australian state Victoria, the earliest data were available from end of March 2020, Fig. 4D, capturing the second part of the first SARS-CoV-2 wave. Detection probabilities in the first interval, until 14th April were changed proportionally to the test capacities during that interval Fig. 4D (upper and lower panel). On 14th April 2020, the testing criteria were expanded, allowing anyone with COVID-like symptoms to be tested. Unlike the situation in Switzerland, where we observed a downward trend in case detection after expanding the testing criteria (Fig. 4C), the detection probability in Victoria remains stable until end of April. In contrast to Switzerland, testing capacities were increased when testing criteria were expanded. On 30th April, the government initiated a two-week ‘testing blitz’, a large, coordinated testing campaign to locate viral spread. The ‘testing blitz’ was accompanied by mass sewerage testing and matched with a massive increase of testing capacities, which led, according to our simulations, to a 4-fold increase in the probability to detect infected individuals. At the end of the ‘testing blitz’, testing capacities steadily decreased and the proportion of detected infections decreased drastically (by roughly 9-fold). At the beginning of June, testing capacities rose again, matched by a rise in the proportion of detected cases. From 1st July onwards, several ‘testing blitzes’ were conducted in outbreak regions, which seemed to have stabilized case detection rates during the second wave of infections. After the second wave (end of August–September, Fig. 3D), case detection rates drop. From October 2020 onwards, our predictions become highly unreliable, as the incidence estimates credibility interval includes zero (compare Fig. 3D), which concludes that the case detection rate cannot be determined anymore.
+In general, we make two striking observations: Firstly, and quite intuitively, whenever more tests were conducted, the proportion of detected SARS-CoV-2 cases increases. Secondly, and unexpectedly, whenever testing criteria were relaxed, this led to a drop in the probability of case detection. We see this drop in mid May in Denmark and Scotland and in mid April in Switzerland. Importantly, the expansions of testing criteria were not-, or insufficiently matched by increased testing capacities. Quite surprisingly, our simulations for Switzerland suggested a drop in case detection when antigen-based RDT self-testing became part of the national diagnostic strategies.
+
+Discussion
+
+SARS-CoV-2 continues to spread around the world, making epidemiological and molecular surveillance indispensable for the evaluation and guidance of public health interventions.
+Many national and international sequencing efforts are underway that closely monitor the dynamics and evolution of the virus. In the global fight against SARS-CoV-2, the vast majority of reconstructed sequence data has been made broadly available through public databases, such as GISAID [14, 54] and the COVID data portal. In this work, we introduce GInPipe, a pipeline that utilizes this data to reconstruct SARS-CoV-2 incidence histories.
+Viral infections are often characterized by a transmission bottleneck [32], where only a very small number of viruses initiate the infection and subsequently replicate within the host. A sufficient number of viruses (viral load) is required for further transmission. Hence, the temporal window of infectiousness begins with the intra-host viral population reaching a sufficiently large abundance and ends with the virus becoming eliminated by the immune system (or drugs). In contrast to HIV or HCV, SARS-CoV-2 is almost always transmitted within days after infection [30, 17]. If neutral or favourable mutations occur during this time, they may become abundant enough to be passed on to other hosts [32]. The consequence is a well-defined duration of intra-patient evolutionary time in which the virus can randomly mutate and become transmitted subsequently. In SARS-CoV-2, this intra-patient evolutionary time appears to be short and the analysis of outbreak clusters indicates that the virus genomes from linked cases were separated by either none, or very few mutations [4, 18, 52]. Taken together, these lines of evidence suggest that evolutionary change of SARS-CoV-2, the effective viral population size, and the number of infected people are correlated.
+In the past, numerous approaches have been published, with the aim to estimate the effective population size from genetic properties (reviewed in [62, 38]). A variety of methods utilize the information of temporal changes in allele frequency (reviewed in [62]), while others build on population genetic theory and phylodynamic reconstruction [16, 59, 27]. GInPipe is rather related to the first class of methods as it adapts recent works of Khatri and Burt, 2019 [23]. Essentially, GInPipe considers snap-shots of inter-patient evolution to estimate a mutational input parameter \( \phi(t) \). The latter is proportional to the effective population size, which correlates with incidence. Taken together, GInPipe uses time-stamped SARS-CoV-2 sequences and divides them into bins of inter-patient virus evolution to estimate time-dependent incidence correlates \( \phi_b \). From the set of \( \phi_b \) estimates, the entire incidence history \( \phi(t) \) can be reconstructed.
+We assessed the suitability of GInPipe using in silico simulated outbreaks, in comparison with phylodynamics and by comparing to reported case statistics. Using simulated data, the method accurately reconstructed incidence histories (Fig. 1). It also performed robustly with incomplete data, and when foreign sequence variants were introduced ( Supplementary Note 1). The method even worked when the introduced variants made up a considerable fraction of the population and did not contribute to the mutational input of the outbreak.
+We also compared the method with epidemiological estimates from phylodynamic reconstruction using BDSKY [55] in BEAST2 [5], shown in Figure 2. Bayesian phylodynamic methods use Monte Carlo Markov Chain (MCMC) or similar
+techniques to allow for a Bayesian estimation of phylogenetic relatedness of genomes, by both estimating evolutionary parameters, as well as parameters governing an underlying epidemiological model [64, 60]. The MCMC sampling procedure makes phylodynamic inference computationally demanding and often requires to ‘down-sample’ data sets.
+When the epidemiological model entails time-varying parameters, changes in the effective reproduction numbers \( R_e(\tau) \) can be computed. However, to enable their estimation (practical parameter identifiability), parameters of the underlying epidemiological model are typically considered to be piecewise constant or to change smoothly. In Figure 2, we show the phylodynamic estimates of the effective reproduction numbers \( R_e^{\text{BEAST}}(\tau) \). Corresponding reproductive numbers \( R_e^{\phi}(\tau) \) were computed with GInPipe by applying the method of Wallinga-Teunis [61] to the estimated incidence correlates \( \phi \). We compared the medians over the temporal windows used in the phylodynamic analysis. Overall, this methodological comparison yielded highly congruent predictions, with the exception of Switzerland in the first- (mid March – May 2020) and final intervals (mid September 2020 – January 2021). The ETH Zurich provides a visualization\(^1\) for the daily \( R_e \) estimates, based on reporting data. The ETH data, similarly to our daily \( R_e^{\phi} \) estimates with GInPipe, shows a peak, followed by a decline in the daily \( R_e \) for the first interval. This could explain why the median \( R_e^{\phi} \) is only slightly smaller than 1 in this first interval, unlike the BEAST2 estimate, which is \( \approx 0.6 \). For the final intervals (mid September 2020 – January 2021) \( R_e^{\phi} \) estimates fluctuate around- or slightly above \( R_e(t)=1 \), in line with the predictions of the ETH, and slightly below the BEAST2 estimate that resulted in a median \( R_e \) around 1.2. For the sake of this comparison, a relatively crude transmission cluster detection method was employed for the phylodynamic analyses, which may be causing a slight bias in the estimated effective reproduction numbers.
+Overall, it appears that both methods yield similar results with respect to inferring the pandemic trajectories in the majority of cases. The power of GInPipe lies in the swift reconstruction of incidence histories with a fine temporal resolution, without requiring phylodynamic inference, construction of a multiple sequence alignment, down-sampling, clustering by e.g. lineages, or masking of problematic sites in the virus genomes. Moreover, GInPipe performs robustly, even in case of large proportions of introduced variants, which would also include lab-specific errors (Supplementary Note 1). However, \( R_e \) estimation is obviously only a side-product of phylodynamic inference, which has many more applications such as the identification and analysis of transmission clusters [19, 47], which GInPipe is not suited for. Hence, the two approaches could complement one another.
+To simplify the use of GInPipe, we provide an automatic workflow that can be directly applied to data downloads from GISAID or the COVID Data Portal. The execution time appears to scale linearly with the number of sequences to be analyzed (\( \approx 1,500 \) sequences per minute on a 2,3 Ghz computer with 2 cores).
+When we applied GInPipe to available GISAID data from Denmark, Scotland, Switzerland, and Victoria (Australia), we observed that the reconstructed incidence histories agree well with the daily numbers of new reported infections (Fig. 3). Particularly for Denmark, reconstructed incidence histories match the reporting data quite well. Of the analyzed countries, Denmark conducted the largest number of SARS-CoV-2 tests per capita (see also \( P(\text{tested}) \) in Supplementary Figure 1). This could imply that the pandemic was relatively well tracked, as also suggested by relatively small changes in the diagnostic rate (Fig. 4). Moreover, a large fraction of the diagnosed cases were sequenced, providing a comprehensive genomic profile of the virus population.
+For the first wave in Scotland and Victoria, we determined a much higher incidence than reported. Notably, the number of SARS-CoV-2 tests per capita was very low in Scotland, as well as in Victoria until May 2020 (\( P(\text{tested}) \) in Supplementary Figure 1). Thus, a large proportion of infected individuals may not have been diagnosed during this time. In Victoria and Scotland, testing capacities were increased in May, i.e. after the peak of the first wave.
+Another striking difference of our predictions in comparison to the reported cases is that GInPipe indicates a rise of infections in August 2020 in all European countries. Notably, this increase in infections coincides with the introduction and community spread of B.1.177 (the ‘Spanish’ variant, 20E (EU1)) in most Western European countries as suggested by phylodynamic analyses [28, 20]. Our results, when compared with the reported cases, therefore imply an under-reporting of cases during the onset of community transmission of B.1.177.
+Quantifying case detection is usually not feasible without knowing, or approximating the proportion of infected individuals (compare Eq. (2)). In order to do so, others have used mathematical models to predict the proportion of infected individuals [6, 1] and with this, to estimate the level of under-reporting of SARS-CoV-2. However, these mathematical models cannot be fitted to the reported cases under the presumption of an unknown trajectory of under-reporting. It therefore remains extremely difficult to parameterize suitable models for the task of assessing under-reporting, in particular for non-monotonic pandemic trajectories.
+Random testing may inform the number of incidents, as well as asymptomatic infections [41]. Yet, usually only snap-shots of the incidence may be derived, which are insufficient to parameterize the aforementioned models. Moreover, it is not clear, whether the samples in the random testing scheme were representative. Sero-prevalence studies remain the gold-standard to estimate the cumulative number of infections [6, 1], as well as cumulative under-detection. Nevertheless, these studies only
+
+\(^1\)https://ibz-shiny.ethz.ch/covid-19-re-international/
+provide very coarse time resolution (if any) and require large sample sizes for robust analysis.
+A methodologically related approach uses a semi-Bayesian approach to assess under-detection in the US [65]. To enable estimation, the probability of case detection is constrained by the assumption of particular prior distributions.
+With regards to the aforementioned approaches, our method to quantify case detecting profiles has the advantage that no complex mathematical modelling is needed, and no constraints are necessary. Instead, we use information about the conducted tests and the test positive rate, in combination with the incidence correlate \( \phi \). This makes the proposed approach simple, interpretable and independent of additional assumptions.
+Using this method, we observed that broad testing with little, or no suspicion of SARS-CoV-2 infection coincides with apparent under-reporting of infections from the second quarter of 2020. This coincides with a drastic decrease in the proportion of positive test results. From the latter, it is possible to compute the conditional probability that a tested person is actually infected (\( P(\text{inf} \mid \text{tested}) \), Supplementary Figure 1). A drop in \( P(\text{inf} \mid \text{tested}) \) coinciding with a steady amount of tests can negatively affect the probability to detect infected individuals \( P(\text{tested} \mid \text{inf}) \), which may have happened in the European summer of 2020.
+In other words, the scarce testing resources available during that time, may not have been employed in the most effective way. This suggests that it may be advisable to focus on testing symptomatic individuals when testing capacity is low.
+Nevertheless, the apparent under-reporting was overcome relatively quickly by either increasing testing capacities (Denmark, Scotland, Victoria) or re-focusing capacities or both (Switzerland), Fig. 4. Interestingly, our method predicts a decline in case detection in Switzerland after the broad introduction of antigen self-testing in November 2020. A potential explanation for this observation is that only a fraction of positive antigen self-tests is confirmed by PCR and hence enters the Swiss reporting system. At the time of writing, the final interpretation of this observation is still unclear and will require further analysis.
+In summary, we have developed a method that allows to reconstruct incidence histories solely based on time-stamped genetic sequences of SARS-CoV-2. We implemented the method in a fully automated workflow that can be applied to publicly available data. Moreover, this method can be used to assess the impact of testing strategies on case reporting. Finally, we envision that the method will be particularly useful to estimate the extent of the SARS-CoV-2 pandemic in regions where diagnostic surveillance is insufficient for monitoring, but may still yield a few samples for sequencing. In some of these regions pandemic control may be impossible or cause more harm than benefit [56] and hence these regions may constitute reservoirs for the emergence of novel SARS-CoV-2 variants. Gaining insight in the pandemic dynamics in these regions through alternative methods, such as GInPipe, could yield valuable information that helps to direct global SARS-CoV-2 control efforts.
+
+Methods
+
+Data and data pre-processing
+Sequences and meta data for Denmark, Scotland, Switzerland, and Victoria (Australia) were downloaded from the GISAID EpiCoV database [14, 54] (Supplementary Note 4). Sequences, where only the year of collection was provided were omitted. If year and month are specified, the 15th day of the month was added to the meta data.
+The retained sequences were individually mapped to the reference (NCBI Wuhan Reference Sequence: NC_045512.2 [39]) with minimap2 version 2.17 (r941) [29]. From the mapping files (SAM), we deduced the nucleotide substitutions for each sequence. Point mutations appearing less than three times in the whole data set were filtered out, as they may occur due to sequencing errors [58].
+
+Construction of temporal sequence bins
+SARS-CoV-2 sequences were sorted chronologically by collection date and assigned to temporal bins \( b \) in a redundant manner. We subdivided the sequence set into bins of
+
+• equal size (proportions of 2%, 5%, 7% of all samples)
+
+• spanning an equal amount of days (10, 15, and 20, and one calendar week).
+
+Bins that contain a proportion of sequences should however span at least 3 days and maximally 21 days, and bins that span a predefined time period should contain at least 15 sequences. The date assigned to a bin is the mean collection date of the comprised sequences.
+The redundant binning ('re-sampling') allows to evaluate cases where there is insufficient data along the time line (Figure 1A), and makes the proposed method statistically more robust to outliers.
+
+Incidence correlate \( \phi_b \)
+The proposed method is inspired by the work of Khatri and Burt, 2019 [23], who derived a simple relation between the mean number of independent origins of soft selective sweeps in a population sample \( \overline{\eta} \), the current number of an allele \( m \) and mutational input, i.e. the scaled (haploid) effective population size \( \theta = 2N_{\text{eff}} \mu \), with \( \mu \) denoting the mutation rate:
+\[
+\overline{\eta}(t) = \theta \log \left(1 + \frac{m}{\theta}\right).
+\]
+Unlike Khatri and Burt, who aim at estimating the recent effective population size utilizing the recurrent mutations which have been fixated in the population, we seek to reconstruct the history of incidences of a population over time. We adapted the equation accordingly, also under the presumption that the de novo occurrence of mutations is driven by random chance events, whose likelihood may increase with the number of infected individuals [25, 10]. Seeking to estimate the incidence correlate \( \phi = c \cdot N_{\text{eff}} \), with the incidence being equivalent to the effective population size \( N_{\text{eff}} \), scaled by a constant factor \( c \), we parameterize the equation as follows: For each temporal bin \( b \) we estimate incidence correlate \( \phi_b \) at time \( t_b \). From the sequences comprised in bin \( b \), i.e. dated within a certain time frame \( \Delta d_b \) (Fig 1A), we infer the number of haplotypes \( h_b \) and the total number of mutant sequences \( m_b \) in the bin (Fig 1B). The mutations are determined with respect to a given reference sequence. In the original equation, we replace the mean number of origins \( \overline{\eta} \) with the number of distinct variants \( h_b \). In each temporal bin, however, haplotypes and mutants are accumulated over the time span \( \Delta d_b \). To correct for biases that result from this accumulation, especially for large time spans, we normalize the inputs \( h_b \) and \( m_b \) using a logistic function \( w_b = (\log(\sqrt{\Delta d_b}) + 1)^{-1} \).
+
+The parameter \( \phi_b \) is derived by numerically solving
+
+\[
+\phi_b^* = \arg\min_{\phi_b} \quad h_b \cdot w_b - \phi_b \log \left(1 + \frac{m_b \cdot w_b}{\phi_b}\right).
+\]
+
+Reconstructing the incidence history
+Incidence point estimates \( \phi_b \) are assigned to the mean collection date \( t_b \) of the sequences contained in the bin. We applied a convolution filter with window size 7 days to derive a continuous, smoothed trajectory (Fig. 1C). For uncertainty estimation, we sub-sampled \( \phi \) trajectories 1000 times, by randomly leaving out 50% of the point estimates and reconstructed the trajectory by smoothing and linear interpolation between the remaining point estimates.
+
+Implementation and availability
+All methods were implemented in Python version 3.9 and R version 4.0. A fully automated workflow has been generated using Snakemake [26] and is available from https://github.com/KleistLab/GInPipe.
+
+Simulation study
+To test the proposed incidence reconstruction method, we stochastically simulated the evolutionary dynamics of a viral outbreak using a Poisson process formalism. We started with \( N(t_0) = 50 \) copies of a random sequence of length \( L = 200 \) nt, that evolved in 120 discrete time steps, depending on a population dynamic. A succeeding generation was modelled to consist of \( N(t+1) \sim Poiss(N(t) \cdot \rho(t)) \) sequences (= effective population size), where we chose a sinodial rate \( \rho(t) = \frac{\sin(t/0.11)}{15} + 1.03 \). Thus, \( N(t+1) \) sequences from the actual generation were randomly chosen with replacement and copied over to the next generation. We then introduced \( n_{\text{mut}} \sim Poiss(\mu \cdot N(t+1) \cdot L) \) random mutations into these sequences with per site mutation rate \( \mu = 0.0001 \).
+For each generation, a fasta file with all sequences was stored and used as input for the incidence reconstruction pipeline. We ran 10 stochastic simulations with the settings stated above to compare the ground truth effective population sizes \( N(t) \) from our simulations with the corresponding inferred incidence trajectories \( \phi \).
+In Supplementary Note 1, we evaluated scenarios where only a fraction of the sequences were sampled (10-90%) or, to rule out sampling biases, we sub-sampled equal amounts of sequences at each time point, independent of \( N(t) \). Moreover, we assessed whether our predictions were affected by the introduction of unrelated sequence variants into the population.
+
+Effective reproduction number \( R_e \)
+Based on the reconstructed incidence histories, the effective reproduction number \( R_e(t) \) was computed using the established method by Wallinga and Teunis [61] (R package R0 [40]). Daily estimates of \( \phi \) were assigned a pseudocount of one and rounded to the nearest integer. For the generation time distribution \( g(\tau) \) of SARS-CoV-2, we chose the Gamma distribution with a mean of 5 days and a standard deviation of 1 day [15, 7].
+
+Phylodynamic analyses
+Phylodynamic analyses were performed on subsampled sets of the data described above (Data and data pre-processing) using a birth-death-sampling process as implemented in the BDSKY [55] model in BEAST2 [5]. Here the precise collection day of sequence samples with only information on year and month was inferred during the analysis and not a priori set to the 15th. The full data sets were first grouped by Pango lineage [48, 8] and then subsampled by randomly selecting a specific percentage of sequences per week (Victoria: 10% for lineage D.2, 50% for other lineages; Switzerland: 50% for all lineages; Scotland: 20% for all lineages; Denmark: 5% for all lineages). In addition, sequences were excluded if they belonged to a lineage with
+less than two representatives in the analyzed set and lineages with periods longer than 75 days without any sample were split into parts. Retained sequences were aligned to the reference genome (Genbank-ID MN908947.3 [3]) in MAFFT [22] using the –keeplength option and problematic sites were masked by replacing the them with 'N' in the alignment [11].
+For each remaining approximate cluster a separate phylogeny was reconstructed. A strict clock model with a fixed rate of \( 8 \cdot 10^{-4} \) substitutions per site per year and an HKY substitution model were used. In the embedded transmission model, transmission (\( \lambda \)), recovery (\( \mu \)) and sampling (\( \psi \)) rates were assumed to be piecewise constant with changes allowed either when intervention measures changed, or in a uniform manner (Supplementary Note 2). The reproductive number \( R_e(t) = \lambda(t)/(\mu(t) + \psi(t)) \) was drawn from a log-normal distribution \( R_e(t) \sim \log \mathcal{N}(0, 4) \), the rate to become non-infectious \( \delta(t) = \mu(t) + \psi(t) \) from a narrow normal distribution with \( \delta(t) \sim \mathcal{N}(27.11, 1) \) which is changed to \( \mathcal{N}(48.8, 1) \) after first control measures are implemented in the respective area. The sampling proportion \( s(t) = \psi(t)/(\psi(t) + \mu(t)) \) was *a priori* assumed to arise from a uniform distribution with a lower limit of zero and the upper limit determined by the ratio of analyzed sequences over diagnosed cases \( s \sim U(0, q_i/d_i) \) where \( d_i \) is the number of diagnoses and \( q_i \) the number of sequences included in the analysis in interval \( i \). To account for the lineage specific subsampling, a separate sampling proportion for lineage D.2, \( s_{D.2} \), was modelled in the analysis of the Victoria data. A uniform distribution with an upper limit corresponding to the subsampling percentage was thus used as prior distribution of the D.2 specific-, as well as general sampling proportion \( s_g \), i.e. \( s_{D.2} \sim U(0, 0.1) \) and \( s_g \sim U(0, 0.5) \).
+Setup files for all four analyses can be found as Supplementary Files.
+
+MCMC chains were run until all parameters converged, which took about 300 million steps for analyses of data from Denmark, Scotland and Switzerland. Because of the large D.2 cluster consisting of more than 900 sequences, about 750 million steps were needed for convergence using data from Victoria. On an Intel Xeon CPU E5-2687W (3.1 Ghz; 2 x 12 cores), this corresponded to about 15 hours to run one analysis for at least 300 million MCMC steps (about 3min/Msample). Log files were assessed using Tracer [49] and are included as Supplementary Files. TreeAnnotator was used to summarize the posterior sample of phylogenetic trees to a maximum clade credibility tree using median node heights. Lineage through time plots of all summary trees were calculated using the R package ape [42] and are shown in Supplementary Note 2.
+
+Relative case detection rate
+We used GInPipe to detect changes in SARS-CoV-2 case detection. Let us denote by \( P_t(\text{tested}|\text{infected}) \) the proportion of infected individuals that are actually diagnosed with the virus in week \( t \). According to Bayes’ theorem we have
+
+\[
+P_t(\text{tested}|\text{infected}) = \frac{P_t(\text{infected}|\text{tested}) \cdot P_t(\text{tested})}{P_t(\text{infected})}
+\]
+
+where \( P_t(\text{infected}|\text{tested}) \) denotes the proportion of tested individuals that are infected, \( P_t(\text{tested}) \) the proportion of individuals that are tested and \( P_t(\text{infected}) \) the proportion currently infected in week \( t \). We calculate \( P_t(\text{infected}|t) = \frac{r_{pos} - (1 - spec)}{sens - (1 - spec)} \) from the positivity rate \( r_{pos} \) of the conducted tests, corrected for the clinical sensitivity \( sens = 0.7 \) and specificity \( spec = 0.999 \) of the diagnostic tests [57]. For calculating the probability of being tested \( P(\text{tested}) \), we considered linear-, Poisson- and Binomial models, all of which yielded identical results. For all illustrations herein, we used the latter, yielding \( P_t(\text{tested}) = 1 - (1 - 1/pop)^{n_t} \), with \( pop \) denoting the population size in the respective regions or country and \( n_t \) denoting the number of tests conducted in the respective week.
+The probability of currently being infected \( P(\text{infected}) \approx \frac{N_{eff}}{pop} \) is unknown. However, since we know that \( N_{eff} \) is linearly correlated with the incidence estimate \( \phi \), we have \( P(\text{infected}) \approx c \cdot \frac{\phi}{pop} \). Putting everything together we can estimate the relative case detection rate:
+
+\[
+P_t(\text{tested}|\text{infected}) \cdot c = \frac{pop}{\phi_t} \cdot \frac{r_{pos} - (1 - spec)}{sens - (1 - spec)} \cdot \left( 1 - \left( 1 - \frac{1}{pop} \right)^{n_t} \right)
+\]
+
+Sources for the weekly number of performed tests, as well as test positive rates are stated in Supplementary Note 3.
+
+Author Contributions
+Conceptualization, M.R.S., M.T. and M.v.K.; Methodology, M.R.S., M.T., A.W., D.K. and M.v.K.; Investigation, M.R.S., M.T., A.W., Y.D. Writing - Original Draft, M.R.S., M.T., A.W. and M.v.K.; Writing-Review and Editing, M.R.S., M.T., A.W., D.K. and M.v.K.; Funding Acquisition, A.W., D.K. and M.v.K.; Supervision, D.K. and M.v.K.;
+
+Acknowledgements
+The authors acknowledge all labs contributing SARS-CoV-2 sequences to the GISAID EpiCoV database as stated in Supplementary Note 4.
+Funding
+M.R.S., M.T., Y.D. and MvK acknowledge funding from the Germany ministry for science and education (BMBF; grant numbers 01KI2016 and 031L0176A). D.K. and A.W. acknowledge funding from the Max Planck Society. A.W. acknowledges financial support through a scholarship (Landesgraduiertenstipendium), funded by the State of Thuringia, Germany. The funders had no role in designing the research or the decision to publish.
+
+Conflicts of interest
+The authors declare that no conflicts of interest exist.
+
+References
+[1] Frederick J Angulo, Lyn Finelli, and David L Swerdlow. “Estimation of US SARS-CoV-2 infections, symptomatic infections, hospitalizations, and deaths using seroprevalence surveys”. In: JAMA network open 4.1 (2021), e2033706–e2033706.
+[2] Lindsey R Baden et al. “Efficacy and safety of the mRNA-1273 SARS-CoV-2 vaccine”. In: New England Journal of Medicine 384.5 (2021), pp. 403–416.
+[3] Dennis A Benson et al. “GenBank”. In: Nucleic acids research 41.D1 (2012), pp. D36–D42.
+[4] Merle M Böhmer et al. “Investigation of a COVID-19 outbreak in Germany resulting from a single travel-associated primary case: a case series”. In: The Lancet Infectious Diseases 20.8 (2020), pp. 920–928.
+[5] Remco Bouckaert et al. “BEAST 2: a software platform for Bayesian evolutionary analysis”. In: PLoS Comput Biol 10.4 (2014), e1003537.
+[6] Lewis F Buss et al. “Three-quarters attack rate of SARS-CoV-2 in the Brazilian Amazon during a largely unmitigated epidemic”. In: Science 371.6526 (2021), pp. 288–292.
+[7] Robert Challen et al. “Meta-analysis of the SARS-CoV-2 serial interval and the impact of parameter uncertainty on the COVID-19 reproduction number”. In: medRxiv (2020). DOI: 10.1101/2020.11.17.20231548.
+[8] cov-lineages/pangolin. Oct. 2020. URL: https://github.com/cov-lineages/pangolin.
+[9] Christian G Daughton. “Wastewater surveillance for population-wide Covid-19: the present and future”. In: Science of the Total Environment 736 (2020), p. 139631.
+[10] Troy Day et al. “On the evolutionary epidemiology of SARS-CoV-2”. In: Current Biology 30.15 (2020), R849–R857.
+[11] Nicola De Maio, Conor Walker, and Rui Borges. Issues with SARS-CoV-2 Sequencing Data - SARS-CoV-2 Coronavirus / nCoV-2019 Genomic Epidemiology. Virological. 2020. URL: https://virological.org/t/issues-with-sars-cov-2-sequencing-data/473 (visited on 05/11/2021).
+[12] Andreas Deckert et al. “Effectiveness and cost-effectiveness of four different strategies for SARS-CoV-2 surveillance in the general population (CoV-Surv Study): a structured summary of a study protocol for a cluster-randomised, two-factorial controlled trial”. In: Trials 22.1 (2021), pp. 1–4.
+[13] Jacqueline Dinnes et al. “Rapid, point-of-care antigen and molecular-based tests for diagnosis of SARS-CoV-2 infection”. In: Cochrane Database of Systematic Reviews 3 (2021), p. CD013705.
+[14] Stefan Elbe and Gemma Buckland-Merrett. “Data, disease and diplomacy: GISAID’s innovative contribution to global health”. In: Global Challenges 1.1 (2017), pp. 33–46.
+[15] Luca Ferretti et al. “Quantifying SARS-CoV-2 transmission suggests epidemic control with digital contact tracing”. In: Science 368.6491 (2020). DOI: 10.1126/science.abb6936.
+[16] Simon DW Frost and Erik M Volz. “Viral phylodynamics and the search for an ‘effective number of infections’”. In: Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 365.1548 (2010), pp. 1879–1890.
+[17] John Griffin et al. “Rapid review of available evidence on the serial interval and generation time of COVID-19”. In: BMJ open 10.11 (2020), e040263.
+[18] Thomas Günther et al. “SARS-CoV-2 outbreak investigation in a German meat processing plant”. In: EMBO molecular medicine 12.12 (2020), e13296.
+[19] Kirsten Hanke et al. “Reconstruction of the genetic history and the current spread of HIV-1 subtype A in Germany”. In: Journal of virology 93.12 (2019).
+[20] Emma B Hodcroft et al. “Emergence and Spread of a SARS-CoV-2 Variant through Europe in the Summer of 2020”. In: medRxiv (2020). DOI: 10.1101/2020.10.25.20219063.
+
+[21] Markus Hoffmann et al. “SARS-CoV-2 Variants B.1.351 and P.1 Escape from Neutralizing Antibodies”. In: Cell 184.9 (2021), 2384–2393.e12.
+
+[22] Kazutaka Katoh and Daron M Standley. “MAFFT multiple sequence alignment software version 7: improvements in performance and usability”. In: Molecular biology and evolution 30.4 (2013), pp. 772–780.
+
+[23] Bhavin S Khatri and Austin Burt. “Robust estimation of recent effective population size from number of independent origins in soft sweeps”. In: Molecular biology and evolution 36.9 (2019), pp. 2040–2052.
+
+[24] Stephen M Kissler et al. “Projecting the transmission dynamics of SARS-CoV-2 through the postpandemic period”. In: Science 368.6493 (2020), pp. 860–868.
+
+[25] Max von Kleist et al. “HIV quasispecies dynamics during pro-active treatment switching: impact on multi-drug resistance and resistance archiving in latent reservoirs”. In: PloS one 6.3 (2011), e18204.
+
+[26] Johannes Köster and Sven Rahmann. “Snakemake—a scalable bioinformatics workflow engine”. In: Bioinformatics 28.19 (2012), pp. 2520–2522.
+
+[27] Denise Kühnert, Chieh-Hsi Wu, and Alexei J Drummond. “Phylogenetic and epidemic modeling of rapidly evolving infectious diseases”. In: Infection, genetics and evolution 11.8 (2011), pp. 1825–1841.
+
+[28] Philippe Lemey et al. “SARS-CoV-2 European resurgence foretold: interplay of introductions and persistence by leveraging genomic and mobility data”. In: Research Square (2021). DOI: 10.21203/rs.3.rs-208849/v1.
+
+[29] Heng Li. “Minimap2: pairwise alignment for nucleotide sequences”. In: Bioinformatics 34.18 (2018), pp. 3094–3100.
+
+[30] Qun Li et al. “Early transmission dynamics in Wuhan, China, of novel coronavirus–infected pneumonia”. In: New England journal of medicine (2020). DOI: 10.1056/NEJMoa2001316.
+
+[31] Milena Loprete et al. “Early warnings of COVID-19 outbreaks across Europe from social media”. In: Scientific reports 11.1 (2021), pp. 1–7.
+
+[32] Katrina A Lythgoe et al. “SARS-CoV-2 within-host diversity and transmission”. In: Science 372.6539 (2021).
+
+[33] Shabir A. Madhi et al. “Efficacy of the ChAdOx1 nCoV-19 Covid-19 Vaccine against the B.1.351 Variant”. In: The New England Journal of Medicine (2021). DOI: 10.1056/NEJMoa2102214.
+
+[34] Anup Malani et al. “Seroprevalence of SARS-CoV-2 in slums versus non-slums in Mumbai, India”. In: The Lancet Global Health 9.2 (2021), e110–e111.
+
+[35] Emmanuel Margolin et al. “Prospects for SARS-CoV-2 diagnostics, therapeutics and vaccines in Africa”. In: Nature Reviews Microbiology 18.12 (2020), pp. 690–704.
+
+[36] Asher Mullard. “How COVID Vaccines Are Being Divvied up around the World”. In: Nature (2020). DOI: 10.1038/d41586-020-03370-6.
+
+[37] Firzan Nainu et al. “SARS-CoV-2 reinfection and implications for vaccine development”. In: Human Vaccines & Immunotherapeutics 16.12 (2020), pp. 3061–3073.
+
+[38] Masatoshi Nei and Fumio Tajima. “Genetic drift and estimation of effective population size”. In: Genetics 98.3 (1981), pp. 625–640.
+
+[39] Nuala A. O’Leary et al. “Reference sequence (RefSeq) database at NCBI: current status, taxonomic expansion, and functional annotation”. In: Nucleic Acids Research 44.D1 (2015), pp. D733–D745.
+
+[40] Thomas Obadia, Romana Haneef, and Pierre-Yves Boëlle. “The R0 package: a toolbox to estimate reproduction numbers for epidemic outbreaks”. In: BMC medical informatics and decision making 12.1 (2012), pp. 1–9.
+
+[41] Daniel P Oran and Eric J Topol. “Prevalence of asymptomatic SARS-CoV-2 infection: a narrative review”. In: Annals of internal medicine 173.5 (2020), pp. 362–367.
+
+[42] Emmanuel Paradis and Klaus Schliep. “Ape 5.0: An Environment for Modern Phylogenetics and Evolutionary Analyses in R”. In: Bioinformatics (Oxford, England) 35.3 (2019), pp. 526–528.
+
+[43] Jordan Peccia et al. “Measurement of SARS-CoV-2 RNA in wastewater tracks community infection dynamics”. In: Nature Biotechnology 38.10 (2020), pp. 1164–1167.
+
+[44] Nicky Phillips. “The Coronavirus Is Here to Stay - Here’s What That Means”. In: Nature 590.7846 (2021), pp. 382–384.
+[45] Delphine Planas et al. “Sensitivity of infectious SARS-CoV-2 B. 1.1. 7 and B. 1.351 variants to neutralizing antibodies”. In: Nature medicine (2021). DOI: 10.1038/s41591-021-01318-5.
+
+[46] Fernando P Polack et al. “Safety and efficacy of the BNT162b2 mRNA Covid-19 vaccine”. In: New England Journal of Medicine 383.27 (2020), pp. 2603–2615.
+
+[47] Kaveh Pouran Yousef et al. “Inferring HIV-1 transmission dynamics in Germany from recently transmitted viruses”. In: JAIDS Journal of Acquired Immune Deficiency Syndromes 73.3 (2016), pp. 356–363.
+
+[48] Andrew Rambaut et al. “A Dynamic Nomenclature Proposal for SARS-CoV-2 Lineages to Assist Genomic Epidemiology”. In: Nature Microbiology 5.11 (2020), pp. 1403–1407.
+
+[49] Andrew Rambaut et al. “Posterior Summarization in Bayesian Phylogenetics Using Tracer 1.7”. In: Systematic Biology 67.5 (2018), pp. 901–904.
+
+[50] Chadi M Saad-Roy et al. “Immune life history, vaccination, and the dynamics of SARS-CoV-2 over the next 5 years”. In: Science 370.6518 (2020), pp. 811–818.
+
+[51] Ester C Sabino et al. “Resurgence of COVID-19 in Manaus, Brazil, despite high seroprevalence”. In: The Lancet 397.10273 (2021), pp. 452–455.
+
+[52] Sinaye Ngcapu San Emmanuel James et al. “High Resolution Analysis of Transmission Dynamics of Sars-Cov-2 in Two Major Hospital Outbreaks in South Africa Leveraging Intrahost Diversity”. In: medRxiv: The Preprint Server for Health Sciences (2020). DOI: 10.1101/2020.11.15.20231993.
+
+[53] Cuilhua Shen et al. “Using reports of symptoms and diagnoses on social media to predict COVID-19 case counts in mainland China: Observational infoveillance study”. In: Journal of medical Internet research 22.5 (2020), e19421.
+
+[54] Yuelong Shu and John McCauley. “GISAID: Global initiative on sharing all influenza data–from vision to reality”. In: Eurosurveillance 22.13 (2017), p. 30494.
+
+[55] Tanja Stadler et al. “Birth-death skyline plot reveals temporal changes of epidemic spread in HIV and hepatitis C virus (HCV)”. In: Proc Natl Acad Sci U S A 110.1 (2013), pp. 228–33.
+
+[56] John Stover et al. “The Risks and Benefits of Providing HIV Services during the COVID-19 Pandemic”. In: medRxiv (2021), p. 2021.03.01.21252663. DOI: 10.1101/2021.03.01.21252663.
+
+[57] Wiep van der Toorn et al. “An intra-host SARS-CoV-2 dynamics model to assess testing and quarantine strategies for incoming travelers, contact person management and de-isolation”. In: Patterns (2021). DOI: 10.1016/j.patter.2021.100262.
+
+[58] Yatish Turakhia et al. “Stability of SARS-CoV-2 phylogenies”. In: PLOS Genetics 16.11 (2020), pp. 1–34.
+
+[59] Erik M Volz, Katia Koelle, and Trevor Bedford. “Viral phylodynamics”. In: PLoS Comput Biol 9.3 (2013), e1002947.
+
+[60] Erik M Volz et al. “Phylodynamics of infectious disease epidemics”. In: Genetics 183.4 (2009), pp. 1421–1430.
+
+[61] Jacco Wallinga and Peter Teunis. “Different epidemic curves for severe acute respiratory syndrome reveal similar impacts of control measures”. In: American Journal of epidemiology 160.6 (2004), pp. 509–516.
+
+[62] J Wang, E Santiago, and Armando Caballero. “Prediction and estimation of effective population size”. In: Heredity 117.4 (2016), pp. 193–206.
+
+[63] Olivier J Wouters et al. “Challenges in ensuring global access to COVID-19 vaccines: production, affordability, allocation, and deployment”. In: The Lancet 397.10278 (2021), pp. 1023–1034.
+
+[64] Chieh-Hsi Wu and Alexei J Drummond. “Joint inference of microsatellite mutation models, population history and genealogies using transdimensional Markov Chain Monte Carlo”. In: Genetics 188.1 (2011), pp. 151–164.
+
+[65] Sean L Wu et al. “Substantial underestimation of SARS-CoV-2 infection in the United States”. In: Nature communications 11.1 (2020), pp. 1–10.
+
+[66] Daming Zhou et al. “Evidence of Escape of SARS-CoV-2 Variant B.1.351 from Natural and Vaccine-Induced Sera”. In: Cell 184.9 (2021), 2348–2361.e6.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• BEAST2ConfigurationFiles.zip
+• nCovPopDynAppendix.pdf

03b0b6a014cc46268783cf3d9b76467437f21edb16a2598fe060abe63f57631f/peer_review/peer_review.md

@@ -0,0 +1,207 @@
+Peer Review File
+
+Regulation of leptin signaling and diet-induced obesity by SEL1L-HRD1 ER-associated degradation in POMC expressing neurons
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+This appears to be an extended study from a previous study (PMID: 29457782), which showed that deletion of SEL1L in POMC neurons led to late-onset obesity with defects in processing POMC peptides. In the current study, the results showed that the same deletion led to an increased sensitivity to DIO. Additional data presented including the SEL1L pathway showing a reversible increase in expression level in response to HFD, the observed DIO by the deletion was not responsive to leptin, and the SEL1L deletion caused a similar defect in the retention of LepR to POMC were used by the authors to support the conclusion that the SEL1L pathway is required for leptin sensitivity in POMC neurons. Multidisciplinary approaches were used, the experimental designs are generally rigorous and the data analysis deems to be appropriate. The data on leptin receptor trafficking within the cell, although from in vitro analysis, appears to be exciting and may lead to a novel way of explanation on leptin resistance.
+
+However, there seem to some major concerns with the general theme, which if unanswered, may cast some doubts on the major conclusions presented here.
+
+Major:
+
+1) The issue with POMC-Cre being used. Previous study (PMID: 20348924) suggests that a significant portion of POMC-Cre expressing progenitors can also develop into other non-POMC neurons including AgRP neurons. Therefore the current model will have deletion in a substantial amount of non-POMC neurons, and the contribution of those non-POMC neurons should not be ignored.
+
+2) The issue on leptin action within the POMC neurons. Previous studies on leptin deletion in POMC neurons including both specific POMC neuron deletion (PMID: 29528284) or with POMC-Cre PMID:15207242) showed either no effects on body weight (the former) or very mild effect on body weight (later), which is in contrast to the strong obesity effects reported here. In principle, if the SEL1L pathway mediates the leptin action, then the deletion of SEL1L should not cause more dramatic effects then the receptor deletion itself. This discrepancy has to be resolved.
+
+3) The data presented in Fig. 3D seem to lack saline controls. The existing data may only support that the knockout mice could gain obesity with leptin treatment. The obesity development with
+leptin treatment should be directly compared with that from the same strain of mice with saline treatment to assess the response to leptin treatment.
+
+Minor:
+
+1) The leptin receptor trafficking experiments were done mostly in cell culture. Discussion with caveats associated with this in vitro approach should be provided.
+
+2) This reviewer is confused by the data presented in Fig. 4. If insulin and leptin were not increased in controls of Group III, would it suggest that parabiosis is not successful? Also how do the results support that DIO is due to altered leptin signaling?
+
+Reviewer #2 (Remarks to the Author):
+
+This study establishes a novel role for the SEL1L-HRD1 protein degradation system in the regulation of leptin receptor. signaling and energy balance, especially in conditions of nutrient overload. This manuscript reveals defects in SEL1L-HRD1 ERAD in POMC neurons can lead to diet-induced obesity related to issues involving leptin resistance and hyperphagia. The researchers argue that the effect of SEL1L-HRD1 ERAD is not linked to the IRE1α -XBP1 pathway of cellular stress response or cell death, contradicting previous studies. They propose, that the capability of SEL1L-HRD1 ERAD to maintain a conducive environment for proper protein folding in the endoplasmic reticulum can be a potential therapeutic target for diseases caused by protein misfolding and aggregation.
+
+Overall, I think this study would be a nice addition to the SEL1-HRD1 field, with some important modifications. I think this story should be re-titled to reflect the results (Leptin receptor degradation – NOT maturation). In addition, I have several major concerns.
+
+I believe the authors have overclaimed what their results demonstrate. So, I would urge some restraint in the claims made. Specifically, the discussion around the function of the ERAD pathway around mechanisms of leptin receptor folding is problematic. All of the evidence is indirect, and there is no evidence that ERAD is assisting folding, or promoting folding, or any involvement in the leptin signaling. Therefore, I think the authors should not be claiming that “SEL1-HRD1 ERAD is required for the maturation of LepRb”. In contrast, LepRb is simply a target of ERAD. When ERAD doesn’t degrade LepRb, it results in the non-degraded misfolded LepRb preventing proper maturation of properly folded LepRb. Downstream signaling is affected, but this is an indirect effect.
+It is very surprising that inactivation of SEL1L-HRD1 does not lead to any UPR induction. This is contrary to the many other studies on this topic, although not in these specific neurons. It seems unlikely that disrupting this highly-relevant system would be a completely different phenotype, when neurons are usually very sensitive to impaired ERAD function. I think this part raises concerns about these data, and their interpretation. The authors should address how this is possible in the text.
+
+How does a denaturing IP with FLAG work? The interaction between the ant-FLAG antibody and FLAG epitope are disrupted under denaturing conditions. There are no methods for this set of experiments. In addition the immunoprecipitated ubiquitin smear starts well below where the LepRb protein. How do the authors believe this is ubiquitin conjugated to LepRb? Without methods, these data impossible to believe and should not be used.
+
+Minor concerns:
+
+Figure 7D, why is the control missing in the IP sample (HSP90)? This is an important part of the experiment. Same for 7E, 7F.
+
+In figure 7A, the HRD1-/- line appears to have a band present only in the knockout line, although at a smaller size. How do the authors explain this?
+
+Figure 8 seems irrelevant for this story. I would just remove it from this manuscript, or at least move it into the supplementary material.
+
+I would encourage a careful review of the text, because there are a number of misspellings and missing definite articles throughout.
+Reviewer #1
+This appears to be an extended study from a previous study (PMID: 29457782), which showed that deletion of SEL1L in POMC neurons led to late-onset obesity with defects in processing POMC peptides. In the current study, the results showed that the same deletion led to an increased sensitivity to DIO. Additional data presented including the SEL1L pathway showing a reversible increase in expression level in response to HFD, the observed DIO by the deletion was not responsive to leptin, and the SEL1L deletion caused a similar defect in the retention of LepR to POMC were used by the authors to support the conclusion that the SEL1L pathway is required for leptin sensitivity in POMC neurons. Multidisciplinary approaches were used, the experimental designs are generally rigorous and the data analysis deems to be appropriate. The data on leptin receptor trafficking within the cell, although from in vitro analysis, appears to be exciting and may lead to a novel way of explanation on leptin resistance. However, there seem to some major concerns with the general theme, which if unanswered, may cast some doubts on the major conclusions presented here.
+We thank the reviewer for the enthusiastic and constructive comments, which have been very helpful for us to improve our work.
+
+Major:
+1) The issue with POMC-Cre being used. Previous study (PMID: 20348924) suggests that a significant portion of POMC-Cre expressing progenitors can also develop into other non-POMC neurons including AgRP neurons. Therefore the current model will have deletion in a substantial amount of non-POMC neurons, and the contribution of those non-POMC neurons should not be ignored.
+We thank for the reviewer for this great point with which we agree. As this is the best Cre line that is available, we now have included a discussion on this issue in the Discussion on page 10 and pasted below.
+
+“As POMC-Cre expressing progenitors give raise to nearly one-fourth of the hypothalamic AgRP/NPY neurons 1, Sel1L deletion hence occurs in both POMC neurons and a fraction of AgRP/NPY neurons in our mouse model. Since leptin signaling inhibits appetite via both POMC and hypothalamic AgRP/NPY neurons 2, our findings and conclusions reported in this study remains valid. Future studies will be required to examine the effect of SEL1L-HRD1 ERAD in hypothalamic AgRP/NPY neurons.”
+
+2) The issue on leptin action within the POMC neurons. Previous studies on leptin deletion in POMC neurons including both specific POMC neuron deletion (PMID: 29528284) or with POMC-Cre PMID:15207242) showed either no effects on body weight (the former) or very mild effect on body weight (later), which is in contrast to the strong obesity effects reported here. In principle, if the SEL1L pathway mediates the leptin action, then the deletion of SEL1L should not cause more dramatic effects then the receptor deletion itself. This discrepancy has to be resolved.
+We thank the reviewer for this great comment. In PMID:152072423, mice with leptin receptor deficiency in POMC neurons (Pomc-Cre;LepRflox/flox) exhibited significantly higher body weight gain starting at 6 weeks of age in male and 4 weeks of age in females3, which is consistent with another study (PMID: 29289646)4. These mice were put on regular chow diet, with only 12.5% Kcal from fat. This is actually quite dramatic bodyweight difference upon normal chow diet comparing to our Sel1Lflox/flox;Pomc-Cre mice upon chow diet, which didn’t show any significant bodyweight difference until 13 weeks of age. Furthermore, the study (PMID: 29289646) showed that LepRflox/flox;Pomc-Cre mice on HFD exhibited significant weight gain and obesity4, which is
+consistent with our findings. These findings are distinct from another study as pointed out by this reviewer (PMID: 29528284)5, where the inducible mouse model (LepRfl/loxP;Pomc-CreERT2) was used to delete LepR in POMC neurons of adult mice. The authors failed to observe significant change in body weight 4 weeks post-tamoxifen administration5. While the reasons for the discrepancies of these different mouse models remain unclear, we speculate that the knockout efficiency in the inducible KO model is less ideal6, with additional concerns about metabolic effects of tamoxifen7,8. We now have discussed this point in the Discussion on page 10-11 and pasted below.
+
+“Moreover, there are conflicting findings in research on LepRb-expressing POMC neurons. Some studies show that mice with congenital deletion of LepRb in POMC neurons exhibit significantly higher body weight gain in both sexes at 4-6 weeks of age on a regular chow diet 3,4. Additionally, their body weight gain and obesity are significantly exacerbated when on a HFD4, consistent with our observations. However, other research using an inducible mouse model that deletes LepRb in POMC neurons in adult mice found no significant weight loss four weeks after tamoxifen administration5. The discrepancies between these studies, which likely arise from different Cre systems, remain unclear. We speculate that the knockout efficiency in the conditional model may be suboptimal6, and that tamoxifen may also induce metabolic alteration7,8.”
+
+3) The data presented in Fig. 3D seem to lack saline controls. The existing data may only support that the knockout mice could gain obesity with leptin treatment. The obesity development with leptin treatment should be directly compared with that from the same strain of mice with saline treatment to assess the response to leptin treatment.
+We thank the reviewer for this great suggestion. We now have added the cohort with saline treatment in Fig. 3d-f of revised manuscript (see below). The data showed that leptin injection reduced body weight and food intake in WT mice compared to PBS-treated WT mice, while this anorexigenic effect of leptin was abolished in Sel1LPOMC mice, indicating leptin resistance in Sel1LPOMC mice, which is consistent with the data shown in the original manuscript.
+
+![Diagram showing leptin or vehicle injections over days of HFD, and bar graphs of body weight change and average daily food intake for Sel1Lff and Sel1LPOMC mice](page_101_1042_1397_377.png)
+
+Figure 3. Hypothalamic ERAD deficiency triggers hyperphagia and leptin resistance. (d-f) Diagram (d) of leptin sensitivity test. 12-week-old mice on HFD were injected daily i.p. first with vehicle PBS and then PBS or leptin (2 mg/kg body weight) for 3 days. (e-f) Body weight change (e), average daily food intake (f) following 3 daily vehicle or leptin injections of the male mice (PBS: n=4 mice per group; leptin: n=6, 4 mice for Sel1Lff and Sel1LPOMC). Body weight change was calculated by end point body weights minus starting point body weights.
+
+Minor:
+1) The leptin receptor trafficking experiments were done mostly in cell culture. Discussion with caveats associated with this in vitro approach should be provided.
+We thank the reviewer for this great suggestion. There is no good antibody for leptin receptor that would allow us to perform studies in vivo, hence we have performed extensive studies in
+vitro to ensure that the results are rigorous, and the conclusions are solid. We now have included the caveats in the Discussion on page 12-13.
+
+“In this study, due to the lack of an efficient antibody for specific LepRb detection in vivo, we conducted extensive mechanistic investigations in vitro instead. We acknowledge the limitations of the in vitro system used, which may not fully represent in vivo neuronal systems. Nevertheless, the extensive use of in vivo and in vitro approaches in this study provides strong support for our overall conclusions.”
+
+2) This reviewer is confused by the data presented in Fig. 4. If insulin and leptin were not increased in controls of Group III, would it suggest that parabiosis is not successful? Also how do the results support that DIO is due to altered leptin signaling?
+We thank the reviewer for the great questions. On the first questions, we were puzzled as well, but speculated that it may be due to short half-life of the hormones and faster clearance as reported previously9,10. On the second question, the reason that results supporting altered leptin signaling in the PKO mice is that in the group III WT-KO parabionts, only WT mice failed to gain weight due to the hyperleptinemia, while having no effect on the body weight gain of the KO parabionts. We have clarified these points in the revised manuscript on page 6-7 and pasted below.
+
+“Two control parabionts, WT:WT (Group I) and Sel1L^{POMC}:Sel1L^{POMC} (Group II), gained weight as expected with the latter pair becoming obese (Fig. 4a-b). By contrast, in WT:Sel1L^{POMC} (Group III) parabionts, body weight gain for WT mice was attenuated compared to WT mice in WT:WT (Group I) parabionts, while body weight gain for Sel1L^{POMC} mice was comparable to that of Sel1L^{POMC}:Sel1L^{POMC} parabionts (Group II) (Fig. 4a-b). Body compositions (i.e., lean vs. fat) in parabionts were not affected by the partner (Fig. 4c). Moreover, serum leptin and insulin levels were highly elevated in Sel1L^{POMC} mice, but unaltered in WT mice regardless of the partners (Fig. 4d, e), likely due to short half-life of circulating hormones as previously reported 9,10. Overall, these data suggested the Sel1L^{POMC} mice are defective in responding to circulating leptin.”
+
+Reviewer #2
+This study establishes a novel role for the SEL1L-HRD1 protein degradation system in the regulation of leptin receptor, signaling and energy balance, especially in conditions of nutrient overload. This manuscript reveals defects in SEL1L-HRD1 ERAD in POMC neurons can lead to diet-induced obesity related to issues involving leptin resistance and hyperphagia. The researchers argue that the effect of SEL1L-HRD1 ERAD is not linked to the IRE1α -XBP1 pathway of cellular stress response or cell death, contradicting previous studies. They propose, that the capability of SEL1L-HRD1 ERAD to maintain a conducive environment for proper protein folding in the endoplasmic reticulum can be a potential therapeutic target for diseases caused by protein misfolding and aggregation. Overall, I think this study would be a nice addition to the SEL1-HRD1 field, with some important modifications. I think this story should be re-titled to reflect the results (Leptin receptor degradation – NOT maturation). In addition, I have several major concerns.
+We thank the reviewer for the enthusiastic and constructive comments, which have been very helpful for us to improve our work. We now have changed the title to better reflect the main findings reported in this study: “POMC neuron-specific SEL1L-HRD1 ER-associated degradation regulates leptin signaling and diet-induced obesity”
+I believe the authors have overclaimed what their results demonstrate. So, I would urge some restraint in the claims made. Specifically, the discussion around the function of the ERAD pathway around mechanisms of leptin receptor folding is problematic. All of the evidence is indirect, and there is no evidence that ERAD is assisting folding, or promoting folding, or any involvement in the leptin signaling. Therefore, I think the authors should not be claiming that “SEL1L-HRD1 ERAD is required for the maturation of LepRb”. In contrast, LepRb is simply a target of ERAD. When ERAD doesn’t degrade LepRb, it results in the non-degraded misfolded LepRb preventing proper maturation of properly folded LepRb. Downstream signaling is affected, but this is an indirect effect.
+
+We thank the reviewer for this great comment. We now have carefully gone through the manuscript to avoid overstatement as much as possible, including the title. We paste some examples below:
+
+“In the absence of SEL1L-HRD1 ERAD, leptin receptors are largely retained in the ER.” -page 2, line 37-38
+
+“Here, we show that SEL1L-HRD1 ERAD in POMC neurons at the arcuate nucleus (ARC) of the hypothalamus controls DIO pathogenesis and leptin sensitivity via the regulation of leptin receptor biogenesis and signaling. POMC-specific Sel1L deficient (Sel1L^{POMC}) mice are hypersensitive to high fat diet (HFD) feeding. Surprisingly, our data suggest that SEL1L-HRD1 ERAD is required for the folding quality control of nascent leptin receptors.” -page 4, line 73-77
+
+“SEL1L-HRD1 is required for the folding and ER exit of nascent leptin receptor (LepRb).” -page 8, line 190
+
+“Taken together, our data showed that SEL1L-HRD1 ERAD is required for the folding of nascent LepRb in the ER and ER exit for maturation.”- page 9, line 213-214
+
+“Hence, we concluded that C604S hLepRb variant is trapped in the ER and targeted for proteasomal degradation by SEL1L-HRD1 ERAD.”- page 10, line 228-230
+
+It is very surprising that inactivation of SEL1L-HRD1 does not lead to any UPR induction. This is contrary to the many other studies on this topic, although not in these specific neurons. It seems unlikely that disrupting this highly-relevant system would be a completely different phenotype, when neurons are usually very sensitive to impaired ERAD function. I think this part raises concerns about these data, and their interpretation. The authors should address how this is possible in the text.
+
+We thank the reviewer for this great comment. We observed mild, if any, UPR activation in the Sel1L-deficient POMC neurons with no detectable cell death or inflammation. The findings in this study are consistent with many recent studies of various types of cells and tissues^{11-16}. Moreover, previous studies suggest protective role of POMC neuron-specific IRE1α and XBP1 in DIO^{17,18}, while others proposed opposing claims^{19}. Hence, we conclude that the effect of POMC neuron- specific SEL1L-HRD1 ERAD is independent of the IRE1α -XBP1 pathway of the UPR and cell death. Further, we recently identified several hypomorphic variants of SEL1L and HRD1 in patients with ERAD-associated neurodevelopmental disorders with infancy onset (EDNI) syndrome ^{20,21}. Notably, neither patient fibroblasts nor knockin HEK293T cells carrying these variants exhibited any overt UPR. These findings point to cellular adaptation in response to ERAD deficiency and the accumulation of misfolded proteins, including the induction of ER chaperones, expansion of ER, and activation of ER-phagy, etc^{22-27}. We now included a
+discussion on this point in the Discussion of the revised manuscript on page 11 and pasted below:
+
+“Sel1L-deficiency in POMC neurons is associated with minimal, if any, UPR activation, and thus no detectable cell death or inflammation, which is in line with many recent studies of different cell and tissue types 11-16. Moreover, previous studies have shown that Ire1α or Xbp1 deficiency in POMC neurons predispose mice to DIO18 and gain-of-function of XBP1 in POMC neurons protects mice against DIO17, while others claim opposingly that mice with IRE1α-deficient POMC neurons are more resistant to DIO19. Therefore, we conclude that the effect of SEL1L-HRD1 ERAD is independent of the IRE1α -XBP1 pathway of the UPR and cell death. Furthermore, we recently identified several hypomorphic variants of SEL1L and HRD1 in patients with ERAD-associated neurodevelopmental disorders with infancy onset (EDNI) syndrome 20,21. Notably, neither patient fibroblasts nor knockin HEK293T cells carrying these variants exhibited any overt UPR. These findings point to cellular adaptation in response to ERAD deficiency and the accumulation of misfolded proteins. Adaptive mechanisms include upregulation of ER chaperones for increased folding efficiency, expansion of ER volume to dilute misfolded protein concentration, enhanced aggregation and sequestration of misfolded proteins to reduce proteotoxicity, and/or activation of ER-phagy to clear misfolded protein aggregates or damaged ER 22-27.”
+
+How does a denaturing IP with FLAG work? The interaction between the ant-FLAG antibody and FLAG epitope are disrupted under denaturing conditions. There are no methods for this set of experiments. In addition the immunoprecipitated ubiquitin smear starts well below where the LepRb protein. How do the authors believe this is ubiquitin conjugated to LepRb? Without methods, these data impossible to believe and should not be used.
+We thank the reviewer for this great comment and apologize for our oversight. We now have included a detailed protocol under “Denaturing IP for ubiquitination assay” in Methods. The purpose of the denaturing IP is to detect Ub directly on LepRb by dissociating any interacting proteins that are potentially Ub under the denaturing conditions. Further before adding Flag antibody for immunoprecipitation, the lysis will be greatly diluted (1:10 in non-denaturing lysis buffer) to prevent denaturation of the antibody used for immunoprecipitation. As Flag antibody recognizes linear Flag epitope (DYKDDDDK), hence denaturing before immunoprecipitation with sufficient dilution during immunoprecipitation procedure won’t affect the immunoprecipitation efficiency. On the second point, we are not sure about signals below the full-length LepRb, which should be background (as similar signal present in negative control lane with no LepRb-3xFLAG in Fig. 7f).
+
+Minor concerns:
+Figure 7D, why is the control missing in the IP sample (HSP90)? This is an important part of the experiment. Same for 7E, 7F.
+We thank the reviewer for this great comment. We now have included those controls in Fig. 7d-f of the revised manuscript (see below)
+(f) Immunoblot analysis of Ub following denaturing immunoprecipitation (IP) of Flag from lysates of HEK293T transfected with mLepRb-3xFlag (n=3 independent samples per group).
+
+In figure 7A, the HRD1-/- line appears to have a band present only in the knockout line, although at a smaller size. How do the authors explain this?
+We thank the reviewer for this great comment. The band showing up in HRD1-/- line with smaller size is the truncated form of HRD1 with a deletion of the first 60AA. Such truncation would abolish the interaction of SEL1L-HRD1, which is known to be essential for HRD1 function28.
+
+Figure 8 seems irrelevant for this story. I would just remove it from this manuscript, or at least move it into the supplementary material.
+We now have moved it into the supplementary material as suggested as Supplementary Fig. 6 of revised manuscript.
+
+I would encourage a careful review of the text, because there are a number of misspellings and missing definite articles throughout.
+We thank the reviewer for this great comment and now have carefully reviewed the manuscript to avoid typos and errors.
+1 Padilla, S. L., Carmody, J. S. & Zeltser, L. M. Pomc-expressing progenitors give rise to antagonistic neuronal populations in hypothalamic feeding circuits. Nat Med 16, 403-405, doi:10.1038/nm.2126 (2010).
+2 Wang, Q. et al. Interactions between leptin and hypothalamic neuropeptide Y neurons in the control of food intake and energy homeostasis in the rat. Diabetes 46, 335-341, doi:10.2337/diab.46.3.335 (1997).
+3 Baltasbar, N. et al. Leptin receptor signaling in POMC neurons is required for normal body weight homeostasis. Neuron 42, 983-991, doi:10.1016/j.neuron.2004.06.004 (2004).
+4 Bell, B. B. et al. Differential contribution of POMC and AgRP neurons to the regulation of regional autonomic nerve activity by leptin. Mol Metab 8, 1-12, doi:10.1016/j.molmet.2017.12.006 (2018).
+5 Caron, A. et al. POMC neurons expressing leptin receptors coordinate metabolic responses to fasting via suppression of leptin levels. Elife 7, doi:10.7554/eLife.33710 (2018).
+6 Ilchuk, L. A. et al. Limitations of Tamoxifen Application for In Vivo Genome Editing Using Cre/ER(T2) System. Int J Mol Sci 23, doi:10.3390/ijms232214077 (2022).
+7 Hesselbarth, N. et al. Tamoxifen affects glucose and lipid metabolism parameters, causes browning of subcutaneous adipose tissue and transient body composition changes in C57BL/6NTac mice. Biochem Biophys Res Commun 464, 724-729, doi:10.1016/j.bbrc.2015.07.015 (2015).
+8 Lopez, M. et al. Tamoxifen-induced anorexia is associated with fatty acid synthase inhibition in the ventromedial nucleus of the hypothalamus and accumulation of malonyl-CoA. Diabetes 55, 1327-1336, doi:10.2337/db05-1356 (2006).
+9 Harris, R. B., Zhou, J., Weigle, D. S. & Kuijper, J. L. Recombinant leptin exchanges between parabiosed mice but does not reach equilibrium. Am J Physiol 272, R1800-1808, doi:10.1152/ajpregu.1997.272.6.R1800 (1997).
+10 Harris, R. B. Contribution made by parabiosis to the understanding of energy balance regulation. Biochim Biophys Acta 1832, 1449-1455, doi:10.1016/j.bbadis.2013.02.021 (2013).
+11 Bhattacharya, A. et al. Hepatic Sel1L-Hrd1 ER-associated degradation (ERAD) manages FGF21 levels and systemic metabolism via CREBH. EMBO J 37, doi:10.15252/embj.201899277 (2018).
+12 Kim, G. H. et al. Hypothalamic ER-associated degradation regulates POMC maturation, feeding, and age-associated obesity. J Clin Invest 128, 1125-1140, doi:10.1172/JCI96420 (2018).
+13 Shi, G. et al. ER-associated degradation is required for vasopressin prohormone processing and systemic water homeostasis. J Clin Invest 127, 3897-3912, doi:10.1172/JCI94771 (2017).
+14 Shrestha, N. et al. Sel1L-Hrd1 ER-associated degradation maintains beta cell identity via TGF-beta signaling. J Clin Invest 130, 3499-3510, doi:10.1172/JCI134874 (2020).
+15 Yoshida, S. et al. Endoplasmic reticulum-associated degradation is required for nephrin maturation and kidney glomerular filtration function. J Clin Invest 131, doi:10.1172/JCI143988 (2021).
+16 Zhou, Z. et al. Endoplasmic reticulum-associated degradation regulates mitochondrial dynamics in brown adipocytes. Science 368, 54-60, doi:10.1126/science.aay2494 (2020).
+17 Williams, K. W. et al. Xbp1s in Pomc neurons connects ER stress with energy balance and glucose homeostasis. Cell Metab 20, 471-482, doi:10.1016/j.cmet.2014.06.002 (2014).
+18 Yao, T. et al. Ire1alpha in Pomc Neurons Is Required for Thermogenesis and Glycemia. Diabetes **66**, 663-673, doi:10.2337/db16-0533 (2017).
+19 Xiao, Y. et al. Knockout of inositol-requiring enzyme 1alpha in pro-opiomelanocortin neurons decreases fat mass via increasing energy expenditure. Open Biol **6**, doi:10.1098/rsob.160131 (2016).
+20 Wang, H. H. et al. Hypomorphic variants of SEL1L-HRD1 ER-associated degradation are associated with neurodevelopmental disorders. J Clin Invest **134**, doi:10.1172/JCI170054 (2024).
+21 Weis, D. et al. Biallelic Cys141Tyr variant of SEL1L is associated with neurodevelopmental disorders, agammaglobulinemia, and premature death. J Clin Invest **134**, doi:10.1172/JCI170882 (2024).
+22 Qi, L., Tsai, B. & Arvan, P. New Insights into the Physiological Role of Endoplasmic Reticulum-Associated Degradation. Trends Cell Biol **27**, 430-440, doi:10.1016/j.tcb.2016.12.002 (2017).
+23 Hwang, J. & Qi, L. Quality Control in the Endoplasmic Reticulum: Crosstalk between ERAD and UPR pathways. Trends Biochem Sci **43**, 593-605, doi:10.1016/j.tibs.2018.06.005 (2018).
+24 Bhattacharya, A. & Qi, L. ER-associated degradation in health and disease - from substrate to organism. J Cell Sci **132**, doi:10.1242/jcs.232850 (2019).
+25 Shrestha, N. et al. Integration of ER protein quality control mechanisms defines beta cell function and ER architecture. J Clin Invest **133**, doi:10.1172/JCI163584 (2023).
+26 Wu, S. A. et al. The mechanisms to dispose of misfolded proteins in the endoplasmic reticulum of adipocytes. Nature communications **14**, 3132, doi:10.1038/s41467-023-38690-4 (2023).
+27 Christianson, J. C., Jarosch, E. & Sommer, T. Mechanisms of substrate processing during ER-associated protein degradation. Nat Rev Mol Cell Biol **24**, 777-796, doi:10.1038/s41580-023-00633-8 (2023).
+28 Lin, L. L. et al. SEL1L-HRD1 interaction is required to form a functional HRD1 ERAD complex. Nat Commun **15**, 1440, doi:10.1038/s41467-024-45633-0 (2024).
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+I appreciated the authors' efforts in addressing my comments. While most of my comments have been sufficiently addressed, I feel that the issue on POMC-Cre remains, which is an important issue as it may cause non-necessary confusion on POMC neuron function related to leptin action. Those citations used by the authors to argue for the effect of Leptin receptor deletion in POMC neurons in causing obesity all used original POMC-Cre mice that also target portion of AgRP as well as other neurons in the arcuate nucleus. The authors also cited the paper that demonstrated no body weight effect was observed by inducible leptin receptor deletion in POMC neurons. The contrasting difference at least to this reviewer would argue that the obesity observed in the original POMC-Cre mice may be due to the leptin deletion in those Cre positive non-POMC neurons. In lieu of this argument, I would strongly suggest the authors to change the narrative to explicitly reflect this possibility.
+
+Reviewer #2 (Remarks to the Author):
+
+The authors have addressed most of my main concerns.
+
+I still believe they should not be claiming that ERAD promotes folding of the leptin receptor. The effect of ERAD on folding of the receptor is likely to be indirect, unless demonstrated otherwise. The specific text examples they highlight are somewhat misleading. For example this section title would lead readers to think ERAD was involved directly in the folding process:
+
+SEL1L-HRD1 is required for the folding and ER exit of nascent leptin receptor (LepRb). -page 8, line 187
+
+or
+
+:Taken together, our data showed that SEL1L-HRD1 ERAD is required for the folding of nascent LepRb in the ER and ER exit for maturation." - page 9, line 213-214"
+I think these statements are somewhat misleading because they indicate that the folding does not happen without ERAD, which is not measured. ER exit does not happen, but this could be for many other reasons. Therefore, I would strongly urge the authors to more carefully word these statements.
+Reviewer #1
+I appreciated the authors' efforts in addressing my comments. While most of my comments have been sufficiently addressed, I feel that the issue on POMC-Cre remains, which is an important issue as it may cause non-necessary confusion on POMC neuron function related to leptin action. Those citations used by the authors to argue for the effect of Leptin receptor deletion in POMC neurons in causing obesity all used original POMC-Cre mice that also target portion of AgRP as well as other neurons in the arcuate nucleus. The authors also cited the paper that demonstrated no body weight effect was observed by inducible leptin receptor deletion in POMC neurons. The contrasting difference at least to this reviewer would argue that the obesity observed in the original POMC-Cre mice may be due to the leptin deletion in those Cre positive non-POMC neurons. In lieu of this argument, I would strongly suggest the authors to change the narrative to explicitly reflect this possibility.
+
+We thank the reviewer for the constructive comments. We now have changed the title and revised our Discussion sections to explicitly the issue of POMC-Cre model per reviewer’s suggestion. In many places, we changed the “POMC neurons” to “POMC-expressing neurons”. Examples are pasted below:
+
+Title: Regulation of leptin signaling and diet-induced obesity by SEL1L-HRD1 ER-associated degradation in POMC expressing neurons
+
+Discussion Page 10 line 241: In our Sel1L^{POMC} mouse model, the POMC promoter is active at embryonic day 10.5 (E10.5) ^1, resulting in Sel1L deletion in POMC-expressing progenitors. As these progenitors differentiate into various neuronal types, including LepRb-expressing POMC neurons as well as non-POMC neurons ^2-6, the impact of SEL1L-HRD1 ERAD in Sel1L^{POMC} mice may reflect effects on both POMC and non-POMC neurons. Notably, non-POMC neurons, such as nearly a quarter of the hypothalamic AgRP/NPY neurons ^2 or functionally distinct POMC neurons ^4-5, could also influence leptin signaling and food intake. Earlier studies have shown that mice with LepRb deficiency in POMC-expressing neurons (using the same POMC-cre line as ours) gain significantly higher body weight in both sexes at 4-6 weeks of age on a regular chow diet ^7,8 or HFD ^8. However, another study using the inducible POMC Cre-ERT2 line to delete LepRb in POMC neurons of adult mice found systemic insulin and leptin resistance but no significant body weight gain four weeks after tamoxifen administration ^9. The discrepancies may be attributed to the possible involvement of other non-POMC neurons in the studies using the original POMC-cre line ^7,8. Therefore, future research is needed to explore the role of SEL1L-HRD1 ERAD in other types of hypothalamic neurons.
+
+Reviewer #2
+The authors have addressed most of my main concerns. I still believe they should not be claiming that ERAD promotes folding of the leptin receptor. The effect of ERAD on folding of the receptor is likely to be indirect, unless demonstrated otherwise. The specific text examples they highlight are somewhat misleading. For example this section title would lead readers to think ERAD was involved directly in the folding process: SEL1L-HRD1 is required for the folding and ER exit of nascent leptin receptor (LepRb). -page 8, line 187 or :Taken together, our data showed that SEL1L-HRD1 ERAD is required for the folding of nascent LepRb in the ER and ER exit for maturation." - page 9, line 213-214". I think these statements are somewhat misleading because they indicate that the folding does not happen without ERAD, which is not measured. ER exit does not happen, but this could be for many other reasons. Therefore, I would strongly urge the authors to more carefully word these statements.
+We thank the reviewer for the constructive comments, which we agree and have changed the text to truly reflect what the data shows. We are very appreciative of this reviewer for pointing this out twice. We pasted some examples below:
+
+Page 2 line 28: Here we report that SEL1L-HRD1 protein complex of the highly conserved ER-associated protein degradation (ERAD) machinery in POMC-expressing neurons ameliorates diet-induced obesity and its associated complications, partly by regulating the turnover of the long isoform of Leptin receptors (LepRb).
+
+Page 4 line 74, and Page 9 line 210: We further show that nascent LepRb protein is unstable and degraded by SEL1L-HRD1 ERAD, a process required for functional LepRb to reach the cell surface.
+
+Page 8 line 190: LepRb protein is an endogenous substrate of SEL1L-HRD1 ERAD
+
+Page 10 line 232: Our data show that native LepRb is an endogenous substrate of SEL1L-HRD1 ERAD. This degradation event is crucial for the production of functional LepRb proteins at the cell surface as ERAD deficiency results in LepRb being retained in the ER (Fig. 9).
+
+References:
+1 McNay, D. E., Pelling, M., Claxton, S., Guillemot, F. & Ang, S. L. Mash1 is required for generic and subtype differentiation of hypothalamic neuroendocrine cells. Mol Endocrinol 20, 1623-1632, doi:10.1210/me.2005-0518 (2006).
+2 Padilla, S. L., Carmody, J. S. & Zeltser, L. M. Pomc-expressing progenitors give rise to antagonistic neuronal populations in hypothalamic feeding circuits. Nat Med 16, 403-405, doi:10.1038/nm.2126 (2010).
+3 Toda, C., Santoro, A., Kim, J. D. & Diano, S. POMC Neurons: From Birth to Death. Annu Rev Physiol 79, 209-236, doi:10.1146/annurev-physiol-022516-034110 (2017).
+4 Yu, H., Rubinstein, M. & Low, M. J. Developmental single-cell transcriptomics of hypothalamic POMC neurons reveal the genetic trajectories of multiple neuropeptidergic phenotypes. Elife 11, doi:10.7554/eLife.72883 (2022).
+5 Biglari, N. et al. Functionally distinct POMC-expressing neuron subpopulations in hypothalamus revealed by intersectional targeting. Nat Neurosci 24, 913-929, doi:10.1038/s41593-021-00854-0 (2021).
+6 Padilla, S. L., Reef, D. & Zeltser, L. M. Defining POMC neurons using transgenic reagents: impact of transient Pomc expression in diverse immature neuronal populations. Endocrinology 153, 1219-1231, doi:10.1210/en.2011-1665 (2012).
+7 Balthasar, N. et al. Leptin receptor signaling in POMC neurons is required for normal body weight homeostasis. Neuron 42, 983-991, doi:10.1016/j.neuron.2004.06.004 (2004).
+8 Bell, B. B. et al. Differential contribution of POMC and AgRP neurons to the regulation of regional autonomic nerve activity by leptin. Mol Metab 8, 1-12, doi:10.1016/j.molmet.2017.12.006 (2018).
+9 Caron, A. et al. POMC neurons expressing leptin receptors coordinate metabolic responses to fasting via suppression of leptin levels. Elife 7, doi:10.7554/eLife.33710 (2018).
+REVIEWERS' COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The authors have sufficiently addressed my concerns and recommend acceptance.
+
+Reviewer #2 (Remarks to the Author):
+
+I have no further concerns.

03b0b6a014cc46268783cf3d9b76467437f21edb16a2598fe060abe63f57631f/preprint/preprint.md

@@ -0,0 +1,505 @@
+SEL1L-HRD1 ER-associated degradation regulates leptin receptor maturation and signaling in POMC neurons in diet-induced obesity
+
+Ling Qi
+xvr2hm@virginia.edu
+
+University of Virginia   https://orcid.org/0000-0001-8229-0184
+
+Hancheng Mao
+Department of Molecular & Integrative Physiology, University of Michigan Medical School
+https://orcid.org/0000-0003-2546-6774
+
+Geun Hyang Kim
+Regeneron Pharmaceuticals, Inc.
+
+Article
+
+Keywords: SEL1L-HRD1 ERAD, POMC, diet-induced obesity, leptin signaling, leptin receptor, parabiosis
+
+Posted Date: January 12th, 2024
+
+DOI: https://doi.org/10.21203/rs.3.rs-3768472/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on September 29th, 2024. See the published version at https://doi.org/10.1038/s41467-024-52743-2.
+SEL1L-HRD1 ER-associated degradation regulates leptin receptor maturation and signaling in POMC neurons in diet-induced obesity
+
+Hancheng Mao¹, Geun Hyang Kim¹,³, Ling Qi²*
+
+¹Department of Molecular & Integrative Physiology, University of Michigan Medical School, Ann Arbor, MI 48105, USA
+² Department of Molecular Physiology and Biological Physics, University of Virginia, School of Medicine, Charlottesville, VA 22903, USA
+³Present address: Regeneron Pharmaceuticals, Inc., 777 Old Saw Mill River Road, Tarrytown, New York 10591, USA
+*Correspondence: xvr2hm@virginia.edu
+
+The authors have declared that no conflict of interest exists.
+
+Short title: Regulation of leptin receptor maturation and signaling by SEL1L-HRD1 ERAD
+
+Summary: Here we report that POMC-specific SEL1L-HRD1 ER-associated degradation is indispensable for leptin signaling in diet-induced obesity by controlling the turnover and maturation of nascent leptin receptor in the ER.
+
+Keywords: SEL1L-HRD1 ERAD, POMC, diet-induced obesity, leptin signaling, leptin receptor, parabiosis
+ABSTRACT
+
+Endoplasmic reticulum (ER) homeostasis in the hypothalamus has been implicated in the pathogenesis of certain patho-physiological conditions such as diet-induced obesity (DIO) and type 2 diabetes; however, the significance of ER quality control mechanism(s) and its underlying mechanism remain largely unclear and highly controversial in some cases. Moreover, how the biogenesis of nascent leptin receptor in the ER is regulated remains largely unexplored. Here we report that the SEL1L-HRD1 protein complex of the highly conserved ER-associated protein degradation (ERAD) machinery in POMC neurons is indispensable for leptin signaling in diet-induced obesity. SEL1L-HRD1 ERAD is constitutively expressed in hypothalamic POMC neurons. Loss of SEL1L in POMC neurons attenuates leptin signaling and predisposes mice to HFD-associated pathologies including leptin resistance. Mechanistically, newly synthesized leptin receptors, both wildtype and disease-associated human mutant Cys604Ser (Cys602Ser in mice), are misfolding prone and bona fide substrates of SEL1L-HRD1 ERAD. Indeed, defects in SEL1L-HRD1 ERAD markedly impair the maturation of these receptors and causes their ER retention. This study not only uncovers a new role of SEL1L-HRD1 ERAD in the pathogenesis of diet-induced obesity and central leptin resistance, but a new regulatory mechanism for leptin signaling.
+INTRODUCTION
+
+Hypothalamic neurons play important roles in the adaptation and mal-adaptation to pathophysiological conditions such as diet-induced obesity (DIO) and type-2 diabetes \(^{1-7}\). Homeostasis in the endoplasmic reticulum (ER) regulates many physiological processes such as systemic inflammation, inter-organelar crosstalk and mitochondrial dynamics \(^{8-14}\). It has been proposed that hypothalamic ER stress or unfolded protein response (UPR) may play a causal role in inflammation and leptin resistance in DIO and type-2 diabetes \(^{15-21}\). However, others have reported a protective role of UPR in similar experimental settings \(^{22,23}\). Hence, the significance of ER quality control pathways and its underlying mechanisms remain controversial.
+
+In addition to UPR that respond to misfolded proteins in the ER, ER-associated protein degradation (ERAD) is a constitutively active and highly conserved process responsible for recruiting unfolded or misfolded proteins in ER for cytosolic proteasomal degradation \(^{24-31}\). Among over a dozen of putative ERAD complexes, the SEL1L-HRD1 protein complex represents the most evolutionarily conserved ERAD branch where SEL1L/Hrd3p is an obligatory cofactor for the E3 ligase HRD1 \(^{28-30,32-34}\). Recent studies using cell type-specific SEL1L or HRD1 knockout mouse models have revealed the patho-physiological importance of SEL1L-HRD1 ERAD in a substrate-specific manner \(^{35-42}\). Particularly relevant to this study, SEL1L-HRD1 ERAD has been reported as indispensable for AVP and POMC neurons to control water balance and food intake via the maturation of prohormones, proAVP and POMC, respectively \(^{39,40}\). POMC neuron-specific *Sel1l* deletion leads to hyperphagia and age-associated obesity starting around 13 weeks of age when fed a low-fat chow diet \(^{39}\). Given the importance of POMC neurons in maintaining energy homeostasis under various nutritional status, one outstanding question is the relevance and significance of SEL1L-HRD1 ERAD in POMC neurons under pathophysiological conditions, including DIO.
+Here, we show that SEL1L-HRD1 ERAD in POMC neurons at the arcuate nucleus (ARC) of the hypothalamus, a key group of metabolic neurons that control food intake and energy expenditure \(^{43}\), controls DIO pathogenesis and leptin sensitivity via the regulation of leptin receptor biogenesis and signaling. Soon after weaning, POMC-specific Sel1l deficient (\(Sel1l^{POMC}\)) mice are hypersensitive to DIO. Much to our surprise, SEL1L-HRD1 ERAD is indispensable for the maturation of nascent leptin receptor to reach to the cell surface. Hence, SEL1L-HRD1 ERAD is a critical regulator of the maturation of leptin receptor in the ER and thereby leptin signaling in POMC neurons.
+
+RESULTS
+
+Transient upregulation of SEL1L-HRD1 ERAD expression in the hypothalamus in response to high fat diet (HFD) feeding.
+
+We previously showed that the SEL1L-HRD1 protein complex is constitutively expressed in the ARC of the hypothalamus \(^{39}\). Here we first asked whether its expression in the ARC region is regulated in response to overnutrition by placing the mice on 60% HFD (60% calories derived from fat) for 1 or 8 weeks. HFD feeding expectedly reduced the expression of Pomc, Npy and Agrp (Supplementary Fig. 1A), while enhance the protein levels of POMC derivatives β-Endorphin and α-MSH (Supplementary Fig. 1B-E). Moreover, HFD feeding enhanced neuronal activity in PVH region as measured by nuclear c-FOS following both 1- and 8-week HFD (Supplementary Fig. 1D, E). One-week HFD significantly induced \(Hrd1\) mRNA level, but not \(Sel1l\) mRNA level, while 8-week HFD feeding had no such effect (Fig. 1A). At the protein levels, both SEL1L and HRD1 proteins, were elevated at 1-week HFD, and returned to the basal levels after 8-week HFD (Fig. 1B, C), pointing to the transient response to SEL1L-HRD1 expression in the hypothalamus in response to HFD challenge. We next performed confocal microscopy to visualize SEL1L-HRD1 expression in the ARC regions in response to HFD. To visualize POMC neurons, we used POMC-eGFP transgenic mice where eGFP is under the control of POMC
+promoter \(^{39,44}\). SEL1L protein level was increased specifically in POMC neurons upon 1-week HFD, and returned to the basal level with prolonged HFD feeding (Fig. 1D, E). Similar observation was obtained for HRD1 protein levels in POMC neurons, but unlike SEL1L, HRD1 protein level was transiently upregulated in non POMC neurons as well (Fig. 1F, G). Hence, SEL1L-HRD1 expression in POMC neurons are responsive to acute, but not chronic, nutrient overload.
+
+Hypothalamic POMC-specific ERAD deficiency leads to early-set DIO and its pathologies.
+To delineate the significance of hypothalamic ERAD in DIO, we next characterized the phenotypes of \(Sel1L^{POMC}\) mice, generated by crossing \(Sel1L^{ff}\) with the Pomc-Cre mouse line \(^{39}\), following 8-week HFD feeding from 5 weeks of age. While, in line with our previous report, \(Sel1L^{POMC}\) mice appeared comparably to WT littermates in terms of body weight on chow diet for the first 13 weeks of age \(^{39}\) (Fig. 2A), \(Sel1L^{POMC}\) mice, both sexes, gained significantly more body weight soon after HFD feeding (Fig. 2A). Body composition analysis showed that fat content was significantly increased in \(Sel1L^{POMC}\) mice, reaching over 50% of body mass after 8-week HFD (Fig. 2B and Supplementary Fig. 2A) with more lipid deposition in the livers, as well as both white and brown adipose tissues (WAT and BAT) (Fig. 2C). \(Sel1L^{POMC}\) mice became highly glucose intolerant and insulin resistant following 8-week HFD (Fig. 2D, E), with elevated ad libitum and fasting blood glucose (Fig. 2F) and ad libitum insulin levels (Fig. 2G). In addition, glucagon and corticosterone levels were elevated in \(Sel1L^{POMC}\) mice (Supplementary Fig. 2B, C), while rectal temperature in \(Sel1L^{POMC}\) mice was decreased by 2 degrees compared to that of WT littermates (Supplementary Fig. 2D). Hence, we concluded that mice with POMC-specific ERAD defects exhibit early onset DIO and its pathologies including glucose and insulin resistance.
+
+Hypothalamic ERAD deficiency triggers hyperphagia and leptin resistance.
+We next explored the possible mechanism underlying the susceptibility to DIO in Sel1L^{POMC} mice. Sel1L^{POMC} mice consumed ~ 40% more food daily, i.e., hyperphagia, upon both 1- and 8-week HFD feeding (Fig. 3A). To directly demonstrate the direct causal link between food intake and weight gain, we performed pair feeding (giving the same amount of the food as WT littermates consume) following 8-week ad libitum HFD feeding. Sel1L^{POMC} mice gained weight quite rapidly under ad libitum feeding of HFD; however, their weight gain was significantly slowed down following pair-feeding and recovered when placed on ad libitum HFD feeding again (Fig. 3B). Indeed, weight gain of Sel1L^{POMC} mice was comparable to that of WT littermates if pair-feeding was performed at the beginning of HFD feeding (Fig. 3C). We then tested whether hyperphagia of Sel1L^{POMC} mice is caused by leptin resistance by leptin injection (Fig. 3D). Leptin injection was expected to induce body weight loss in WT mice, but not Sel1L^{POMC} mice. Indeed, unlike WT mice, Sel1L^{POMC} mice continued to gain body weight following leptin injection (Fig. 3D, E). This difference in body weight gain was likely due to the differences in food intake in response to leptin injection (Fig. 3F), pointing to a significant leptin resistance in Sel1L^{POMC} mice. Sel1L^{POMC} mice exhibited progressively marked hyperleptinemia with HFD feeding (Fig. 3G). Hence, we concluded that hypothalamic POMC neurons-specific ERAD deficiency triggers hyperphagia and leptin resistance.
+
+The effect of hypothalamic SEL1L-HRD1 ERAD in DIO is mediated by leptin resistance.
+To further establish the effect of leptin resistance in ERAD deficiency-associated DIO, we next performed parabiosis where two littermates were surgically stitched together to allow the sharing of the circulation (Fig. 4A). Following two weeks of recovery on chow diet, the parabionts WT: Sel1L^{POMC} (Group III) were placed on HFD for 8 weeks (Fig. 4A). Two control parabionts, WT: WT (Group I) and Sel1L^{POMC}:Sel1L^{POMC} (Group II), gained weight as expected with the latter pair becoming obese (Fig. 4B). However, in WT: Sel1L^{POMC} (Group III) parabionts, body weight gain for WT mice was attenuated compared to WT mice in WT: WT (Group I)
+control parabionts (P=0.08), while body weight gain for Sel1L^{POMC} mice was comparable to that of Sel1L^{POMC}:Sel1L^{POMC} parabionts (Group II) (Fig. 4B). Body compositions (i.e., lean vs. fat) in parabionts were not affected by the partner (Fig. 4C). Moreover, serum leptin and insulin levels were highly elevated in the Sel1L^{POMC} mice, but unaltered in WT mice regardless of the partners (Fig. 4D, E). Hence, these data suggested that hypothalamic SEL1L-HRD1 ERAD controls DIO pathogenesis via hyperleptinemia.
+
+Hypothalamic SEL1L-HRD1 ERAD deficiency impairs leptin-pSTAT3 signaling.
+We next asked how POMC-specific SEL1L-HRD1 ERAD regulates leptin sensitivity. As leptin signaling induces phosphorylation of STAT3 (pSTAT3), we next examined the levels of pSTAT3 in POMC neurons following leptin challenge. To visualize POMC neurons, we generated Sel1L^{POMC} mice on the POMC-eGFP background (Sel1L^{POMC};POMC-eGFP) ^{39,44}. HFD feeding progressively blunted leptin-induced pSTAT3 in the POMC neurons of the ARC region of WT mice, but to a much greater extent, in Sel1L^{POMC} mice (Fig. 5A-D and Supplementary Fig. 3). In keeping with the notion that pSTAT3 a critical transcription factor for the Pomc gene ^{45}, hypothalamic Pomc mRNA expression was markedly decreased in Sel1L^{POMC} mice with HFD (Fig. 5E). Moreover, Western blot analysis of pSTAT3 of the ARC region also showed a greater reduction of the percent of STAT3 being phosphorylated following HFD feeding (Fig. 5F, G).
+Thus, our data suggested that SEL1L-HRD1 ERAD in POMC neurons is vital for maintaining central leptin sensitivity during DIO pathogenesis.
+
+The effect of POMC-specific ERAD in DIO is likely uncoupled from UPR or inflammation.
+As ERAD deficiency expectedly causes the accumulation of unfolded/misfolded proteins in the ER that can potentially trigger UPR and given the reported role of UPR in DIO pathogenesis, we next tested whether ERAD deficiency activates UPR and if so, to what extent. There was no detectable activation of the PERK pathway as measured by phosphorylation of PERK and its
+downstream phosphorylation of eIF2α (Fig. 6A and Supplementary Fig. 4A). Phosphorylation of IRE1α, on the other hand, was moderately elevated in the ARC of Sel1LPOMC mice, so was the splicing of Xbp1 mRNA (a downstream effector of IRE1α) (Fig. 6B, C and Supplementary Fig. 4B, C). Consistently, ER chaperons BiP (an XBP1 target) was mildly elevated in the ARC of Sel1LPOMC mice (Fig. 6A and Supplementary Fig. 4A, D). In vitro, treatment with an ER stress inducer thapsigargin (Tg) induced strong ER stress, but failed to affect leptin signaling in WT HEK293T cells transfected with long isoform of mouse Leptin receptors (mLepRb) (Fig. 6D and Supplementary Fig. 4E), indicating that UPR is not sufficient to induce leptin resistance.
+
+Importantly, we found no significant POMC neuronal loss in the ARC of Sel1LPOMC;POMC-eGFP mice (Fig. 6E). Inflammatory markers were largely comparable in the ARC of Sel1LPOMC mice compared to those in WT littermates as measured by phosphorylation and protein levels of c-Jun N-terminal Kinase (JNK) as well as protein levels of I kappa B alpha (IkBα) (Fig. 6F, G). Chronic HFD feeding mildly increased astrogliosis in the ARC regions of both Sel1LPOMC and Sel1LPOMC mice as measured by both Western blot and immunofluorescence staining of astrocyte marker Glial Fibrillary acidic protein (GFAP) and/or microglia marker Ionized calcium-binding adaptor molecule 1 (IBA1) (Fig. 6F-H and Supplementary Fig. 4F). Taken together, these data demonstrate that Sel1L deficiency in POMC neurons triggers leptin resistance, independently of UPR, neuronal cell death and inflammation.
+
+SEL1L-HRD1 is required for the maturation of nascent leptin receptor (LepR).
+The forementioned data suggested that SEL1L-HRD1 ERAD regulates leptin sensitivity upstream of STAT3. To further explore the underlying mechanism, we generated leptin-responsive HEK293T cell system expressing the long isoform of LepR (LepRb) responsible for leptin-induced JAK2-STAT3 signaling \(^{46-49}\). Indeed, in line with decreased leptin sensitivity in vivo, mLepRb-positive HRD1\(^{−/−}\) HEK293T cells exhibited impaired phosphorylation of JAK2 and STAT3 compared to those in transfected \(WT\) cells in response to leptin stimulation (Fig. 7A, B).
+Surprisingly, the protein level of mLepRb was significantly higher in \( HRD1^{+/+} \) HEK293T cells compared to that of WT cells, under both serum-deprived and -supplemented conditions (Fig. 7B, C). Moreover, SEL1L interaction with in LepRb-transfected cells was markedly enhanced in \( HRD1^{+/+} \) cells where substrate-SEL1L interaction is known to be stabilized \( ^{29,50,51} \) (Fig. 7D, E). LepRb was ubiquitinated in an HRD1-dependent manner (Fig. 7F) and was significantly stabilized in \( HRD1^{+/+} \) cells compared to that in \( WT \) cells (Fig. 7G).
+
+We next assess the consequence of ERAD deficiency on LepRb maturation in the ER. Endoglycosidase H (EndoH) digestion, which cleaves asparagine-linked high mannose or hybrid glycans of the immature glycoproteins predominantly in ER \( ^{52} \), revealed significantly lower fraction of EndoH resistant form of LepRb that were able to exit the ER for complete maturation in \( HRD1^{+/+} \) HEK293T cells (Fig. 7H). This was further confirmed by surface biotinylation assay followed by immunoprecipitation with streptavidin-beads, which indicated reduced proportion of surface LepRb in \( HRD1^{+/+} \) cells (Fig. 7I and Supplementary Fig. 5A). Moreover, confocal microscopy following immunofluorescence staining further demonstrated an altered distribution of LepRb with increased intracellular, but decreased surface, expression in ERAD-deficient cells (Fig. 7J and Supplementary Fig. 5B). In the absence of SEL1L-HRD1, LepRb protein was prone to form high molecular weight aggregates via disulfide bonds (Fig. 7K) Taken together, our data show that SEL1L-HRD1 ERAD is required for the maturation of LepRb by targeting the misfolding-prone or misfolded LepRb for proteasomal degradation.
+
+SEL1L-HRD1 ERAD degrades and limits the pathogenicity of human LepRb Cys604Ser (C604S) mutant.
+
+To demonstrate the clinical relevance of our findings, we asked whether human LepRb (hLepRb) mutants \( ^{53,54} \) are SEL1L-HRD1 ERAD substrates. Here, we focused on hLepRb mutant C604S, a recessive point mutation due to missense homozygous substitution T > A at position 1810, identified in two brothers at 1- and 5-years old with severely early onset obesity
+C604-C674 forms a disulfide bond in human LepRb corresponding to C602-C672 in mouse LepRb (Fig. 8A, B) \(^{56-58}\). This mutation has been predicted as loss-of-function likely due to defects in folding \(^{54,56-58}\). C602S mLepRb significantly impaired leptin response compared to WT mLepRb in \(WT\) cells, which was further diminished in \(HRD1^{-/-}\) cells (Fig. 8C). Similar to WT mLepRb, C602S mLepRb was stabilized in the absence of HRD1 (Fig. 8D). Notably, C602S mLepRb readily formed HMW aggregates in \(WT\) HEK293T cells, and to much greater extent, in \(HRD1^{-/-}\) cells (Fig. 8E). Such aggregates likely formed in the ER as demonstrated by their colocalization with the ER chaperone BiP based on immunostaining (Fig. 8F-I). Hence, SEL1L-HRD1 ERAD is indispensable for the degradation of nascent WT and, at least a subset of, disease mutant LepRb, which ensures the maturation, trafficking and membrane display of functional LepRb.
+
+DISCUSSION
+
+This study not only identifies a novel regulatory mechanism for leptin receptor and signaling, but also reports a key role of hypothalamic ERAD in maintaining energy homeostasis under nutrient overload conditions. SEL1L-HRD1 ERAD defects in POMC neurons predispose mice to DIO and its pathologies, due to hyperphagia and hypothalamic leptin resistance. Our mechanistic studies establish LepRb as a bona fide endogenous substrate of SEL1L-HRD1 ERAD. Pointing to the clinical relevance of our findings, human recessive LepRb C604S variant is trapped in the ER and degraded by SEL1L-HRD1 ERAD (Fig. 9). In the absence of SEL1L-HRD1 ERAD, both WT and C604S LepRb are trapped in the ER in the form of HMW aggregates, with attenuated cell surface expression (Fig. 8E-I and Fig. 9). While this reported effect of ERAD in POMC neurons is in keeping with recent studies demonstrating the profound physiological importance of SEL1L-HRD1 ERAD in vivo \(^{39,40}\), it uncovers a novel function of SEL1L-HRD1 ERAD in leptin signaling and a novel regulatory mechanism for leptin biology.
+Our data show that hypothalamic SEL1L deficiency markedly increases the progression and pathogenesis of DIO in mice. Sel1L-deficient POMC neurons exhibit mild alterations in ER homeostasis including elevated activation of the IRE1α -XBP1 pathway and expression of ER chaperones, but without any detectable cell death. As previous studies have shown that deficiency of Ire1a or Xbp1 in POMC neurons predispose mice to DIO \(^{21}\), while gain-of-function of XBP1s in POMC neurons had an opposite effect \(^{23}\), we conclude that the effect of SEL1L-HRD1 ERAD is uncoupled from IRE1α -XBP1 pathway of the UPR and cell death, which is in line with many recent studies of various tissue-specific Sel1L- or Hrd1-deficient models \(^{37,39-42,59}\). These findings point to the cellular adaption in response to ERAD deficiency \(^{25}\). Such mild UPR activation and chaperone expression are potentially cyto-protective in response to the accumulation of misfolding proteins in the ER.
+
+Previous reports have suggested that UPR may play a causal role in leptin resistance due to impaired leptin signaling \(^{15,17,60}\). These studies were performed via the administration of ER stress inducers tunicamycin and thapsigargin which can be fraught with artefacts. Indeed, tunicamycin can inhibit glycosylation of the glycoproteins\(^{61}\) including LepRb, and thus the impaired leptin signaling can be directly due to defective glycosylation and concomitant functionality of LepRb instead of UPR activation as a general outcome of numerous dysregulation of glycoproteins. Further, high dosage of ER stress inducers included in previous studies may fall far from any physiological relevance\(^{15,17,60}\). In our study, thapsigargin treatment induced a range of ER stress response in a dose dependent manner, but failed to alter leptin signaling in *WT* HEK293T cells transfected with mLepRb even at the high level of UPR. Hence, collective evidence suggests that UPR is likely uncoupled from leptin signaling. The reason for these discrepancies remains unknown. Careful future studies are needed to validate either model.
+This study demonstrates an important role of SEL1L-HRD1 ERAD in leptin signaling, at least in part via the regulation of the maturation of nascent LepRb protein. We previously showed that SEL1L-HRD1 ERAD is required for the posttranslational maturation of POMC prohormone in mice on chow diet and that Sel1l deficiency in POMC neurons cause age-associate obesity in mice on chow diet due to the ER retention of POMC prohormone \(^{39}\). In DIO mouse models, we found defects in \(Sel1l^{POMC}\) mice occurring upstream of POMC transcription as leptin-induced STAT3 phosphorylation is impaired in the absence of SEL1L-HRD1 ERAD \(^{45-49}\). Further mechanistic studies identify partial loss-of-function of LepRb resulted from attenuated ER exit of nascent LepRb in SEL1L-HRD1 ERAD deficient cells. This study suggests that nascent LepRb protein is likely misfolding prone in the ER, likely due to multiple glycosylation and the formation of disulfide bonds, and hence relies on SEL1L-HRD1 ERAD to generate an ER environment conducive for the proper folding and conformation of bystander LepRb.
+
+Several human mutants have also been identified as SEL1L-HRD1 ERAD substrates that readily form aggregates and become resistant to and bypassing the quality control mediated by ERAD, leading to loss-of-function disease phenotype. These misfolded substrates with highly reactive cysteine thiols accumulate and promote the formation of inter- or intra-molecular disulfide-bonded aggregates \(^{39-41}\). Hence, SEL1L-HRD1 ERAD-mediated degradation of nascent unfolded and misfolded substrates, including LepRb in this study, may effectively prevent protein aggregation and maintain the folding environment in the ER. Efforts to target SEL1L-HRD1 ERAD function may represent a viable means for the treatment of certain diseases caused by a dominant-negative disease allele or a general collapse of the folding environment in the ER.
+
+METHODS
+Mice. As described previously 39, POMC-specific Sel1L-deficient mice (Sel1LPOMC) and control littermates (Sel1Ltf) were generated. The mice were further crossed with Pomc-eGFP reporter mice to generate Sel1LPOMC;POMC-eGFP and control littermates Sel1Ltf;POMC-eGFP. WT B6 mice were purchased from JAX and bred in our mouse facility. Mice were fed a chow diet (13% fat, 57% carbohydrate and 30% protein, PicoLab Rodent Diet 5L0D) and placed on a high-fat diet (HFD, calories provided by 60% fat, 20% carbohydrate and 20% protein, Research Diet D12492) from 5 weeks of age for 1 week or 8 weeks. All mice were housed in a temperature-controlled room with a 12-hour light/12-hour dark cycle.
+
+Food intake measurement and pair-feeding. Food intake were measured as previously described 39. Briefly, to perform daily food intake measurement, mice were first acclimatized to single housing 24 hours before the experiment. Daily food intake was measured 1 hour before the onset of the dark cycle each day. For the pair-feeding at later stage of HFD feeding, Sel1LPOMC and WT littermates had continuous free access to HFD for eight weeks and were then single housed and fed ~2.5 g, which was determined by the average of daily food intake of WT littermates, at the start of the dark cycle. For the pair-feeding at early stage of HFD feeding, 5-week-old Sel1LPOMC mice were split into two groups: One group of Sel1LPOMC and WT littermates had continuous free access to food; the other group of Sel1LPOMC mice (pair-fed) was fed ~2.5 g at the start of dark hours. Weekly bodyweight gains were monitored.
+
+Leptin treatment in mice. Twelve-week-old mice were intraperitoneally (i.p.) injected PBS followed by leptin (2 mg/kg body weight, R&D systems; catalog 498-OB-05M) 1 hour before the onset of dark cycle for three consecutive days as described 39. Body weight and food intake were monitored daily during the treatment period. For phosphorylated STAT3 staining, 2 mg/kg leptin were i.p. injected to mice, followed by overnight fasting. Mice were anesthetized by isoflurane for fixation-perfusion 30 min after injection.
+Tissue and blood collection. These procedures were carried out as previously described39. Briefly, blood was collected from anesthetized mice via cardiac puncture, transferred to 1.5ml microcentrifuge tubes, kept at room temperature for 30 minutes prior to centrifugation at 2,000 \( g \) for 15 minutes. Serum was aliquoted and stored at -80°C until analysis. For brain microdissection, Adult Mouse Brain Slicer Matrix (BSMAA001-1, Zivic Instruments) was used to collect coronal brain slices containing ARC region with further microdissection to obtain ARC-enriched region. All tissues were snap-frozen in liquid nitrogen and stored at -80°C before use.
+
+Preparation of brain sections. Mice were anesthetized with isoflurane, perfused with PBS followed by 4% paraformaldehyde (PFA) (Electron Microscopy Sciences; catalog 19210) for fixation. Brains were then postfixated in 4% PFA for overnight at 4°C, dehydrated in 15% sucrose and then 30% sucrose consecutively overnights at 4°C, and sectioned (30 \( \mu \)m) on a cryostat (Microm HM550 Cryostat, Thermo Fisher Scientific). The sections were stored in DEPC-containing anti-freezing media (50% 0.05 M sodium phosphate pH 7.3, 30% ethylene glycol, 20% glycerol) at –20°C. Different brain regions were identified using the Paxinos and Franklin atlas. Counted as distance from bregma, the following coordinates were used: PVN (–0.82 mm to –0.94 mm) and ARC (–1.58 mm to –1.7 mm).
+
+Western blot and antibodies. Frozen tissue or cells were homogenized by sonication in lysis buffer [150mM NaCl, 50mM Tris pH 7.5, 10 mM EDTA, 1% Triton X-100] with freshly added protease inhibitors (Sigma; catalog P8340), phosphatase inhibitors (Sigma; catalog P5726) and 10 mM N-ethylmaleimide (Thermo Scientific; catalog 23030). Lysates were incubated on ice for 30 min followed by centrifugation (13,000 g, 10 min at 4 °C). Supernatants were collected and analyzed for protein concentration using Bradford assay (Bio-Rad; catalog 5000006). For
+denaturing SDS-PAGE, samples were further supplied with 1mM DTT and denatured at 95°C for 5 min in 5x SDS sample buffer (250 mM Tris-HCl pH 6.8, 10% sodium dodecyl sulfate, 0.05% Bromophenol blue, 50% glycerol, and 1.44 M β-mercaptoethanol). For non-reducing SDA-PAGE, samples were prepared in 5x non-denaturing sample buffer (250 mM Tris-HCl pH 6.8, 10% sodium dodecyl sulfate, 0.05% bromophenol blue, 50% glycerol). For phostag gel analysis based on phos-tag system as described\(^{62,63}\), SDS-PAGE gel was supplemented by 50μM MnCl2 (Sigma) and 25μM phostag reagent (NARD Institute; catalog AAL-107) and must be protected from light until finishing running. Protein isolated from the liver of mice treated with tunicamycin (TM, 1 mg/kg, i.p.) for 24 hours was used as a positive control to indicate the position of phosphorylated PERK and IRE1a. For phosphatase treatment, 100 μg tissue lysates were incubated with 1 μl lambda phosphatase (\(λ\)PPase, New England BioLabs; catalog P0753S) in 1× PMP buffer (New England BioLabs; catalog B0761S) with 1 mM MnCl\(_2\) (New England BioLabs; catalog B1761S) at 30°C for 30 min. Reaction was stopped by adding 5× SDS sample buffer and incubated at 90°C for 5 min.
+
+All samples were incubated in 65°C for 10min and run with 15-30 μg total lysate on SDS-PAGE gel for separation followed by electrophoretic transfer to PVDF membrane (0.45μm, Millipore; catalog IPFL00010). The blots were incubated in 2% BSA/Tri-buffered saline tween-20 (TBST) with primary antibodies overnight at 4°C, washed with TBST followed by 1hr incubation with goat anti-rabbit or mouse IgG HRP at room temperature. Band density was quantitated using the Image Lab software on the ChemiDOC XRS+ system (Bio-Rad).
+
+Antibodies for Western blot were as follows: SEL1L (rabbit, 1:8000, Abclonal; catalog E112049), HRD1 (rabbit, 1:2000, ABclonal; catalog E15102), GRP78 BiP (rabbit, 1:5000, Abcam; catalog ab21685), HSP90 (rabbit, 1:5,000, Santa Cruz Biotechnology Inc.; catalog sc-7947), FLAG (mouse, 1:2000, Sigma-Aldrich; catalog F-1804), IRE1α (rabbit, 1:2,000, Cell Signaling Technology; catalog 3294), p-eIF2α (rabbit, 1:2000, Cell Signaling Technology; catalog 3597), eIF2α (rabbit, 1:2000, Cell Signaling Technology; catalog 9722), p-JNK (mouse, 1:2000, Cell
+Signaling Technology; catalog 9255), JNK (rabbit, 1:1000, Cell Signaling Technology; catalog 9252), PERK (Rabbit, 1:1000, Cell Signaling Technology; catalog 3192), pSTAT3 (Tyr705) (rabbit, 1:1000, catalog 9131, Cell Signaling Technology), STAT3 (rabbit, 1:1000, Cell Signaling Technology; catalog 9132), pJAK2 (Tyr1007/1008) (rabbit, 1:1000, Cell Signaling Technology; catalog 3771), JAK2 (rabbit, 1:1000, ABclonal; catalog A19629), Tubulin (mouse, 1:5000, Santa Cruz Biotechnology Inc.; catalog sc-5286), IkBα (rabbit, 1:1000, Cell Signaling Technology; catalog 9242) and IBA1 (rabbit, 1:1000, Proteintech; catalog 10904-1-AP)
+
+Secondary antibodies for Western blot were goat anti-rabbit IgG HRP and goat anti-mouse IgG HRP at 1:5,000, both from Bio-Rad.
+
+Immunostaining and antibodies. For fluorescent immunostaining in free-floating brain sections, samples were picked out of anti-freezing buffer followed by 3 washes with PBS. Free-floating sections were simultaneously incubated with primary antibodies in blocking buffer (0.3% donkey serum and 0.25% Triton X-100 in 0.1 M PBS) overnight at 4°C. Following 3 washes with PBS, sections were incubated with secondary antibodies for 2 hours at room temperature. Brain sections were then mounted on gelatin-coated slides (Southern Biotech; catalog SLD01-CS). Counterstaining and mounting were performed with mounting medium containing DAPI (Vector Laboratories; catalog H-1200) and Fisherfinest Premium Cover Glasses (Fisher Scientific; catalog 12-548-5P). For immunostaining in cells, 24 hours after transfection of LepRb-3xFLAG constructs, cells were placed on Poly-L-Lysine (Advanced Biomatrix; catalog 5048) coated Millicell EZ SLIDE 8-well glasses (Millipore; catalog PEZGS0816) for 24 hours before treatment and fixation. For staining surface bound leptin, samples were washed by ice cold PBS for 5 times and fixed by 4% formaldehyde (VWR; catalog 89370-094) for 15 minutes on ice followed by 3 washes with PBS. No permeabilization reagents were involved. For staining other markers, permeabilization was included and the overall process were the same as described above. To quantify immunoreactivity, identical acquisition settings were used for imaging each brain
+section from all groups within an experiment. The numbers of immunoreactivity-positive soma analysis and intensity of immunoreaction were quantified in 3D stack volumes after uniform background subtraction using the NIS Elements AR software (Nikon) and FIJI (National Institute of Health, USA).
+
+Antibodies for immunostaining were as follows: HRD1 (rabbit, 1:500, homemade), GRP78 BiP (rabbit, 1:500, Abcam; catalog ab21685), α-MSH (sheep, 1:2,000, Millipore; catalog AB5087), β-endorphin (rabbit, 1:2,000, Phoenix Pharmaceuticals; catalog H-022-33, provided by Carol Elisa), and GFP (chicken IgY, 1:300, Abcam; catalog ab13970), p-Y705 STAT3 (rabbit, 1:200, Cell Signaling Technology; catalog 9145), GFAP (rabbit, 1:500, Agilent; Z033429-2), FLAG (mouse, 1:500, Sigma-Aldrich; catalog F-1804), KDEL (rabbit, 1:500, Novus Biologicals; catalog NBP2-75549), eIF3η (goat, 1:500, Santa Cruz Biotechnology; catalog sc-16377).
+
+Secondary antibodies for fluorescent immunostaining (all 1:500) were as follows: Anti-rabbit IgG Alexa Fluor 647; anti-goat IgG Alexa Fluor 488 & 647; anti-sheep IgG Cy5 were from Jackson ImmunoResearch. Donkey anti-mouse IgG Alexa flour 555 was from Invitrogen (catalog A32773) and goat anti-chicken IgY FITC was from Aves Labs (catalog F-1005).
+
+Plasmids. Mouse LepRb cDNA was provided by Dr. Martin Myer at University of Michigan Medical School. The *LepRb* coding region was amplified by PCR using a primer set containing HindIII and XbaI restriction site at 5’ and 3’ respectively.
+
+F: 5'- CCG AAGCTT ATGATGTGTCAGAAATTCTATGTGGTT-3'
+R: 5'- TGC TCTAGA CACAGTTAAGTCACACATCTTTATT-3'
+
+Both PCR products and the backbone vector p3xFLAG-CMV14 were digested using HindIII and XbaI restriction enzymes in the double digestion system from New England BioLabs. For construction of LepRb point mutants, quick change mutagenesis was performed using PFU
+DNA polymerase (600140, Agilent). The following primers were used for mutagenesis to construct LepRb-C602S:
+F: 5'- CCTGCTGGGTGTCAGACCTCAGTGCAGTCTATG-3'
+R: 5'- CATAGACTGCACTGAGGTCTGACACCAGCAGG-3'
+
+CRISPR/Cas9-based knockout (KO) in HEK293T cells. HEK293T cells were cultured at 37°C with 5% CO₂ in DMEM with 10% fetal bovine serum (Fisher Scientific). To generate HRD1-deficient HEK293T cells, sgRNA oligonucleotides designed for human *HRD1* (5'-GGACAAAGGCCTGGATGTAC-3') was inserted into lentiCRISPR v2 (plasmid 52961, Addgene). Cells transfected with empty plasmids without sgRNA were used as wild type control. Cells grown in 10 cm petri dishes were transfected with indicated plasmids using 5μl 1 mg/ml polyethylenimine (PEI, Sigma) per 1μg of plasmids for HEK293T cells. Cells were cultured 24 hours after transfection in medium containing 2 μg/ml puromycin for 48 hours and then in normal growth media.
+
+Statistics. Results are expressed as the mean ± SEM unless otherwise stated. Statistical analyses were performed in GraphPad Prism version 8.0 (GraphPad Software Inc.). Comparisons between the groups were made by unpaired two-tailed Student's t test for two groups, or one-way ANOVA or two-way ANOVA followed by multiple comparisons test for more than two groups. *P* value < 0.05 was considered as statistically significant. All experiments were repeated at least twice and/or performed with several independent biological samples, and representative data are shown.
+
+Study Approval. All experiments performed with mice were in compliance with University of Michigan (Ann Arbor, MI) Institutional Animal Care and Use Committee (#PRO00006888) guidelines.
+Data and material availability. The materials and reagents used are either commercially available or available upon the request, with detailed information included in Methods. The predicted structure of mLepRb is available at AlphaFold ID AF-P48356-F1. All data supporting the findings and materials for the manuscript are available within the article and the Supplementary Information.
+AUTHOR CONTRIBUTION
+
+H.M. and G.H.K. designed the most of experiments and H.M., with the help of G.H.K., performed most of the experiments and data analysis. H.M., with the help of G.H.K., wrote the methods and figure legends. L.Q. and H.M. wrote the manuscript. All authors have approved the manuscript.
+ACKNOWLEDGEMENTS
+
+We thank Drs. Richard Wojcikiewicz and Martin Myers for reagents; Drs. Peter Arvan, Carol Elias and Daniel Klionsky for critical comments and suggestions, and members of the Qi and Arvan laboratories for comments and technical assistance. This work was supported by NIH grants 1R01DK11174 (to P.A. and L.Q.), 1R01DK105393, 1R01DK120047, and American Diabetes Association (ADA) 1-19-IBS-235 (to L.Q.).
+1 McLean, F. H. et al. A high-fat diet induces rapid changes in the mouse hypothalamic proteome. Nutr Metab (Lond) **16**, 26, doi:10.1186/s12986-019-0352-9 (2019).
+2 Dalvi, P. S. et al. High fat induces acute and chronic inflammation in the hypothalamus: effect of high-fat diet, palmitate and TNF-alpha on appetite-regulating NPY neurons. Int J Obes (Lond) **41**, 149-158, doi:10.1038/ijo.2016.183 (2017).
+3 Beutler, L. R. et al. Obesity causes selective and long-lasting desensitization of AgRP neurons to dietary fat. eLife **9**, doi:10.7554/eLife.55909 (2020).
+4 Horvath, T. L. et al. Synaptic input organization of the melanocortin system predicts diet-induced hypothalamic reactive gliosis and obesity. Proc Natl Acad Sci U S A **107**, 14875-14880, doi:10.1073/pnas.1004282107 (2010).
+5 Souza, G. F. et al. Defective regulation of POMC precedes hypothalamic inflammation in diet-induced obesity. Sci Rep **6**, 29290, doi:10.1038/srep29290 (2016).
+6 Poon, K. Behavioral Feeding Circuit: Dietary Fat-Induced Effects of Inflammatory Mediators in the Hypothalamus. Front Endocrinol (Lausanne) **11**, 591559, doi:10.3389/fendo.2020.591559 (2020).
+7 Veloso, L. A. & Schwartz, M. W. Altered hypothalamic function in diet-induced obesity. Int J Obes (Lond) **35**, 1455-1465, doi:10.1038/ijo.2011.56 (2011).
+8 Zhang, K. & Kaufman, R. J. From endoplasmic-reticulum stress to the inflammatory response. Nature **454**, 455-462, doi:10.1038/nature07203 (2008).
+9 Chaudhari, N., Talwar, P., Parimisetty, A., Lefebvre d'Hellencourt, C. & Ravanan, P. A molecular web: endoplasmic reticulum stress, inflammation, and oxidative stress. Front Cell Neurosci **8**, 213, doi:10.3389/fncel.2014.00213 (2014).
+10 Zhang, Y. et al. Synergistic mechanism between the endoplasmic reticulum and mitochondria and their crosstalk with other organelles. Cell Death Discov **9**, 51, doi:10.1038/s41420-023-01353-w (2023).
+11 Wu, H., Carvalho, P. & Voeltz, G. K. Here, there, and everywhere: The importance of ER membrane contact sites. Science **361**, doi:10.1126/science.aan5835 (2018).
+12 Kornmann, B. et al. An ER-mitochondria tethering complex revealed by a synthetic biology screen. Science **325**, 477-481, doi:10.1126/science.1175088 (2009).
+13 Rowland, A. A. & Voeltz, G. K. Endoplasmic reticulum-mitochondria contacts: function of the junction. Nat Rev Mol Cell Biol **13**, 607-625, doi:10.1038/nrm3440 (2012).
+14 Marchi, S., Paterngani, S. & Pinton, P. The endoplasmic reticulum-mitochondria connection: one touch, multiple functions. Biochim Biophys Acta **1837**, 461-469, doi:10.1016/j.bbabio.2013.10.015 (2014).
+15 Ozcan, L. et al. Endoplasmic reticulum stress plays a central role in development of leptin resistance. Cell Metab **9**, 35-51, doi:10.1016/j.cmet.2008.12.004 (2009).
+16 Purkayastha, S. et al. Neural dysregulation of peripheral insulin action and blood pressure by brain endoplasmic reticulum stress. Proc Natl Acad Sci U S A **108**, 2939-2944, doi:10.1073/pnas.1006875108 (2011).
+Ramirez, S. & Claret, M. Hypothalamic ER stress: A bridge between leptin resistance and obesity. FEBS Lett 589, 1678-1687, doi:10.1016/j.febslet.2015.04.025 (2015).
+Schneeberger, M. et al. Mitofusin 2 in POMC neurons connects ER stress with leptin resistance and energy imbalance. Cell 155, 172-187, doi:10.1016/j.cell.2013.09.003 (2013).
+Ye, Z., Liu, G., Guo, J. & Su, Z. Hypothalamic endoplasmic reticulum stress as a key mediator of obesity-induced leptin resistance. Obes Rev 19, 770-785, doi:10.1111/obr.12673 (2018).
+Zhang, X. et al. Hypothalamic IKKbeta/NF-kappaB and ER stress link overnutrition to energy imbalance and obesity. Cell 135, 61-73, doi:10.1016/j.cell.2008.07.043 (2008).
+Yao, T. et al. Ire1alpha in Pomp Neurons Is Required for Thermogenesis and Glycemia. Diabetes 66, 663-673, doi:10.2337/db16-0533 (2017).
+Xiao, Y. et al. Knockout of inositol-requiring enzyme 1alpha in pro-opiomelanocortin neurons decreases fat mass via increasing energy expenditure. Open Biol 6, doi:10.1098/rsob.160131 (2016).
+Williams, K. W. et al. Xbp1s in Pomp neurons connects ER stress with energy balance and glucose homeostasis. Cell Metab 20, 471-482, doi:10.1016/j.cmet.2014.06.002 (2014).
+Friedlander, R., Jarosch, E., Urban, J., Volkwein, C. & Sommer, T. A regulatory link between ER-associated protein degradation and the unfolded-protein response. Nat Cell Biol 2, 379-384, doi:10.1038/35017001 (2000).
+Qi, L., Tsai, B. & Arvan, P. New Insights into the Physiological Role of Endoplasmic Reticulum-Associated Degradation. Trends Cell Biol 27, 430-440, doi:10.1016/j.tcb.2016.12.002 (2017).
+Travers, K. J. et al. Functional and genomic analyses reveal an essential coordination between the unfolded protein response and ER-associated degradation. Cell 101, 249-258, doi:10.1016/s0092-8674(00)80835-1 (2000).
+Hwang, J. & Qi, L. Quality Control in the Endoplasmic Reticulum: Crosstalk between ERAD and UPR pathways. Trends Biochem Sci 43, 593-605, doi:10.1016/j.tibs.2018.06.005 (2018).
+Carvalho, P., Goder, V. & Rapoport, T. A. Distinct ubiquitin-ligase complexes define convergent pathways for the degradation of ER proteins. Cell 126, 361-373, doi:10.1016/j.cell.2006.05.043 (2006).
+Gardner, R. G. et al. Endoplasmic reticulum degradation requires lumen to cytosol signaling. Transmembrane control of Hrd1p by Hrd3p. J Cell Biol 151, 69-82 (2000).
+Hampton, R. Y., Gardner, R. G. & Rine, J. Role of 26S proteasome and HRD genes in the degradation of 3-hydroxy-3-methylglutaryl-CoA reductase, an integral endoplasmic reticulum membrane protein. Mol Biol Cell 7, 2029-2044, doi:10.1091/mbc.7.12.2029 (1996).
+Bhattacharya, A. & Qi, L. ER-associated degradation in health and disease - from substrate to organism. J Cell Sci 132, doi:10.1242/jcs.232850 (2019).
+Vashistha, N., Neal, S. E., Singh, A., Carroll, S. M. & Hampton, R. Y. Direct and essential function for Hrd3 in ER-associated degradation. Proc Natl Acad Sci U S A **113**, 5934-5939, doi:10.1073/pnas.1603079113 (2016).
+
+Wu, X. & Rapoport, T. A. Mechanistic insights into ER-associated protein degradation. Curr Opin Cell Biol **53**, 22-28, doi:10.1016/j.ceb.2018.04.004 (2018).
+
+Schoebel, S. *et al.* Cryo-EM structure of the protein-conducting ERAD channel Hrd1 in complex with Hrd3. *Nature* **548**, 352-355, doi:10.1038/nature23314 (2017).
+
+Sha, H. *et al.* The ER-associated degradation adaptor protein Sel1L regulates LPL secretion and lipid metabolism. *Cell Metab* **20**, 458-470, doi:10.1016/j.cmet.2014.06.015 (2014).
+
+Wu, S. A. *et al.* The mechanisms to dispose of misfolded proteins in the endoplasmic reticulum of adipocytes. *Nat Commun* **14**, 3132, doi:10.1038/s41467-023-38690-4 (2023).
+
+Bhattacharya, A. *et al.* Hepatic Sel1L-Hrd1 ER-associated degradation (ERAD) manages FGF21 levels and systemic metabolism via CREBH. *EMBO J* **37**, doi:10.15252/embj.201899277 (2018).
+
+Wei, J. *et al.* HRD1-ERAD controls production of the hepatokine FGF21 through CREBH polyubiquitination. *EMBO J* **37**, doi:10.15252/embj.201898942 (2018).
+
+Kim, G. H. *et al.* Hypothalamic ER-associated degradation regulates POMC maturation, feeding, and age-associated obesity. *J Clin Invest* **128**, 1125-1140, doi:10.1172/JCI96420 (2018).
+
+Shi, G. *et al.* ER-associated degradation is required for vasopressin prohormone processing and systemic water homeostasis. *J Clin Invest* **127**, 3897-3912, doi:10.1172/JCI94771 (2017).
+
+Yoshida, S. *et al.* Endoplasmic reticulum-associated degradation is required for nephrin maturation and kidney glomerular filtration function. *J Clin Invest* **131**, doi:10.1172/JCI143988 (2021).
+
+Shrestha, N. *et al.* Sel1L-Hrd1 ER-associated degradation maintains beta cell identity via TGF-beta signaling. *J Clin Invest* **130**, 3499-3510, doi:10.1172/JCI134874 (2020).
+
+Toda, C., Santoro, A., Kim, J. D. & Diano, S. POMC Neurons: From Birth to Death. *Annu Rev Physiol* **79**, 209-236, doi:10.1146/annurev-physiol-022516-034110 (2017).
+
+Bumaschny, V. F. *et al.* Obesity-programmed mice are rescued by early genetic intervention. *J Clin Invest* **122**, 4203-4212, doi:10.1172/JCI62543 (2012).
+
+Munzberg, H., Huo, L., Nilini, E. A., Hollenberg, A. N. & Bjorbaek, C. Role of signal transducer and activator of transcription 3 in regulation of hypothalamic proopiomelanocortin gene expression by leptin. *Endocrinology* **144**, 2121-2131, doi:10.1210/en.2002-221037 (2003).
+
+Liu, H., Du, T., Li, C. & Yang, G. STAT3 phosphorylation in central leptin resistance. *Nutr Metab (Lond)* **18**, 39, doi:10.1186/s12986-021-00569-w (2021).
+
+Baumann, H. *et al.* The full-length leptin receptor has signaling capabilities of interleukin 6-type cytokine receptors. *Proc Natl Acad Sci U S A* **93**, 8374-8378, doi:10.1073/pnas.93.16.8374 (1996).
+Chen, H. et al. Evidence that the diabetes gene encodes the leptin receptor: identification of a mutation in the leptin receptor gene in db/db mice. Cell **84**, 491-495, doi:10.1016/s0009-8674(00)81294-5 (1996).
+
+Uotani, S., Bjorbaek, C., Tornoe, J. & Flier, J. S. Functional properties of leptin receptor isoforms: internalization and degradation of leptin and ligand-induced receptor downregulation. Diabetes **48**, 279-286, doi:10.2337/diabetes.48.2.279 (1999).
+
+Plempner, R. K. et al. Genetic interactions of Hrd3p and Der3p/Hrd1p with Sec61p suggest a retro-translocation complex mediating protein transport for ER degradation. *J Cell Sci* **112** (*Pt 22*), 4123-4134, doi:10.1242/jcs.112.22.4123 (1999).
+
+Sun, S. et al. Sel1L is indispensable for mammalian endoplasmic reticulum-associated degradation, endoplasmic reticulum homeostasis, and survival. *Proc Natl Acad Sci U S A* **111**, E582-591, doi:10.1073/pnas.1318114111 (2014).
+
+Cao, L. et al. Global site-specific analysis of glycoprotein N-glycan processing. *Nat Protoc* **13**, 1196-1212, doi:10.1038/nprot.2018.024 (2018).
+
+Nunziata, A. et al. Functional and Phenotypic Characteristics of Human Leptin Receptor Mutations. *J Endocr Soc* **3**, 27-41, doi:10.1210/js.2018-00123 (2019).
+
+Saeed, S. et al. Genetic variants in LEP, LEPR, and MC4R explain 30% of severe obesity in children from a consanguineous population. *Obesity (Silver Spring)* **23**, 1687-1695, doi:10.1002/oby.21142 (2015).
+
+Saeed, S. et al. High morbidity and mortality in children with untreated congenital deficiency of leptin or its receptor. *Cell Rep Med* **4**, 101187, doi:10.1016/j.xcrm.2023.101187 (2023).
+
+Peelman, F., Zabeau, L., Moharana, K., Savvides, S. N. & Tavernier, J. 20 years of leptin: insights into signaling assemblies of the leptin receptor. *J Endocrinol* **223**, T9-23, doi:10.1530/JOE-14-0264 (2014).
+
+Moharana, K. et al. Structural and mechanistic paradigm of leptin receptor activation revealed by complexes with wild-type and antagonist leptins. *Structure* **22**, 866-877, doi:10.1016/j.str.2014.04.012 (2014).
+
+Tsirigotaki, A. et al. Mechanism of receptor assembly via the pleiotropic adipokine Leptin. *Nat Struct Mol Biol* **30**, 551-563, doi:10.1038/s41594-023-00941-9 (2023).
+
+Zhou, Z. et al. Endoplasmic reticulum-associated degradation regulates mitochondrial dynamics in brown adipocytes. *Science* **368**, 54-60, doi:10.1126/science.aay2494 (2020).
+
+Hosoi, T. et al. Endoplasmic reticulum stress induces leptin resistance. *Mol Pharmacol* **74**, 1610-1619, doi:10.1124/mol.108.050070 (2008).
+
+Heifetz, A., Keenan, R. W. & Elbein, A. D. Mechanism of action of tunicamycin on the UDP-GlcNAc:dolichyl-phosphate Glc-NAc-1-phosphate transferase. *Biochemistry* **18**, 2186-2192, doi:10.1021/bi000578a008 (1979).
+
+Qi, L., Yang, L. & Chen, H. Detecting and quantitating physiological endoplasmic reticulum stress. *Methods Enzymol* **490**, 137-146, doi:10.1016/B978-0-12-385114-7.00008-8 (2011).
+Yang, L. et al. A Phos-tag-based approach reveals the extent of physiological endoplasmic reticulum stress. PLoS One 5, e11621, doi:10.1371/journal.pone.0011621 (2010).
+Fig. 1: Transient upregulation of SEL1L-HRD1 ERAD expression in the hypothalamus in response to high fat diet (HFD) feeding.
+(A) Quantitative PCR (qPCR) analysis of Sel1L and Hrd1 mRNA levels in the arcuate nucleus (ARC) of the C57BL/6J male mice fed on normal chow diet (NCD), 1w- and 8w-HFD (n=3-4 mice per group).
+(B-C) Representative Western blot of SEL1L and HRD1 in the ARC of the C57BL/6J male mice fed on NCD, 1w- and 8w-HFD, with quantitation shown on the right (n=13-15 mice per group).
+(D-E, F-G) Representative images and quantitation of IF staining of SEL1L (D-E) and HRD1 (F-G) in the ARC of POMC-eGFP mice fed NCD, or HFD for 1-week or 8-week (n=3-4 mice per group, 70-100 POMC and non-POMC cells respectively per mouse). Yellow arrows, GFP-positive POMC neurons; White arrows, GFP-negative non-POMC neurons.
+Values, mean ± SEM. ns., not significant; *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001 by one-way ANOVA followed by Tukey’s multiple comparisons test (A, C, E, G).
+Fig. 2: Hypothalamic POMC-specific ERAD deficiency leads to early-set DIO and its pathologies.
+(A) Growth curve of Sel1Lff and Sel1LPOMC mice, male (left) and female (right), fed on NCD (open symbols/dotted lines) or HFD (solid symbols/lines) (n=18-24 per group for male mice, n=10-16 per group for female mice).
+(B) Body composition of Sel1Lff and Sel1LPOMC male mice after 8w-HFD (n=4-7 mice per group).
+(C) H&E images of peripheral tissues from male mice fed HFD for 8 weeks (n=3 mice per group). iWAT and gWAT, inguinal and gonadal white adipose tissues; BAT, brown adipose tissues.
+(D-E) Glucose tolerance (D) and insulin tolerance tests (E) in male mice fed HFD for 8 weeks. Mice were fasted for 16 or 6 hours prior to glucose (2 g/kg body weight) or insulin (1 unit/kg body weight) injection, respectively (n=6 mice per group).
+(F) Serum glucose in 8w-HFD male mice, either ad-lib or after 6h-fasting (n=7-10 mice per group).
+(G) Insulin levels in 8w-HFD male mice under ad-lib condition (n=5-6 mice per group).
+Values, mean ± SEM. ns, not significant; *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001 by two-way ANOVA followed by multiple comparisons test (A-B, D-F) or two-tailed Student’s t-test (G).
+Fig. 3: Hypothalamic ERAD deficiency triggers hyperphagia and leptin resistance.
+(A) Daily food intake of male Sel1L\textsuperscript{ff} and Sel1L\textsuperscript{POMC} mice at 1w- and 8w-HFD (n=9-11 mice per group).
+(B) Growth curve of male Sel1L\textsuperscript{POMC} mice fed with either NCD or HFD under ad libitum or pair feeding as indicated (n=3 mice per group, blue solid circles). Male Sel1L\textsuperscript{ff} mice fed ad libitum with the same diets were included as controls (n=3 mice per group, black open circles)
+(C) Growth of Sel1L\textsuperscript{POMC} male mice with either ad libitum or pair-feeding of HFD starting at 5 weeks of age (n=3-5 mice per group).
+(D) Body weights of 12-week-old mice put on HFD (at day 0) followed by daily i.p. injected with vehicle (PBS) and leptin (2 mg/kg body weight) for 3 days (n=2 per group for male mice, indicated in dots; n=2-3 per group for female mice, indicated in squares).
+(E-F) Percentage of body weight change (E), average daily food intake (F) following 3 daily vehicle and leptin injections of the mice (n=2 per group for male mice, indicated in dots; n=2-3 per group for female mice). % Body weight is calculated based on the body weights at the end point over those at the starting point for each treatment.
+(G) Serum leptin levels in mice fed on NCD, 1w- and 8w-HFD (n=5-13 mice per group).
+Values, mean ± SEM. ns, not significant; *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001 by two-way ANOVA followed by multiple comparisons test (A-G).
+Fig. 4: Hypothalamic SEL1L-HRD1 deficiency leads to DIO via leptin signaling.
+(A) Schematic diagram for parabiosis and pictures (right) of Sel1Ltf and Sel1LPOMC female mice after parabiosis HFD for 8 weeks (n=3 pairs in group I, n=1 pair in group II, n=5 pairs in group III).
+(B-C) Body weights (B) of mice before and after parabiosis and body composition (C) after parabiosis following 8-week HFD for 8 weeks (n=6 mice in group I, n=2 mice in group II, n=5 mice per genotype in group III).
+(D-E) Serum leptin (D) and insulin (E) levels of mice after parabiosis HFD for 8 weeks (n=6 mice in group I, n=2 mice in group II, n=5 mice per genotype in group III).
+Values, mean ± SEM. ns, not significant; *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001 by two-way ANOVA followed by multiple comparisons test (B-E).
+Fig. 5: Hypothalamic SEL1L-HRD1 ERAD deficiency impairs leptin-pSTAT3 signaling.
+(A-D) Representative immunofluorescence (IF) staining of pSTAT3 in Sel1Lft;POMC-eGFP and Sel1LPOMC;POMC-eGFP mice at NCD (A), 1w-HFD (B) and 8w-HFD (C), with quantitation shown in D. Mice were fasted for overnight (16hrs) and administrated with leptin (i.p., 2 mg/kg body weight) for 30 min (n=3-4 mice per group). Yellow arrows, pSTAT3 positive POMC neurons; White arrows, pSTAT3 negative POMC neurons. PBS-injected mice were included as negative controls and shown in Supplementary Fig. 3.
+(E) Quantitative PCR (qPCR) analysis of Pomc mRNA expression levels in ARC of Sel1Lft and Sel1LPOMC mice at 8w-HFD (n=3 mice per group).
+(F-G) Representative Western blot for pSTAT3 in ARC of Sel1Lft and Sel1LPOMC mice at NCD or 8w-HFD, injected with leptin or PBS for 30 min (n=4 male mice per group), with quantitation shown in G.
+Values, mean ± SEM. ns, not significant; *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001 by two-way ANOVA followed by multiple comparisons test (D, E, G).
+Fig. 6: The effect of POMC-specific ERAD in DIO is likely uncoupled from UPR and inflammation.
+
+(A) Representative Western blot for the PERK pathway of UPR in the ARC of Sel1Lff and Sel1LPOMC mice fed on 8w-HFD (n=6 mice per group with 3 male mice and 3 female), with quantitation shown on the right. Livers of mice treated with tunicamycin (TM, 1 mg/kg, i.p.) for 24 hours (Liver_TM) or not (Liver_CON), as well as lysates treated with Lambda protein phosphatase, included as controls.
+(B) Phostag gel (P-T)-based Western blot for IRE1α phosphorylation in the ARC of Sel1Lff and Sel1LPOMC mice fed on 8w-HFD, with quantitation shown on the right (n=3 mice per group with 2 male and 1 female).
+(C) Reverse transcriptase PCR (RT-PCR) analysis of Xbp1 mRNA splicing (u, unspliced; s, spliced) in ARC of Sel1Lff and Sel1LPOMC mice fed on 8w-HFD (n=2-3 male mice and n=2-3 female mice per group), with quantitation shown on the right. ARC of mice treated with tunicamycin (TM, 1 mg/kg, i.p.) for 24 hours (ARC_TM) included as a positive control.
+(D) Representative assays for UPR and pSTAT3 in mLepRb-transfected HEK293T treated with leptin with/without Thapsigargin (Tg) (n=5 independent cell samples for SDS-PAGE gel, n=3 for P-T gel, two independent repeats for RT-PCR).
+(E) Representative confocal images of the number of GFP-expressing POMC neurons in Sel1Lff;POMC-eGFP and Sel1LPOMC;POMC-eGFP mice after 8w-HFD, with quantitation shown on the right (n=6-9 mice per group).
+(F-G) Representative Western blot analysis of inflammatory markers in the ARC of Sel1Lff and Sel1LPOMC mice fed on 8w-HFD, with quantitation shown in G (n=3 mice per group).
+(H) Representative confocal images of GFAP, a marker of astrocytes, in the ARC of male Sel1Lff and Sel1LPOMC mice fed on 8w-HFD (n=3 mice per group).
+Values, mean ± SEM. ns, not significant; *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001 by two-way ANOVA followed by multiple comparisons test (D) or two-tailed Student’s t-test (A-C, E, G).
+A
+Empty   LepRb-3xFLAG
+Leptin
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+
+B
+LepRb-3xFLAG
+pJAK2/JAK2 (a.u.)
+pSTAT3/STAT3 (a.u.)
+pSTAT3/JAK2 (a.u.)
+Serum deprived
+FLAG/HSP90 (a.u.)
+WT   HRD1+/-   PBS   Leptin
+p-0.06
+
+C
+Empty   LepRb-3xFLAG
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+WT   HRD1+/-
+
+D
+IP: FLAG Input
+WT   HRD1+/-
+WT   HRD1+/-
+LepRb-3xFLAG
+SEL1L
+HRD1
+HSP90
+
+E
+IP: IgG SEL1L Input
+WT   HRD1+/-
+WT   HRD1+/-
+LepRb-3xFLAG
+SEL1L
+HRD1
+Tubulin
+
+F
+Input   IP: FLAG
+WT   HRD1+/-
+WT   HRD1+/-
+Ub
+LepRb-3xFLAG
+HSP90
+HRD1
+
+G
+BFA CHX 0 2 4 0 2 4 Hrs
+WT   HRD1+-
+LepRb-3xFLAG
+HSP90
+LepRb (% of initial)
+WT   HRD1+-
+CHX (hr)
+
+H
+EndoH PNGase
+WT   HRD1+-
+LepRb-3xFLAG
+SEL1L
+HSP90
+% EndoR Resistant
+WT   HRD1+-
+
+I
+100 nM Leptin
+Total Surface
+WT   HRD1+-
+LepRb-3xFLAG
+HSP90
+HRD1
+Surface/Total 1.051±0.06
+
+J
+WT   HRD1+-
+DAPI
+LepRb
+10 μm
+
+K
+Normeducing
+WT   HRD1+-
+HMW-Monomer
+FLAG
+HSP90
+HRD1
+Reducing
+WT   HRD1+-
+HMW/HSP90 (a.u.)
+HMW/Monomer (a.u.)
+Fig. 7: SEL1L-HRD1 is required for the maturation of nascent LepRb.
+(A-B) Representative Western blot analysis for pJAK2, pSTAT3 and LepRb in HEK293T transfected with or without mLepR, treated with or without leptin (A), with quantitation shown in (B) (n=4-7 individual cell samples per group).
+(C) Representative Western blot analysis of mLepRb protein levels in mLepRb-transfected HEK293T in complete medium (DMEM w/ 10% FBS), with quantitation shown on the right (n=8 individual cell samples per group).
+(D-E) Representative Western blot analysis of interaction between SEL1L-HRD and mLepRb following immunoprecipitation (IP) of Flag (D) or SEL1L (E) from lysates of mLepRb-transfected HEK293T (n=2-3 individual cell samples).
+(F) Representative Western blot analysis of Ub following denaturing immunoprecipitation (IP) of Flag from lysates of mLepRb-transfected HEK293T, with quantitation shown on the right (n=3 individual cell samples per group).
+(G) Representative Western blot analysis of LepRb protein decay in LepRb-transfected HEK293T cells co-treated with protein trafficking inhibitor Brefeldin-A and translation inhibitor cycloheximide (CHX) for the 0, 2 and 4 hours, with quantitation shown below (n=4 individual cell samples per group).
+(H) Representative Western blot analysis of LepRb glycosylation in LepRb-transfected HEK293T with EndoH and PNGase treatment, with quantitation shown on the right (n=3 individual cell samples per group).
+(I) Representative Western blot analysis of mLepRb membrane display by surface biotinylation and streptavidin-bead pull down assay in mLepRb-transfected HEK293T treated with leptin. T, total lysate; S, surface fraction. (n=2 individual cell samples per group).
+(J) Representative IF images of LepRb in mLepRb-transfected HEK293T treated with leptin, with quantitation of %surface signals over total shown on the right (n=28 cells per genotype from 3 independent repeats).
+(K) Reducing and non-reducing SDS-PAGE and Western blot analysis of LepRb high molecular-weight aggregates of LepRb in mLepRb-transfected WT and HRD1-/- HEK293T, with quantitation shown on the right (n=3 individual cell samples per group).
+Values, mean ± SEM. ns, not significant; *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001 by two-tailed Student’s t-test (A, C, F, H, J, K) or two-way ANOVA followed by multiple comparisons test (B, G).
+A
+C602-672 in mouse
+C604-674 in human
+LUMENAL/EXTRACELLULAR
+CYTOSOL
+1 Leptin receptor isoform B (LepRb)
+1162 in mouse
+1164 in human
+
+B
+WT LepRb
+Fibronectin Type III
+Transmembrane
+WT LepRb CYS602
+CYS672
+C602S LepRb
+CYS672
+SER602
+
+C
+LepRb-WT   LepRb-C602S
+WT HRD1+/−   WT HRD1+/−
+Leptin - + - + - +
+FLAG
+pSTAT3
+STAT3
+HRD1
+HSP90
+
+D
+LepRb-WT   LepRb-C602S
+WT HRD1+/−   WT HRD1+/−
+0 1 2 0 1 2 CHX (hr)
+FLAG
+HSP90
+HRD1
+LepRb levels (a.u.)
+ns
+*
+*
+WT
+HRD1+/−
+WT
+HRD1+/−
+
+E
+Empty   LepRb-WT   LepRb-C602S
+WT HRD1+/−   WT HRD1+/−
+Nonreducing
+Reducing
+HMW
+Monomer
+HSP90
+HRD1
+FLAG
+HSP90
+HRD1
+HMW/Monomer (a.u.)
+***
+***
+WT
+HRD1+/−
+WT
+HRD1+/−
+
+F
+WT
+LepRb-WT
+LepRb-C602S
+DAPI
+Leprb-3xFLAG BIP
+10μm
+10μm
+
+G
+%Surface/Total (a.u.)
+WT
+HRD1+/−
+LepRb-WT
+LepRb-C602S
+*
+
+H
+Pearson's coefficient values
+Fraction of LepRb-3xFLAG overlapping BIP
+WT
+HRD1+/−
+LepRb-WT
+LepRb-C602S
+***
+
+I
+Fig. 8: SEL1L-HRD1 ERAD degrades and limits the pathogenicity of LepRb Cys602Ser disease mutant.
+
+(A) Schematic diagram of mouse LepRb. “SP”, Signal Peptide; “TM”, Transmembrane. Star symbols, N-glycosylation sites; Green lines, disulfide bonds.
+(B) Structural modeling of mouse LepRb by AlphaFold2. Red arrow, location of human mutation C604S (mouse C602S).
+(C) Representative Western blot analysis for pSTAT3 in HEK293T transfected with mLepRb-WT or mLepRb-C602S with or without leptin treatment, with quantitation shown below (n=3 individual cell samples per group).
+(D) Representative Western blot analysis of LepRb protein decay in WT and HRD1−/− HEK293T transfected with mLepRb-WT or -C602S, treated with brefeldin-A and cycloheximide (CHX) for the 0, 1 and 2 hours, with quantitation shown below (n=4 individual cell samples per group).
+(E) Reducing and non-reducing SDS-PAGE and Western blot analysis of LepRb high molecular-weight (HMW) aggregates of LepRb in WT and HRD1−/− HEK293T transfected with mLepRb-WT or -C602S, with quantitation shown on the right (n=6 individual cell samples per group).
+(F-I) Representative IF images of mLepRb-WT and -C602S in transfected WT and HRD1−/− HEK293T cells (F) with quantitation %surface signals over total (G) (n=11-17 cells per group) and analysis of co-localization of LepRb with BiP signals by Pearson correlation coefficient (H) and Manders overlap coefficient (I) (n=10-14 cells per group).
+Values, mean ± SEM. ns, not significant; *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001 by two-way ANOVA followed by multiple comparisons test (C, D, E, G, H, I).
+Fig.9: Proposed models for SEL1L-HRD1 ERAD degradation of wildtype LepRb and C604S disease mutant.
+In the basal conditions, SEL1L-HRD1 ERAD constitutively degrades misfolded LepRb and ensures the proper folding, maturation and surface expression of the LepRb. In the absence of ERAD, the accumulation of misfolded receptors forms aggregates, interferes with the folding and maturation of the nascent LepRb with attenuated surface display. In the context of recessive LepRb C604S mutant, though degraded by SEL1L-HRD1 ERAD, C604S LepRb readily forms aggregates to the extent beyond the capacity of ERAD, resulting in impaired maturation and surface display of the receptors.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• HFDPKOSupplementaryNC.pdf

04b173197c545f10945642520e847eff78301d76d085be0937577c7b3426ca9d/peer_review/peer_review.md

@@ -0,0 +1,441 @@
+Peer Review File
+
+Hydrogen evolution with hot electrons on a plasmonic-molecular catalyst hybrid system
+
+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/.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+In this work, the authors have been developed a system consisting of NiO/Au/[CoII(phen-NH2)2(H2O)2] for photocatalytic hydrogen evolution. The reaction mechanism in this system could be explained by situ and ultrafast spectroscopic measurements. However, several issues still remain controversial and need to be solved. Therefore, I do not recommend the publication of this work.
+Q1. It was claimed that the CoII(phen-NH2)2(H2O)2 was connected on the surface of AuNP. However, no evidence was provided. Could you provide some evidence? What is the interaction? A shift of XPS Au 4f spectra may help to reveal the interaction. In addition, the amino groups are difficult to be bounded on AuNP due to the steric effect.
+Q2. The authors applied the composite structure of NiO/Au/[CoII(phen-NH2)2(H2O)2] for photocatalytic hydrogen evolution. However, no data on HER rate was provided. In addition, the HER rate of this reaction system needs to be compared with that reported in other photocatalytic systems.
+Q3. This article repeatedly mentions in the abstract, conclusion, and introduction that this catalytic process is not related to local thermal effects. However, in Line 30-31, Page 6, from the results of photocurrent(Figure 2C), it is concluded that the results are related to heating effects from the illumination and plasmonic decay. This is clearly a conflict. Please give an explanation. On the other hand, an increase in temperature may definitely speed up the reaction based on basic chemistry.
+Q4. As you said, Lu group reported a composite system of plasmonic AuNPs and molecular catalysts. In this report, both plasmonic hot carriers and localized thermal effects contribute to the efficient photocatalytic systems. Your statement on this report is not accurate. Please correct the expression in the text.
+Q5. Can the authors provide morphology of the AuNPs sprayed onto NiO films? Previous literature has reported that the morphology of AuNPs could change significantly after high-temperature annealing (Langmuir, 2012, 28, 25, 9885.) So it is necessary to provide the morphology of the AuNPs before and after thermal annealing.
+Q6. All NAP-XPS spectra need to be reanalyzed after peak deconvolution. The position of the binding energy of the element is accurate only when the peaks were properly deconvoluted.
+Q7. Can the authors clarify the absence of the peak of XPS Co 2p? The peak shift of Co 2p was not evidenced by the results. Therefore, the mechanism in Figure 5 is based on guess not evidence.
+Q8. The results in Figure 2a, claimed a stepwise reduction process. However, no further evidence was provided. In addition, the results can also be explained by electrochemical reduction of acetic acid.
+Q9. It was claimed that the disappearance of the amino bonds in FTIR evidence the binding of amino group on AuNPs. This claim is not accurate. The disappearance of this peak can also be explained by the small amount of the molecules adsorbed on AuNPs. This explanation seems more possible. It is better to show the peaks of other vibrational bands from adsorbed molecules.
+Q10. How the molecules were adsorbed on AuNPs? After the dipping, was the substrate rinsed with solvent to remove excess molecules?
+Q11. In Figure 2a-b, the unit for x axis need to be clarified. Was this voltage referenced to Ag|AgCl or standard hydrogen potential?
+Q12. On page 6, it was claimed that no hydrogen was detected in absence of molecular catalyst? However, no data was provided.
+Q13. It was claimed that the XPS O 1s peak at 532.4 eV can be attributed to the liquid water from thin electrolyte layer. However, this may also, at least partially, come from the acetic acid added in the system.
+Q14. The data of the Au catalyst in the control group is incomplete, so readers cannot clearly understand the comparison of photocatalytic performance of the newly synthesized catalyst. I advise an additional set of experiments of the effect of light in the chronoamperometry of the Au applying -0.65V potential in 3 mM acetic acid and 0.1M LiCl is suggested (pH = 3.5) is shown in Figure 2C.
+Q15. There is a problem with the abbreviation of the phrase "proton-coupled electron transfers" (CEPT) in the text, which should be (PCET). This error also appears in Figure (5), which easily affects readers' reading, please note the correction.
+Q16. The writing needs further polishing.
+
+Reviewer #2 (Remarks to the Author):
+
+This manuscript explores the combination of a plasmonic particle with a molecular complex for hydrogen generation under plasmon excitation. The system is nicely design and the authors used a very elegant combination of conventional characterization techniques with ultrafast and NAP-XPS, which are less standard in this type of studies. As such they proposed that plasmonic hot-carriers are involved in the process and they discarded thermal effects due to the low thermal stability of the molecular complex which seems not to degrade. Overall this is an interesting study, that besides the results, presents interesting combination of techniques to further debate the role of plasmons in enhanced catalytic processes. I think this manuscript could be of potential interest after sorting some key points:
+
+Why do the authors neglect the effect of enhanced near-fields that can promote electronic excitations in the molecular complex (i.e. resembling homogeneous photocatalysis processes)? From the abstract, introduction and discussion the dispute is heat or hot-carriers but the enhanced fields are never discussed nor considered.
+
+I am also wondering if (under a field-driven scenario), the e-ph coupling times could also be explained without needing the charge-transfer.
+
+Also, it has been shown that the highly concentrated electric fields can change the water molecules adsorption orientation and reactivity, modifying reaction energy barriers. As such, the hypothesis that thermal water-splitting needs more than 500 degrees, I'm not sure if still holds under high e-fields conditions.
+Figure 2 should have the illumination conditions.
+
+The thermal argument is debatable. I agree with the authors, but it's also true that they should show (by external heating) the degradation of the molecular complex (to support the idea that under high temperatures the catalyst breaks down).
+
+The NAP-XPS and TAS measurements are very nice and not usually found together in the same manuscript. So I thank the authors for trying to do a very deep mechanistic understanding with non-conventional (or not the most commonly used) techniques.
+
+I think the scheme in figure 5 can be highly enhance. For instance, the authors could put the times measured (instead of "very fast"). We all know these processes are very fast. Saying only very fast is not new/relevant. Also, where are the holes in that scheme? I think omitting the NiO doesn't help to understand the mechanism. The first arrow is the plasmon excitation and decay (I guess, because it says nothing).
+
+Reviewer #3 (Remarks to the Author):
+
+The authors report a plasmonic photocatalytic system for plasmon-assisted electrochemical hydrogen evolution and use photoelectrochemistry, transient spectroscopy, and XPS to provide evidence for the electron and hole transfer from plasmonic Au nanoparticles to NiO hole acceptor and Co-catalyst as an electron acceptor. Using transient spectroscopies and XPS, the authors provide evidence for the charge transfer from Au nanoparticles to holes and electron acceptors. However, the following concerns remain unaddressed and I recommend this study to be published after major revision.
+
+1) In the introduction, the authors state that “it remains challenging to disentangle charge carrier catalysis from photothermal effects”. I agree with the authors here however, the same challenge still persists in the current study. Even though the authors have provided evidence for the charge transfer catalysis, authors have not provided any evidence to prove that photothermal effects are not playing a role in the photocatalytic system reported by the authors. The authors state that the photocatalytic system is not stable at which the water’s thermolysis takes place, which is why photothermal catalysis may not be occurring in their experiments. Even a small increase in the temperature due to photothermal heating can decrease the activation energy and can catalyze the electrochemical hydrogen evolution. The temperature does not have to reach 500 – 2000C (as mentioned by the authors in the manuscript) for photothermal heating to catalyze electrochemical hydrogen evolution. Hence, due to the lack of evidence to prove that photothermal heating is not participating in the electrochemical HER in the photocatalytic system reported in this study, both charge transfer and heating are likely to be playing the role in the photocatalytic HER reported by the authors unlike claimed otherwise by the authors. Authors are requested to include relevant discussion.
+2) To understand the relative contribution of the charge carriers and photothermal heating in catalyzing electrochemical HER, using a photocatalytic system involving non-plasmonic Au (smooth Au film) may help. NiO/smooth Au film/Co-Cat (non-plasmonic substrate) when used for the photocatalytic experiments, smooth Au being non-plasmonic, laser light excitation will mainly lead to photothermal heating/interband transition of Au and it would possible to only understand the contribution of only photothermal heating and this seems to be the primary motivation behind the study.
+
+3) Although the authors provide chronoamperometry data for the photoelectrocatalytic HER (Figure 2C) with light illumination on and off, CV data reporting electrocatalytic HER (with NiO/AuNPs/Co-Cat) with and without light illumination should be provided. Monitoring the onset potential for HER in conditions with lights on and off will provide further insights into the charge transfer process. Reporting the above-mentioned CV experiment with different light intensities will also provide additional evidence for the charge transfer.
+
+4) Hot electron generation and transfer are dependent on the wavelength of the light. The authors have performed experiments only at a single wavelength of light. Control experiments reporting photoelectrochemistry and transient spectroscopies using light of wavelength which off-resonance (for example 642 nm which does not excite the plasmon resonance) of the Au particles should be reported. Comparing results obtained at 550 nm excitation (already included in the manuscript) with at least one off-resonance wavelength excitation can provide additional insights.
+
+5) Ex-situ transient spectroscopies make it clear that including NiO in a photocatalytic system alters e-e scattering, and e-ph scattering lifetimes, however, For Figures 2B and 2D, data for only FTO/Au/Co-Cat (no NiO) should also be provided to prove that NiO is playing during photoelectrochemical HER.
+
+6) Was the Argon atmosphere maintained during CV experiments reported in Figure 2B? If not, peak -0.5 V (present in both CVs NiO/Au/Co-Cat and NiO/Au), may also correspond to the oxygen reduction. Authors are requested to include the discussion regarding the same.
+
+7) In Figure 2D, NiO/Au also produced photoelectrochemical current, rather in the opposite way. The authors are requested to include a discussion regarding this.
+
+8) In the proposed mechanism section, authors state that the hot holes are transferred to the NiO and react at the counter electrode leading to O2 production. Authors are requested to provide data for this statement and include relevant discussion.
+
+9) Regarding the characterization of the catalytic system:
+a) Only DLS measurements for the Au NPs are provided. DLS measurements are generally carried out in a colloidal state, however, in the photocatalytic system reported in this study, Au NPs are drop cast on NiO/FTO and then annealed. Drop casting and annealing of the drop casted Au NPs at 500C will result in Au NPs aggregation which may shift the LSPR of the Au NPs. SEM images of Au NPs on FTO before and after annealing should be provided for thoroughness.
+b) In Figure 2A, the authors state that no distinguishable peaks were reported before the catalytic wave which relates to Co-complex redox behavior. However, there is a clear reduction (-1.2 V in acid, -1.4 V in water) and oxidation peak in both CVs (water and acid) which are uncharacteristic of HER because in HER, oxidation peak on the reverse sweep is not usually observed (Figure 2C inset). Hence, out of two catalytic waves, the first catalytic wave is unlikely from HER and may be related to the Co-Cat oxidation-reduction behavior. Authors are requested to provide further discussion regarding this. Further, was the Co-cat free in the solution or deposited on the glassy carbon? Authors are requested to include this information in the SI.
+c) Figure 3D reports the use of Au/ligand. No procedure for preparing Au/ligand (no Co) is reported in the paper. Authors are requested to provide the method of preparation in the SI.
+d) Au nanoparticles are prepared using tannic acid. What is the purpose of using tannic acid as Au NPs of the same size can be prepared using citrate? Please include the relevant discussion for better clarity.
+
+10) Regarding transient spectroscopy:
+a) The lifetime of all the scattering processes changed after including Co-Cat in the photocatalytic system. Short pulses used for the spectroscopy may generate a lot of local heat which may damage the photocatalytic system especially the organic compound Co-complex. Can this decomposition or damage to the catalyst have an impact on the scattering process lifetime? Please include relevant discussion. FTIR, SEM, and UV Vis of the catalytic system after irradiating short laser pulses should be included to ensure the integrity of the catalyst and make sure that the decomposition of the catalyst (if at all) is not the reason behind the scattering process’s lifetime change.
+b) Why was the 490 nm winglet considered for the analysis and not the other winglet? Please provide the discussion.
+
+11) It would be helpful for the readers if you can please provide pictures of the electrodes, electrochemical cell and experimental set up.
+Answers to the Reviewers’ comments
+
+Reviewer #1:
+In this work, the authors have been developed a system consisting of NiO/Au/[Coll(phen-NH2)2(H2O)2] for photocatalytic hydrogen evolution. The reaction mechanism in this system could be explained by situ and ultrafast spectroscopic measurements. However, several issues still remain controversial and need to be solved. Therefore, I do not recommend the publication of this work.
+
+REPLY: We thank the Reviewer for the comprehensive revision and detailed feedback, which helped us improve significantly the quality of the manuscript. We have addressed all the comments below and changed the manuscript and SI accordingly. We also what to thank the Reviewer for the time dedicated to the manuscript revision. We hope the revised version suppressed the earlier reluctance and can be considered for publication.
+
+Q1. It was claimed that the Coll(phen-NH2)2(H2O)2 was connected on the surface of AuNP. However, no evidence was provided. Could you provide some evidence? What is the interaction? A shift of XPS Au 4f spectra may help to reveal the interaction. In addition, the amino groups are difficult to be bounded on AuNP due to the steric effect.
+
+REPLY: We thank the Reviewer for the suggestion. We used infrared to demonstrate that the catalyst was bounded to the Au. We see the disappearance of the NH2 bands (Fig 1C insert) only when Au is present, corroborating our expectations. Additionally, the LSPR maximum shifted, suggesting that the surrounding dielectric medium changed when the catalyst was added. Finally, the XPS analysis requested by the Reviewer showed the Co signal is only present when Au is present, confirming anchoring selectivity.
+The XPS analysis of the Au 4f revealed a doublet related to a single Au specie in all the samples (Au-Cat, NiO-Au, and NiO-Au-Cat). The Au 4f7/2 binding energy for all of them was found to be 84.05 eV, consistent with Au in metallic form. The single doublet suggests that Co 2p induced electronic changes are not localized, confirming good electronic coupling. Since the added catalyst was very small, the induced electronic changes couldn’t be quantified.
+
+Actions taken:
+We updated the manuscript with the following statements:
+The Au LSPR peak shifts to lower energy when the cobalt catalyst is added (Figure 1C), confirming the anchoring. Note that the LSPR peak absorption is sensitive to the surrounding dielectric medium. Consequently, the surface modification by the catalyst should induce a shift in the LSPR maximum as absorbed. Additionally, it is possible to see the complex absorption shoulder located at 370 nm, corroborating the attachment between catalyst and Au NPs. Unfortunately, the glass support (FTO or cover glass) covers the rest of the complex UV-Vis band precluding their measurement.
+It is worth mentioning that the catalyst only bonded when Au NPs were present according to Co 2p XPS, confirming the selective anchoring of the catalyst to the Au NPs via the amino groups. The Co 2p3/2 of Au/Co-cat and NiO/Au/Co-cat measured in vacuum, at a prominent peak is centred at around 780.5 eV, consistent with Co is oxidation +2. The Au 4f7/2 for the samples Au/Co-cat., NiO/Au and NiO/Au/Co-cat. appeared at 84.0 eV (figure S9), consistent with gold in the metallic state. There was only one species in all the samples, coherent with the idea that any electronic change due to NiO and Co-catalyst is delocalized over all the gold atoms, suggest good electronic coupling.
+The anchoring was also confirmed by the disappearance of the amino bands in the infrared (Figure 1C insert). Before anchoring the complex has three small peaks located between 3450-3300 cm^{-1} and 3250-3200 cm^{-1} associated with N-H bending modes of primary amino groups. The correspondent bending modes between 1650-1580 cm^{-1} are also visible but somewhat overlapped by the water O-H bending mode. After attaching the catalyst to the Au NPs, the N-H bands disappear. The complete disappearance suggests that the catalyst coordinates to the Au NPs via both 1,10-Phenanthrolin-5-amine ligands, as shown schematically in Figure 1D. Note that the infrared bands were normalized to the C-N stretch at 1280 cm^{-1} intensity to enable
+direct comparison. The C-N band is unaffected by the attachment, making it suitable for normalization. Unfortunately, the formed Au-N bonds are not infrared active, and the low loading prevented their detection with Raman spectroscopy. Figure 1D shows the a chematic representation of the complete photosystem.
+
+- We added Figure S9, showing Au 4f XPS of Au/Co-cat., NiO/Au and NiO/Au/Co-cat. to the supporting information.
+
+Q2. The authors applied the composite structure of NiO/Au/[Coll(phen-NH2)2(H2O)2] for photocatalytic hydrogen evolution. However, no data on HER rate was provided. In addition, the HER rate of this reaction system needs to be compared with that reported in other photocatalytic systems.
+REPLY: We thank the Reviewer for the suggestion. In the original version of the manuscript, we had the photocurrent per area, which we confirmed by online QMS to be due to hydrogen production. We updated the manuscript with the rate (average of 3.1 nmol/(min*cm^2)). We would like to reiterate that the aim of the manuscript is to demonstrate the direct use of hot carriers in the photocatalytic process. Moreover, because this is a new catalyst and its loading is purposely low, we think it is inappropriate to draw comparisons with unrelatable systems.
+
+Q3. This article repeatedly mentions in the abstract, conclusion, and introduction that this catalytic process is not related to local thermal effects. However, in Line 30-31, Page 6, from the results of photocurrent(Figure 2C), it is concluded that the results are related to heating effects from the illumination and plasmonic decay. This is clearly a conflict. Please give an explanation. On the other hand, an increase in temperature may definitely speed up the reaction based on basic chemistry.
+REPLY: We thank the Reviewer for the comment. What we wanted to say the drift in the baseline is related to heating effects, not the entire light-induced photocurrent signal. We updated the phrase to convey this. Heat cannot justify the increase in photocurrent upon illumination, since the magnitude of the baseline drift is much smaller than the photocurrent. This is a limitation of running experiments under longer illumination, which was mitigated by doing short on/off cycles (see Figure 2C). Nevertheless, we understand the Reviewer’s natural confusion, and thus the reason for updating the text.
+
+Q4. As you said, Lu group reported a composite system of plasmonic AuNPs and molecular catalysts. In this report, both plasmonic hot carriers and localized thermal effects contribute to the efficient photocatalytic systems. Your statement on this report is not accurate. Please correct the expression in the text.
+REPLY: We thank the Reviewer for the comment.
+We updated the manuscript with the following statement:
+"Moreover, the authors suggested a cooperative result between plasmon hot carriers and localized thermal effects for which this study does not have evidence."
+
+Q5. Can the authors provide morphology of the AuNPs sprayed onto NiO films? Previous literature has reported that the morphology of AuNPs could change significantly after high-temperature annealing (Langmuir, 2012, 28, 25, 9885.) So it is necessary to provide the morphology of the AuNPs before and after thermal annealing.
+REPLY: We thank the Reviewer for the comment. Au NPs morphology after annealing is the same according to the AFM analysis (figure S6), which shows an average particle size in the range of 10 nm as expected from DLS analysis. Small particle aggregation cannot be discarded but is not the prevalent morphology. Regarding the NiO, the AFM shows that the deposition of Au and subsequent annealing did not affect its morphology (figure S7). Moreover, the Ni 2p XPS showed no significant change in the Ni binding energy to suggest that the addition of Au affected the NiO significantly from an electronic perspective (figure S8). This is not surprising since the supplier (Solaronix) stated that films can be annealed up to 700 C.
+Actions taken:
+- We added Figure S6, S7 and S8 to the supporting information.
+
+Q6. All NAP-XPS spectra need to be reanalyzed after peak deconvolution. The position of the binding energy of the element is accurate only when the peaks were properly deconvoluted.
+REPLY: We thank the reviewer for this comment. We agree that the shifts of the O 1s shown in the manuscript (Figures 4A and S8A, insets) should be supported by spectra deconvolutions. Indeed, in the first version of the manuscript, we slightly overestimated the shift of the main peak of O 1s. The correct shifts are now reported in the insets of Figure 4A and Figure S8A. Upon fitting of the O 1s, peak components’ assignments were amended in the main text. We deconvoluted all spectra. As suggested by this reviewer, fitting results gave us precise peak positions and a correct estimate of the peak shift upon application of the voltage. In the revised version, we show deconvoluted spectra in Figure S15.
+In the case of Co 2p, because (as written in the manuscript) the S/N is low and the spectral shape broad, a deconvolution would not be meaningful. Line shapes and peak positions are comparable to other reports of cobalt complexes, and this supports our qualitative description. Furthermore, Co 2p peaks position and shape do not change with the applied voltage. In the manuscript, we never report absolute binding energy values. We identify the binding energy of the main peak centroid.
+
+Actions taken:
+- We replaced the insets of Figures 4A in the manuscript
+
+![O 1s and Co 2p XPS spectra under acidic conditions at different potentials](page_320_670_1109_370.png)
+
+- We replaced the insets of Figures S14A in the supporting information
+
+![O 1s and Co 2p XPS spectra under acidic conditions at different potentials](page_320_1100_1109_370.png)
+
+- We added Figure S15, showing O 1s peak deconvolutions, to the supporting information.
+- We added the Table S1 in the methods section describing XPS measurements in the supporting information.
+
+<table>
+  <tr>
+    <th>Position (eV)</th>
+    <th>FWHM</th>
+    <th>% L-G</th>
+  </tr>
+  <tr>
+    <td>529.50</td>
+    <td>1.75</td>
+    <td>0</td>
+  </tr>
+  <tr>
+    <td>531.00</td>
+    <td>1.85</td>
+    <td>25</td>
+  </tr>
+  <tr>
+    <td>532.68 – 532.75</td>
+    <td>1.75</td>
+    <td>0</td>
+  </tr>
+  <tr>
+    <td>534.80 – 535.25</td>
+    <td>0.90</td>
+    <td>0</td>
+  </tr>
+</table>
+
+We added to the supporting text:
+"Deconvolution of the O 1s spectra was performed after removal of a Shirley background. Gaussian and Voigt-shaped components, whose position set according to past literature reports, were used to obtain the best correlation with experimental data (see Figure S15). Fitting parameters (peak positions, full width at half maximum –FWHM- and % of Lorentian-Gaussian) are summarized in Table S1."
+
+We modified the manuscript text with:
+"Three prominent peaks, centered at approximately 531.00 eV, 532.70 eV and 534.90 eV, are assigned to adsorbed hydroxyls, liquid water (thin electrolyte film on top of the electrode) and gas phase water, respectively (fitting parameters are reported in Table S1). The slight shift of the O 1s main peak measured between pure electrolyte and the same in the presence of acid (from 532.68 to 532.75 eV under OCP conditions) may be due to the presence of acetic acid, whose contribution falls within the spectral range of liquid water peak. A fourth peak, centered at 529.50 eV, was detected in the absence of acid and assigned to lattice oxygen (electrode). Such a difference between the two conditions (no lattice oxygen detected in the presence of acid) suggests that the liquid electrolyte film in the absence of acid was thinner than in the presence of acid."
+
+Q7. Can the authors clarify the absence of the peak of XPS Co 2p? The peak shift of Co 2p was not evidenced by the results. Therefore, the mechanism in Figure 5 is based on guess not evidence.
+Reply: we believe that this reviewer refers to the low intensity/noisy signal of Co 2p. Indeed, as reported in Figures 4B and S8B and mentioned in the text, such a signal has always been detected. The low intensity of Co 2p is due to photoelectrons attenuation through the liquid layer of electrolyte stabilized on top of the electrode. The spectra shown in Figures 4B and S8B
+are the sum of more than 100 iterations, corresponding to several hours of acquisition. Due to limited beamtime duration, we had to stop measurements in order to acquire a complete dataset.
+Concerning the mechanism, the reviewer is correct: no peak shift of Co 2p was detected upon voltage application by means of in situ XPS measurements. We justify this in the main text, on page 11: “Figure 4B also shows that at -0.65 V vs Ag/AgCl the cobalt signal decreases in intensity. Such a potential is sufficient for the evolution of H2 since bubbles were detected at the electrode. Thus, the decrease in signal suggests that some of the cobalt centres undergo fast reduction and protonation, leading to hydrogen evolution. However, since these are expected to be transient species, one cannot capture them due to our low signal-to-noise data and temporal resolution. Nevertheless, their presence can be deduced by the drop in the Co2+ signal intensity.”
+To better highlight the presence of Co 2p, and explain the low S/N ratio, we modified the manuscript text with:
+“Despite the low signal-to-noise ratio, due to Co 2p attenuation through the liquid electrolyte layer stabilized on the WE, the main features of Co 2p peaks can be detected on both investigated electrodes (see figures 4B and S8B).”
+
+Q8. The results in Figure 2a, claimed a stepwise reduction process. However, no further evidence was provided. In addition, the results can also be explained by electrochemical reduction of acetic acid.
+REPLY: We thank the Reviewer for the comment. We remeasured several times figure 2a, ensuring that no oxygen was present in the solution. It became clear that peak was related to oxygen reduction and not catalyst reduction, which is consistent with NAP-XPS. The figure was updated accordingly.
+
+Q9. It was claimed that the disappearance of the amino bonds in FTIR evidence the binding of amino group on AuNPs. This claim is not accurate. The disappearance of this peak can also be explained by the small amount of the molecules adsorbed on AuNPs. This explanation seems more possible. It is better to show the peaks of other vibrational bands from adsorbed molecules.
+REPLY: We thank the Reviewer for the comment. The Reviewer is correct in suggesting that low loading of the catalyst could make detection of -NH2 bands. To mitigate this limitation the FTIR signals were normalized to the C-N band at 1280 cm^{-1}, enabling direct comparison between unbound and bounded catalyst (see new figure 2C insert).
+Upon binding, one expects the formation of Au-N bond, which is not active in the mid FTIR region. The bond is active in Raman but the low loading of catalyst made it impossible its detection. Unfortunately, the catalyst has no active FTIR bands that can be used to confirm the attachment.
+
+Actions taken:
+- We rechecked the FTIR and UV-Vis analyses and performed additional XPS to confirm the attachment. See the reply to Q1 for more information.
+
+Q10. How the molecules were adsorbed on AuNPs? After the dipping, was the substrate rinsed with solvent to remove excess molecules?
+REPLY: We thank the Reviewer for the comment. After attaching, the electrode was rinsed several times with the solvent to remove any excess of catalyst.
+We updated the supporting information with the following statement:
+“The system was rinsed with water several times to remove unbounded catalyst molecules.”
+Q11. In Figure 2a-b, the unit for x axis need to be clarified. Was this voltage referenced to Ag|AgCl or standard hydrogen potential?
+REPLY: We thank the Reviewer for the comment. The potential is versus Ag/AgCl as stated in the figure caption. However, we took the Reviewer suggestion and updated the figures to make this clear.
+
+Q12. On page 6, it was claimed that no hydrogen was detected in absence of molecular catalyst? However, no data was provided.
+REPLY: We thank the Reviewer for the comment. The QMS signal for the system without catalyst was the same as the H2 baseline and thus we could confidently state that no hydrogen was produced. We opted against showing that because it is not common to show zero signals. However, to satisfy the Reviewer request we added the O2 trace to Figure S4.
+
+Q13. It was claimed that the XPS O 1s peak at 532.4 eV can be attributed to the liquid water from thin electrolyte layer. However, this may also, at least partially, come from the acetic acid added in the system.
+REPLY: we thank the reviewer for the suggestion. We did not explicitly consider acetic acid in the discussion because its concentration, around 40 μM, is very low. In the revised version of the manuscript, we address this comment. We refer to Q6: “The slight shift of the O 1s main peak measured between pure electrolyte and the same in the presence of acid (from 532.68 to 532.75 eV under OCP conditions) may be due to the presence of acetic acid, whose contribution falls within the spectral range of liquid water peak.”
+We added a reference that supports our claim (J. Phys. Chem. A 2013, 117, 401–409).
+
+Q14. The data of the Au catalyst in the control group is incomplete, so readers cannot clearly understand the comparison of photocatalytic performance of the newly synthesized catalyst. I advise an additional set of experiments of the effect of light in the chronoamperometry of the Au applying -0.65V potential in 3 mM acetic acid and 0.1M LiCl is suggested (pH = 3.5) is shown in Figure 2C.
+REPLY: We thank the Reviewer for the comment. We added the requested data to Figure 2D. It is visible that only when the complete system is present we get significant photocurrents.
+
+Q15. There is a problem with the abbreviation of the phrase "proton-coupled electron transfers" (CEPT) in the text, which should be (PCET). This error also appears in Figure (5), which easily affects readers’ reading, please note the correction.
+REPLY: We thank the Reviewer for the comment. The Reviewer is correct if the abbreviation was for proton-coupled electron transfer but in the present case it is for concerted proton-coupled electron transfer that is commonly abbreviated as CEPT.
+
+Q16. The writing needs further polishing.
+REPLY: We thank the Reviewer for the comment. The manuscript was polished and checked by a third party.
+Reviewer #2:
+This manuscript explores the combination of a plasmonic particle with a molecular complex for hydrogen generation under plasmon excitation. The system is nicely design and the authors used a very elegant combination of conventional characterization techniques with ultrafast and NAP-XPS, which are less standard in this type of studies. As such they proposed that plasmonic hot-carriers are involved in the process and they discarded thermal effects due to the low thermal stability of the molecular complex which seems not to degrade. Overall this is an interesting study, that besides the results, presents interesting combination of techniques to further debate the role of plasmons in enhanced catalytic processes. I think this manuscript could be of potential interest after sorting some key points:
+REPLY: We thank the Reviewer for recognition of the scientific merit of this work and recommendation. We also thank the Reviewer for the comprehensive revision and detailed feedback, which helped us improve significantly the quality of the manuscript. We have taken all the suggestions on board, and modify the manuscript accordingly. A detailed point-by-point clarification is presented below. We hope the revised manuscript matches the Reviewer’s expectations and can be subsequently considered for publication.
+
+Why do the authors neglect the effect of enhanced near-fields that can promote electronic excitations in the molecular complex (i.e. resembling homogeneous photocatalysis processes)? From the abstract, introduction and discussion the dispute is heat or hot-carriers but the enhanced fields are never discussed nor considered.
+I am also wondering if (under a field-driven scenario), the e-ph coupling times could also be explained without needing the charge-transfer.
+REPLY: We thank the Reviewer for the comment. We decided to answer both comments together since they are related. The Reviewer is correct in stating that effect of enhanced near-fields can also contribute to enhancements in catalysis.
+We can consider two possibilities in the case of near-field enhancement:
+A. the plasmon enhances the optical absorption of the catalyst, similar to what Sheng et al. Nat. Commun. 14, 1528 (2023) published, or bahaves as a strongly correlated system as proposed by Rossi et al. Nat. Commun. 10, 3336 (2019). In both cases, the catalyst and plasmon resonance optical absorptions should be close in energy to detect significant enhancements. As we showed our catalyst has an absorption in the UV region (ca. 370 nm), while of Au LSPR is centred around 550 nm, which is considerable far apart in energy to consider this kind of enhancement. Moreover, we perfomed the catalysis under monochromatic illumination with a 532 nm CW laser to ensure that only the plasmon is excited.
+B. Plasmon near-field induces an energy transfer to catalyst leading to increase in activity. The Reviewer is correct in stating that an energy transfer from plasmon to the catalyst would result in a decrease of e-ph lifetime, because lower energy in the resonance would induce faster relaxation of the system, similar to what happens when we transfer hot electrons. However, energy transfer cannot justify the TIRAS signal. The broad and featureless TIRAS signal that we detected when we have NiO and the catalyst, is indicative of free carriers that must have come from the plasmon as hot carriers.
+While we cannot discard that near-field effects are present, the combined TAS and TIRAS measurements seems to be more supportive of hot carriers’ involvement than energy transfer. Other effects such as increased optical absorption or establishment of a strong correlated systems are unlikely due to large energy difference between the LSPR and catalysts absorption. Finally, catalysis promotion by local electric field enhancement is possible but one still needs the hot electrons to reduce the protons.
+We added to the manuscript text:
+"The free carrier signal detected by TIRAS confirms the transference of hot carriers to the acceptors, suggesting that the observed decrease in e-ph lifetime is less likely to be due to energy transfer. Since the Au plasmon resonance absorption is very far from the acceptors’ absorptions, one can also discard the hypothesis of photonic enhancement as the corporate for catalytic performance."
+Also, it has been shown that the highly concentrated electric fields can change the water molecules adsorption orientation and reactivity, modifying reaction energy barriers. As such, the hypothesis that thermal water-splitting needs more than 500 degrees, I’m not sure if still holds under high e-fields conditions.
+
+REPLY: We thank the Reviewer for the comment. Water molecules has strong bonds and consequently one needs high temperature to break them. In the case of water thermolysis in absence of a catalyst, this can be between 800-1000 C. Electric fields induce water molecules dipole alignment facilitating adsorption and increasing coverage, however the most significant effect seems to be on the electrical charges, in which electric fields can mitigate to some extent electron–hole recombination, and enhance the interaction of holes with water molecules at the interface, leading to more facile water intramolecular dissociation (Boyd et al Energies 15, 1553 (2022) and references within).
+The statement suggest once more a promoter role for the electric field but the reaction still needs the carriers. Additionally, the enhancements are small and thus it is not expected a significant decrease in water thermolysis temparature. The reported 500 C value is a conservative estimate, measured in a presence of a catalyst.
+
+We added to the text:
+“There is the possibility for near-field enhancements caused by the local electric fields formed upon Au LSPR excitation. While localized electric fields can promote the catalytic process by improving charge separation and molecule dipole alignment, they act on the hot carriers. Consequently, the local electric fields cannot catalyze the process, i.e., one still needs electrons to reduce the protons that are not generated from the localized electric fields.”
+
+And:
+The complete system’s susceptibility to off-resonant excitation was also evaluated. Excitation at 650 nm (off-resonant) did not yield significant differences in the CV compared with no excitation (Figure S17). Off-resonance excitation creates local effects like an increased temperature (hot spots) and a high electric field. However, when it comes to hot carriers, their energy is low and consequently not valuable to drive photocatalytic processes. The findings suggest that hot electrons are involved in the catalysis, and the increase in catalytic output under illumination is related to hot electrons, not thermal or electric fields. However, this does not mean that they are not present and might help the catalysis; instead, they cannot justify the catalytic performance on their own.
+
+Figure 2 should have the illumination conditions.
+
+REPLY: We thank the Reviewer for the comment. We updated the manuscript accordingly.
+
+The thermal argument is debatable. I agree with the authors, but it’s also true that they should show (by external heating) the degradation of the molecular complex (to support the idea that under high temperatures the catalyst breaks down).
+
+REPLY: We thank the Reviewer for the comment. The catalysts complex was found to thermally decompose at 265° C, well below water thermolysis.
+
+We added to the text:
+The molecular catalyst thermally decomposes at 265 °C.
+
+The NAP-XPS and TAS measurements are very nice and not usually found together in the same manuscript. So I thank the authors for trying to do a very deep mechanistic understanding with non-conventional (or not the most commonly used) techniques.
+
+REPLY: We thank the Reviewer for recognizing this. Combining NAP-XPS with ultrafast is indeed uncommon, and if we must say after doing it, we know why. The NAP-XPS measurements on a mesoporous electrode proved to be extremely challenging and according to the beam line responsible unreported till now. Mesoporous films capture too much electrolyte, requiring several hours removal until an unbroken thin film of electrolyte is established enabling XPS measurement under bias.
+I think the scheme in figure 5 can be highly enhance. For instance, the authors could put the times measured (instead of "very fast"). We all know these processes are very fast. Saying only very fast is not new/relevant. Also, where are the holes in that scheme? I think omitting the NiO doesn't help to understand the mechanism. The first arrow is the plasmon excitation and decay (I guess, because it says nothing).
+REPLY: We thank the Reviewer for the comment. We updated the figure to accommodate the Reviewer's suggestions.
+Reviewer #3:
+The authors report a plasmonic photocatalytic system for plasmon-assisted electrochemical hydrogen evolution and use photoelectrochemistry, transient spectroscopy, and XPS to provide evidence for the electron and hole transfer from plasmonic Au nanoparticles to NiO hole acceptor and Co-catalyst as an electron acceptor. Using transient spectroscopies and XPS, the authors provide evidence for the charge transfer from Au nanoparticles to holes and electron acceptors. However, the following concerns remain unaddressed and I recommend this study to be published after major revision.
+
+REPLY: We thank the Reviewer for recognition of the scientific merit of this work and recommendation. We also thank the Reviewer for the comprehensive revision and detailed feedback, which helped us improve significantly the quality of the manuscript. We have taken all the suggestions on board, and modify the manuscript accordingly. A detailed point-by-point clarification is presented below. We hope the revised manuscript matches the Reviewer’s expectations and can be subsequently considered for publication.
+
+1) In the introduction, the authors state that “it remains challenging to disentangle charge carrier catalysis from photothermal effects”. I agree with the authors here however, the same challenge still persists in the current study. Even though the authors have provided evidence for the charge transfer catalysis, authors have not provided any evidence to prove that photothermal effects are not playing a role in the photocatalytic system reported by the authors. The authors state that the photocatalytic system is not stable at which the water’s thermolysis takes place, which is why photothermal catalysis may not be occurring in their experiments. Even a small increase in the temperature due to photothermal heating can decrease the activation energy and can catalyze the electrochemical hydrogen evolution. The temperature does not have to reach 500 – 2000C (as mentioned by the authors in the manuscript) for photothermal heating to catalyze electrochemical hydrogen evolution. Hence, due to the lack of evidence to prove that photothermal heating is not participating in the electrochemical HER in the photocatalytic system reported in this study, both charge transfer and heating are likely to be playing the role in the photocatalytic HER reported by the authors unlike claimed otherwise by the authors. Authors are requested to include relevant discussion.
+
+REPLY: We thank the Reviewer for the comment. We want to start by emphasizing that we do not argue against some positive effects of the thermal process. We state that the photothermal process cannot alone justify the observed catalytic performance. This becomes even more obvious from the catalysis experiments performed under different light wavelengths, which the Reviewer suggested and will be discussed in point 4. We think the additional data, some of which was requested by the Reviewer, strengthens our conclusions and hopefully suppresses the Reviewer’s original concerns.
+
+2) To understand the relative contribution of the charge carriers and photothermal heating in catalyzing electrochemical HER, using a photocatalytic system involving non-plasmonic Au (smooth Au film) may help. NiO/smooth Au film/Co-Cat (non-plasmonic substrate) when used for the photocatalytic experiments, smooth Au being non-plasmonic, laser light excitation will mainly lead to photothermal heating/interband transition of Au and it would possible to only understand the contribution of only photothermal heating and this seems to be the primary motivation behind the study.
+
+REPLY: We thank the Reviewer for the comment. The Reviewer's suggestion is excellent and was something we have tried. However, attaching the catalyst to any gold film was impossible despite many attempts. Au films prepared by evaporation or electrodeposition showed no signature for catalyst attachment in the CVs, precluding us from making the desired experiment. We suspect the catalyst is attached to unsaturated Au atoms exposed by the annealing and consequent removal of the citrate capping agent. This is why the loading on Au NPs is very low, and the catalyst doesn’t attach to Au films. However, this requires further investigation that is beyond the work scope.
+3) Although the authors provide chronoamperometry data for the photoelectrocatalytic HER (Figure 2C) with light illumination on and off, CV data reporting electrocatalytic HER (with NiO/AuNPs/Co-Cat) with and without light illumination should be provided. Monitoring the onset potential for HER in conditions with lights on and off will provide further insights into the charge transfer process. Reporting the above-mentioned CV experiment with different light intensities will also provide additional evidence for the charge transfer.
+REPLY: We thank the Reviewer for the comment. The requested data was added to supporting information. We think the new figure shows the effect of light that the Reviewer requested. Since the data was collected with the same electrode, the measurements are comparable and quantitative.
+
+Actions taken:
+- We added Figure S17 to the supporting information.
+
+4) Hot electron generation and transfer are dependent on the wavelength of the light. The authors have performed experiments only at a single wavelength of light. Control experiments reporting photoelectrochemistry and transient spectroscopies using light of wavelength which off-resonance (for example 642 nm which does not excite the plasmon resonance) of the Au particles should be reported. Comparing results obtained at 550 nm excitation (already included in the manuscript) with at least one off-resonance wavelength excitation can provide additional insights.
+REPLY: We thank the Reviewer for the comment. The requested data was added to supporting information. The requested data was added to supporting information. We think the new figure shows the importance of exciting at the LSPR resonance maximum that the Reviewer requested. Since the data was collected with the same electrode, the measurements are comparable and quantitative. Moreover, this provides evidence that thermal cannot be the sole culprit for the catalysis since this would yield similar response independent of the excitation wavelength.
+
+Actions taken:
+- We added Figure S17 and S18 to the supporting information.
+
+5) Ex-situ transient spectroscopies make it clear that including NiO in a photocatalytic system alters e-e scattering, and e-ph scattering lifetimes, however, For Figures 2B and 2D, data for only FTO/Au/Co-Cat (no NiO) should also be provided to prove that NiO is playing during photoelectrochemical HER.
+REPLY: We thank the Reviewer for the comment. We updated the figures 2B and 2D to include with the requested data.
+
+6) Was the Argon atmosphere maintained during CV experiments reported in Figure 2B? If not, peak -0.5 V (present in both CVs NiO/Au/Co-Cat and NiO/Au), may also correspond to the oxygen reduction. Authors are requested to include the discussion regarding the same.
+REPLY: We thank the Reviewer for the comment. The peak at -0.5V is related to oxygen on Au reduction since it is always present when Au is in the system.
+
+7) In Figure 2D, NiO/Au also produced photoelectrochemical current, rather in the opposite way. The authors are requested to include a discussion regarding this.
+REPLY: We thank the Reviewer for the comment. We want to start by stating that photocurrents measured with NiO/Au are very small and, thus, challenging to conclude from shape analysis. However, since the QMS detected no H2 from this system and NiO is a known OER catalyst, the current may originate from OER reaction on NiO surface. The amount is meagre and thus not detectable by the QMS but would induce a change in photocurrent direction. We want to emphasize again that the photocurrent is very low, and no products were detected in the QMS, so shape analysis must be done with extreme care. If what is happening
+is what we stated, it further confirms the importance of plasmonic hot carriers to teh overall process.
+
+8) In the proposed mechanism section, authors state that the hot holes are transferred to the NiO and react at the counter electrode leading to O2 production. Authors are requested to provide data for this statement and include relevant discussion.
+REPLY: We thank the Reviewer for the comment. We overlapped to the SI to figure S4 the O2 trace confirming that O2 is produced at the counter electrode. The H2:O2 is roughly 2:1 as expected from the process stoichiometry.
+
+9) Regarding the characterization of the catalytic system:
+a) Only DLS measurements for the Au NPs are provided. DLS measurements are generally carried out in a colloidal state, however, in the photocatalytic system reported in this study, Au NPs are drop cast on NiO/FTO and then annealed. Drop casting and annealing of the drop casted Au NPs at 500C will result in Au NPs aggregation which may shift the LSPR of the Au NPs. SEM images of Au NPs on FTO before and after annealing should be provided for thoroughness.
+REPLY: We thank the Reviewer for the comment. We updated the SI with AFM images of Au nanoparticles after annealing at 500C showing little to no aggregation even when measurements are carried out in a flat Si wafer. We also added AFM of the sample after second annealing showing that NiO morphology including porous structure is preserved. Finally, we did additional XPS analysis confirming that NiO surface oxidation is preserved throughout the preparation of the film.
+
+b) In Figure 2A, the authors state that no distinguishable peaks were reported before the catalytic wave which relates to Co-complex redox behavior. However, there is a clear reduction (-1.2 V in acid, -1.4 V in water) and oxidation peak in both CVs (water and acid) which are uncharacteristic of HER because in HER, oxidation peak on the reverse sweep is not usually observed (Figure 2C inset). Hence, out of two catalytic waves, the first catalytic wave is unlikely from HER and may be related to the Co-Cat oxidation-reduction behavior. Authors are requested to provide further discussion regarding this. Further, was the Co-cat free in the solution or deposited on the glassy carbon? Authors are requested to include this information in the SI.
+REPLY: We thank the Reviewer for the comment. We remeasured several times figure 2a, ensuring that no oxygen was present in the solution. It became clear that peak was related to oxygen reduction and not catalyst reduction, which is consistent with NAP-XPS. The figure was updated accordingly.
+
+Actions taken:
+- We added Figure S6, S7 and S8 to the supporting information.
+
+c) Figure 3D reports the use of Au/ligand. No procedure for preparing Au/ligand (no Co) is reported in the paper. Authors are requested to provide the method of preparation in the SI.
+REPLY: We thank the Reviewer for the comment. We added the description to the supporting information. This was done in the same manner as adding the catalyst by simple immersion in a solution containing the linker.
+
+d) Au nanoparticles are prepared using tannic acid. What is the purpose of using tannic acid as Au NPs of the same size can be prepared using citrate? Please include the relevant discussion for better clarity.
+REPLY: We thank the Reviewer for the comment. We followed Piella et al. Chem. Mater. 28, 1066 (2016) synthesis protocol. The protocol is very reproducible and reliable, and became the cornerstone of our group plasmonic synthesis. According to the authors and some studies perfomed in our group, tannic acid decreases the synthesis pH from 7.7 to 6.4. This promotes formation of smaller particles and a higher uniformity in size and shape. There are ample
+literature that advocates pH role in controlling size and uniformity of particles produce with the Turkevich method.
+
+10) Regarding transient spectroscopy:
+a) The lifetime of all the scattering processes changed after including Co-Cat in the photocatalytic system. Short pulses used for the spectroscopy may generate a lot of local heat which may damage the photocatalytic system especially the organic compound Co-complex. Can this decomposition or damage to the catalyst have an impact on the scattering process lifetime? Please include relevant discussion. FTIR, SEM, and UV Vis of the catalytic system after irradiating short laser pulses should be included to ensure the integrity of the catalyst and make sure that the decomposition of the catalyst (if at all) is not the reason behind the scattering process’s lifetime change.
+REPLY: We thank the Reviewer for the comment. Beam damage is an issue that, unfortunately, has pledged several published data sets. Experient users, which we considered to be, take the beam damage problem very seriously. We adopt several procedures that enable us to ensure that the data is collected unaffected by beam damage, such as:
+- Sample circulation so the same spot is not measured more than 2 or 3 times
+- Comparison of between spectra measured on the same spot
+- Power dependence measurements to determine damage power threshold
+By adopting such practices, one can ensure data without beam damage effects. Unfortunately, UV-Vis and FTIR analysis offer low guarantees because the sport size of the laser system is significantly smaller than the probe area of such analytics, meaning damage is only observed in extreme cases. SEM could detect pinhole formation due to pump laser damage; however, we are using fluencies where this is impossible, at least, from our experience
+
+b) Why was the 490 nm winglet considered for the analysis and not the other winglet? Please provide the discussion.
+REPLY: We thank the Reviewer for the comment. Similar analysis was performed with the other winglet, which yielded the same result.
+We added to the manuscript text:
+"Note that similar findings were obtained when performing the analysis on the winglet to the red of LSPR maximum"
+
+11) It would be helpful for the readers if you can please provide pictures of the electrodes, electrochemical cell and experimental set up.
+REPLY: We thank the Reviewer for the comment. We added some pictures of electrodes, electrochemical cell and experimental setup to the revised SI.
+
+Actions taken:
+- We added Figures S1 and S2 to the supporting information.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+Although readers made some revision to improve the quality of the manuscript, there are still some issue to be resolved.
+1. In response to Q1 from Reviewer 1, author tried to convince the successful binding of Coll(phen-NH2)2(H2O)2 via Au-N bonds. However, I am still not convinced. The IR spectra focused only the wavenumber of 3200-3500 cm-1. It is strongly suggested to show the spectra of other wavenumber region for cross checking. Authors claimed that the binding of molecules was confirmed by the shift in UV-Vis spectra. However, the shift of UV-Vis peak is attributed to the change in refractive index of surrounding dielectric medium. This can also happen in the case of physical adsorption. Therefore, the chemical bonding of Au-N could not be confirmed by UV-Vis spectra. In term of XPS spectra, a shift of Au 4f and N 1s should be presence because of the Au-N bonding. However, no shift was observed. Therefore, the Au-N bonding could not be confirmed.
+2. In response to Q2 from Reviewer 1, authors refused to compare their catalyst with similar one in the literatures by claiming their catalyst is new. It will be great to compare new catalyst with existing ones, which will make readers to understand the efficiency of the newly developed catalyst.
+3. In response to Q6 from Review 1, after deconvolution, the shift of O 1s peak was only 0.15/0.1 eV, which is usually too small for a reliable shift. Usually, the shift smaller than 0.2 might be attributed to measurement error.
+4. In response to Q15 from Review 1, it is still not logical to abbreviate “concerted proton-coupled electron transfers” as CEPT.
+5. Since the claimed Au-N bonding was not convinced and the mechanism was not fully confirmed by results, I think the quality of this work is not high enough for being published in Nature Communications.
+
+Reviewer #2 (Remarks to the Author):
+
+The revised version of the manuscript is highly improved but some answers are still weak, in my opinion. There are some points that I still disagree; I mention them below. Overall, this is a very nice work and as I mentioned earlier in the first round, I thank the authors for putting together a very nice set of techniques that have rarely been used in plasmonic catalysis until now.
+
+1) The explanation that the catalyst absorbs in the UV and that the plasmon resonance is in the visible is not enough - in my opinion - to discard the role of the electric field. There are many examples in the literature using catalytic metals (absorbing in the UV) but operating nicely in the visible when coupled to the plasmonic near-field of a secondary metal. Very similar to this case.
+
+2) The thermolysis of the bounded catalyst should be shown by external heating. What do you detect
+when the catalyst breaks down? This is the main support for discarding thermal effects and it should be better demonstrated.
+
+3) I disagree with the discussion regarding the wavelength-dependent experiments (S17 and S18) and the fact that they further support a hot-carriers pathway. Exciting the plasmon resonance is also the most efficient way to heat the system, so both processes, photothermal heating (that depends on the absorption cross-section) and hot-carriers generation are maximized at the same wavelength. For that reason, I think that point 2 is relevant (to show the thermolysis by external heating more than wavelength-dependence).
+
+As one of the key motivations of the paper is to disentangle the mechanisms behind the reported activity under plasmon excitation, I think it would be necessary to show in a more comprehensive way the role of field enhancement and heat.
+Point-by-point answer to Reviewers comments:
+
+Reviewer #1:
+
+Although readers made some revision to improve the quality of the manuscript, there are still some issue to be resolved.
+We thank the Reviewer for acknowledging the enhancements made to the manuscript and providing valuable feedback to strengthen its quality. Additional experiments were conducted to substantiate our initial claims about catalyst anchoring. Plus, we added some comparisons with catalysts that share similarities to ours regarding their bulk electrolysis. In our view, these were the main aspects highlighted. We appreciate the Reviewer’s constructive feedback, suggestions, and the time invested in revising the manuscript. We trust that the revised version addresses any lingering concerns, paving the way for the acceptance of the manuscript.
+
+1. In response to Q1 from Reviewer 1, author tried to convince the successful binding of CoII(phen-NH2)2(H2O)2 via Au-N bonds. However, I am still not convinced. The IR spectra focused only the wavenumber of 3200-3500 cm-1. It is strongly suggested to show the spectra of other wavenumber region for cross checking. Authors claimed that the binding of molecules was confirmed by the shift in UV-Vis spectra. However, the shift of UV-Vis peak is attributed to the change in refractive index of surrounding dielectric medium. This can also happen in the case of physical adsorption. Therefore, the chemical bonding of Au-N could not be confirmed by UV-Vis spectra. In term of XPS spectra, a shift of Au 4f and N 1s should be presence because of the Au-N bonding. However, no shift was observed. Therefore, the Au-N bonding could not be confirmed.
+We thank the Reviewer for the suggestion. We performed the suggested XPS experiments to substantiate the catalyst anchoring to the Au surface. The XPS comparing the N 1s and Au 4f signals before and after binding to the Au surface are presented in Figures 2E and 2F, respectively. Before attaching, the catalyst has two N1s peaks: the N from the phenanthroline bonded to the Co at 398.8 eV and N at 401.4 eV ascribed to NH2 groups. Upon attaching, the signal related to N of the NH2 group disappeared, and the N from phenanthroline shifted to 399.1 eV and got broader (FWHM before 1.641 and after attaching 1.812). These observations are consistent with attachment via the NH2 groups and formation of Au-N species, which cannot be distinguished from the N of the phenanthroline ligand because of signal-to-noise. Additionally, there were no changes in the Au 4f, which is understandable since the catalyst loading was very low.
+
+Action taken: we added Figure 2A and 2B to the manuscript and added to the manuscript:
+The 10-Phenanthrolin-5-amine ligand was purposely chosen to ensure selective coordination to the gold surface via the amino groups. This first supporting evidence came from XPS measured at low vacuum conditions, which had a Co 2p signal related to the catalyst only when Au NPs were present. The Co 2p½ on Au/Co-cat and NiO/Au/Co-cat measured had single contribution centred at around 780.5 eV, consistent with Co is oxidation +2. The observation that Co signal was only present when Au is present is a strong endorsement to the selective anchoring of the catalyst to the gold surface. The anchoring is believed to occur via the -NH2 groups. This was corroborated by the disappearance of the amino bands in the infrared…
+The UHV-XPS comparing the N 1s and Au 4f signals before and after anchoring the catalyst to the Au surface are presented in Figures 2A and 2B, respectively. Before attaching, the catalyst has two N 1s peaks: the N from the phenanthroline bonded to the cobalt center at 398.8 eV and N at 401.4 eV ascribed to the -NH2 groups. Note that the UHV-XPS also did not show a peak ascribed to the nitrate
+groups, corroborating the its exchange by water molecules. Upon attaching, the N 1s signal related to the -NH2 group disappeared, and the N from phenanthroline shifted to 399.1 eV and got broader (FWHM before 1.641 and after attaching 1.812). Note that one expects the N 1s from the amino group to shift to lower binding energies as it loses the protons at a rate of about 1eV per each hydrogen atom lost. These observations are consistent with attachment via the NH2 groups and formation of Au-N species, which cannot be distinguished from the N of the phenanthroline ligand because of signal-to-noise. The Au 4f7/2 had a binding energy at 83.6 eV for the sample with and without catalyst, consistent with metallic gold. There was only one species in all the samples, coherent with the idea that any electronic change due to NiO and Co-catalyst is delocalized over all the gold atoms, suggest good electronic coupling. Figure 1D shows the a schematic representation of the complete photosystem.
+
+2. In response to Q2 from Reviewer 1, authors refused to compare their catalyst with similar one in the literatures by claiming their catalyst is new. It will be great to compare new catalyst with existing ones, which will make readers to understand the efficiency of the newly developed catalyst.
+We thank the Reviewer for the comment. We did not refuse to compare the catalyst performance with other systems because we wanted to ensure that the comparison is fair and valid. In the revised version of the manuscript we compared our catalyst with Luo’s et al system that served as inspiration for the reported system. In the new version, we compared our system further with Wang’s et al. system, namely [Co(Py3Me-Bpy)(OH2)2]-(PF6)2. Their proposed reaction mechanism is very different from ours (as with the case of Luo’s bipyridine system) but the system has similar structural rigidity and coordination making it possible to compare.
+
+Action taken: we added to the revised text:
+Bulk electrolysis of the cobalt catalyst in water using 0.1 M LiCl as a supporting electrolyte without acid (Figure 3A) shows no unique reduction peaks before the onset of the catalytic wave at -1.18 vs Ag/AgCl. Initiating with the absence of reduction peaks, this discovery diverges from the observations made by Luo et al. (whose system served as inspiration for this study) and the system studied by Wang et al., exhibiting comparable rigidity and coordination. The onset potential for electrolytic hydrogen production without using acid in our catalyst was lower, differing by only ca. 15 mV from Wang et al. system but by a noteworthy 190 mV from the Luo et al. system.
+Predictably, the catalytic wave’s amplitude is heightened with the presence of protons, underscoring proton availability as a pivotal factor in accessing catalytic performance within this system. Consequently, all catalytic data was obtained in a 3 mM acetic acid environment. Although the introduction of acid did not result in new reduction peaks, it did cause a notable shift in the onset potential, reducing it by as much as 120 mV—a deviation from the characteristics of the previously documented system. These observations imply the engagement of two concerted proton-electron transfer mechanisms as opposed to the conventional sequential reduction followed by protonation. Subsequent sections will delve into additional evidence supporting this proposed mechanism.
+
+3. In response to Q6 from Reviewer 1, after deconvolution, the shift of O 1s peak was only 0.15/0.1 eV, which is usually too small for a reliable shift. Usually, the shift smaller than 0.2 might be attributed to measurement error.
+We thank the Reviewer for the comment. We understand the Reviewer’s uncertainty about our in situ XPS results, but we respectfully disagree with the comment. The experiment is complex and final shifts detected are small. However, we capitalized on several years of experience with such measurements and we believe that the shifts shown are real for two reasons:
+    1. Statistics of the measurement. Spectra shown in Figures 4 (main text), S14 and S15 do not display
+single scans of the O 1s. They are the sum of 30 to 50 iterations. The signal-to-noise ratio is high and deconvolutions give good correlation with experimental data.
+2. Deconvolutions of O 1s (Figure S15) show that peaks at 529.5 eV and 531.0 eV, corresponding to lattice oxygen and adsorbed hydroxyls, respectively, do not shift upon voltage application. In contrast, the component centered at approximately 532.7 eV, assigned to the thin electrolyte film on top of the electrode, displays a clear shift (spectra never superimpose in that region). If we consider lattice oxygen and hydroxyls as “internal references”, this is a clear evidence of voltage-induced shift of the 532.7 eV peak.
+
+Actions taken: we modified the text as follows.
+A difference of about 0.1–0.15 V was detected between the applied potential chosen based on the catalysis and seen in the O1s NAP-XPS, which is assigned to calibration shifts of the reference electrode during long experimental times. Therefore, the values specified in the plots are applied for consistency, not measured voltages. It is important to highlight that while the 532.7 eV peak component clearly displays a shift proportional to the applied voltage, components at 529.5 eV and 531.0 eV do not.
+
+4. In response to Q15 from Review 1, it is still not logical to abbreviate “concerted proton-coupled electron transfers” as CEPT.
+We thank the Reviewer for the comment. The Reviewer is correct in stating that it is not concerted proton-coupled electron transfer. Instead, it should be concerted proton-electron transfers (CPET) according to for example Tyburski et al. J. Am. Chem. Soc. 143 (2021) 560, which we corrected. This is indeed the terminology used by the leading groups
+
+Action taken: we fixed the issue in the text and reaction mechanism figure (figure 7) by making sure that is concerted proton-electron transfers (CPET).
+
+5. Since the claimed Au-N bonding was not convinced and the mechanism was not fully confirmed by results, I think the quality of this work is not high enough for being published in Nature Communications.
+We thank the Reviewer for the comment. We are convinced that the newly added XPS and anchoring followed by in situ FTIR substantiates our claims and suppresses the original Reviewer’s reservations.
+Reviewer #2:
+
+The revised version of the manuscript is highly improved but some answers are still weak, in my opinion. There are some points that I still disagree; I mention them below. Overall, this is a very nice work and as I mentioned earlier in the first round, I thank the authors for putting together a very nice set of techniques that have rarely been used in plasmonic catalysis until now.
+
+We thank the Reviewer for acknowledging the enhancements made to the manuscript and providing valuable feedback to strengthen its quality. Additional experiments were conducted to substantiate the direct involvement of hot carriers in the process. More specifically, we performed other chronoamperometry data under light modulation and analyzed the response shape, which provides additional evidence for the involvement of hot electrons in the catalytic process. We appreciate the Reviewer’s constructive feedback, suggestions, and the time invested in revising the manuscript. We trust that the revised version addresses any lingering concerns, paving the way for the acceptance of the manuscript.
+
+1) The explanation that the catalyst absorbs in the UV and that the plasmon resonance is in the visible is not enough - in my opinion - to discard the role of the electric field. There are many examples in the literature using catalytic metals (absorbing in the UV) but operating nicely in the visible when coupled to the plasmonic near-field of a secondary metal. Very similar to this case.
+
+We thank the Reviewer for the comment. Significant enhancements due to near-filed in plasmon-molecule hybrid systems were, to our knowledge, reported only when you have some optical overlap. For systems without optical overlap, those effects were only observed when you have the plasmonic materials in a core-shell architecture (Kholmicheva et al. ACS Nano 2014; 8, 12549 & Nanophotonics 2019; 8, 613), which is not what we have. In such systems, significant near-field effects were observed for Ag, while in the case of Au the effect is negligible (Cushing et al. J Phys Chem C 2015, 119, 16239). According to literature off-resonance excitation is able to generate significant local fields (Sun, et al. Light 2, e118 (2013); He et al. Scie. Rep. 6, 20777 (2016) & Hermann et al. Opt. Exp. 26, 27668 (2018)) but produce very low amount of useful hot carriers (Tagliabue, et al. Nat. Commun. 9, 3394 (2018)). This offered us the possibility to evaluate of the near-fields can affect directly the catalysis. Since this was found to not be the case, and the fact that one established that the ligands were reduced via unbiased ultrafast spectroscopies, it is possible to role near-fields as the corporate for the light-induced enhancement.
+
+Action taken: we added to the revised text (including references):
+
+It is clear from the data that the systems with plasmonic materials are responsive to the 532 nm illumination. Light-mediated plasmon-catalysis is a very complex process due to many potential reaction enhancers. One possibility is the near-field enhancements caused by the local electric fields formed upon Au LSPR excitation. At the most basic level near-fields can enhance charge separation and alignment of molecular dipoles. However, such localized electric fields impact the hot carriers but cannot catalyze the reaction autonomously. The second option employs the plasmon-induced resonance energy transfer (PIRET) process, connecting the plasmon evanescent field to a semiconductor absorber through dipole–dipole interaction. However, these systems necessitate core-
+shell architectures (which are not applicable in this context), and the most substantial enhancements were observed with silver as the plasmonic material rather than gold. The final option explores strong-correlated plasmon-molecule systems, but in this scenario, there must be an optical overlap between the plasmon and molecule, which is once again not present in this case.
+
+To assess local field enhancement contribution, catalytic performance measurements were conducted using off-resonant excitation. Off-resonance excitation induces local effects such as elevated local near fields. However, in the context of hot carriers, off-resonance excitation generates low-energy carriers that are not conducive to driving photocatalytic processes. Excitation at 650 nm (off-resonant) caused no significant differences in the CV compared to experiments performed in the dark (Figure S17). This result was further supported by light switch chronoamperometry (Figure S18). The findings imply that local near-fields do not contribute significantly to enhancing catalysis. Consequently, the observed increase in catalytic output under resonant illumination is likely associated with hot electrons and heat rather than near fields. This, however, does not rule out the potential for near-fields to assist catalysis by enhancing charge separation; they would likely influence hot carriers indirectly engaged in the catalytic process.
+
+2) The thermolysis of the bounded catalyst should be shown by external heating. What do you detect when the catalyst breaks down? This is the main support for discarding thermal effects and it should be better demonstrated.
+
+We thank the Reviewer for the comment. Since the catalysis experiments were performed in an aqueous medium, it is impossible to perform thermolysis experiments at relevant temperatures (at least 500C). However, we can state that experiments performed at about 60C did not change significantly the catalytic performance.
+
+The issue with plasmonic catalysis is that heat will always be present since we don’t use all the charges. Additionally, surface temperature (much higher than the solution) is challenging since it also has ultrafast dynamics. Therefore, we performed light modulation experiments and analysed the shape catalytic response. This is presented in the new figure 3, and discussed in detail in the following comment. We think these additional experiments confirm the direct involvement of hot carriers as the prime culprit for the process while not discarding some positive contribution of heat.
+
+Action taken: we added Figure 4 to the revised version of the manuscript and the text:
+
+Heat is an inherent factor in plasmonic catalysis due to the occasional underutilization of charges, leading to their recombination and the generation of local heat. Although it is acknowledged that the surface temperature of excited plasmonic materials exceeds that of the solution, determining the precise value poses a challenge due to the ultrafast dynamics of thermalization. The molecular catalyst remains stable only up to 265 °C, a temperature considerably lower than what is required for uncatalyzed water thermolysis. Nevertheless, there exists a noteworthy temperature range that remains unexplored, primarily because experiments are conducted in an aqueous medium.
+
+3) I disagree with the discussion regarding the wavelength-dependent experiments (S17 and S18) and the fact that they further support a hot-carriers pathway. Exciting the plasmon resonance is also the most efficient way to heat the system, so both processes, photothermal heating (that depends on the absorption cross-section) and hot-carriers generation are maximized at the same wavelength. For that reason, I think that point 2 is relevant (to show the thermolysis by external heating more than wavelength-dependence).
+
+We thank the Reviewer for the comment. We agree that the off-resonant experiments cannot rule out the heat contribution since one expects low heat generation since we formed low energy carriers. However, the experiment provided evidence against the near-field enhancement.
+To disentangle the heat contribution, we performed light modulation chronoamperometry. Maley et al. (J. Phys. Chem. C 2019, 123, 12390) showed that light absorption at electrode surfaces in nanoparticle arrays created significant local temperature increases and solution flows. These thermal effects were predicted to alter observed electrochemical currents through various mechanisms, including mass transfer enhancements, shifts in equilibrium redox potentials, or conventional temperature-dependent increases in kinetic rates for electrode processes. In particular, the presented analysis predicts that mass transfer enhancements alone would result in sizable current increases, and these enhancements would apply to any electrochemical reaction involving dissolved reactants and/or products. This was found valid for both outer-sphere and inner-sphere reactants. Consequently, heat-induced effects have a characteristic slow rise and decay of the current under light modulation since they operate on processes with time constants in the nanosecond time scale.
+
+The light response of the complete system (NiO/Au-Co catalyst) versus the system without NiO (i.e. Au/Co-catalyst) provides clues for the contribution of heat to the process (figure 2C and D, compounded figure below)). By removing the NiO, the lifetime of the charge-separated state is expected to decrease and thus generate more heat. Consequently, if heat is the main contributor to the reactivity, one should expect a higher current when NiO is not present, which is not the case because we measured four times higher current induced by light when NiO was present. Furthermore, analysis of the response of the Au/Co-catalyst to light mocluation shows a classic heat-mediated process with a relatively slow rise (see figure below) and decay, contrasting with the complete system which shows a faster rise and decay to the light modulation, indicating hot carrier involvement.
+
+![Normalized current vs. Time graph showing two curves: Au/Co-Cat and NiO/Au/Co-Cat](page_420_670_600_350.png)
+
+To decouple heat from hot carriers and further substantiate our mechanistic claim, we perfomed light modulation experiments with a higher repetition rate (new Figure 3). By increasing the repetition rate one expects lower heat accumulation at the electrode and thus clearer signal for hot carriers.
+
+Action taken: we added Figure 4 to the revised version of the manuscript and the text:
+Heat is an inherent factor in plasmonic catalysis due to the underutilization of hot charges, leading to their recombination and the generation of local heat. Although it is acknowledged that the surface temperature of excited plasmonic materials exceeds that of the solution, determining the precise value poses a challenge due to the ultrafast dynamics of thermalization. The molecular catalyst remains stable only up to 265 °C, a temperature considerably lower than what is required for uncatalyzed water thermolysis. Nevertheless, there exists a noteworthy temperature range that remains unexplored, primarily because experiments are conducted in an aqueous medium.
+To disentangle the heat contribution, we performed light modulation chronoamperometry. In a study by Maley et al., it was demonstrated that light absorption at the electrode surfaces within nanoparticle arrays led to significant localized temperature increases and altered solution flows. These thermal effects were anticipated to influence electrochemical currents through diverse mechanisms, encompassing enhancements in mass transfer, shifts in equilibrium redox potentials, and conventional temperature-dependent accelerations in kinetic rates for electrode processes. Notably, their analysis suggests that mass transfer enhancements alone would result in substantial current increases applicable to electrochemical reactions involving dissolved reactants and products, both outer-sphere and inner-sphere reactants alike. Consequently, heat-induced effects exhibit a distinctive gradual rise and decay of the current during light modulation, as they operate on processes with time constants in the nanosecond range.
+
+The light response of the entire system (NiO/Au-Co catalyst) in comparison to the system without NiO (i.e., Au/Co-catalyst) offers insights into the role of heat in the process (see figure 3C and D, compounded figure below). Removing NiO is expected to decrease the lifetime of the charge-separated state, generating more heat. However, contrary to the expectation that heat is the primary contributor to reactivity, we observed a fourfold increase in current induced by light when NiO was present.
+
+Additionally, examining the response of the Au/Co-catalyst to light modulation reveals a classic heat-mediated process with a relatively slow rise and decay, in contrast to the complete system. The complete system demonstrates a faster rise and decay to light modulation, indicating the involvement of hot carriers.
+
+Light absorption by planar electrodes randomly decorated with plasmonic structures acts as a uniform heat source delocalized across the electrode–solution interface, resulting in heat dissipation in a linear geometry with significant temperature changes as a function of time. Consequently, the electrochemical response to light modulation provides a strategy to decouple heat from hot carriers’ contributions to substantiate our mechanistic claim. Commonly, the experiments are performed by modulating the light intensity. However, the tested electrodes are quasi-transparent, making light-intensity modulation studies challenging. Thus, we opted to change the light ON/OFF cycle repetition rate to modulate electrode exposure to light.
+
+Figure 4A shows the changes in measured photocurrent (\( \Delta i \)) as a function of light modulation repetition rate. Unsurprisingly, lower repetition rates (i.e. higher light exposure) resulted in larger \( \Delta i \). In a heat-mediated process, the \( \Delta i \) is expected to scale with \( \sim t^{1/2} \) (t = time), inconsistent with the observed current transients, providing the first substantiation for a hot carrier-mediated process. Additionally, \( \Delta i \) in a heat-mediated electrochemical process follows a linear dependence with increased light exposure, independent of the process occurring via inner or outer sphere reaction. Figure 4B shows that \( \Delta i \) does not show a linear behavior regarding light exposure, thus providing clearer evidence for hot carriers’ involvement.
+
+![Two plots: (A) Photocurrent vs Cycles for different repetition rates; (B) Photocurrent vs Frequency for different repetition rates](page_186_1347_1077_393.png)
+Figure 4. Light-modulated photo-electrocatalytic studies. The effect of light-modulation in the chronoamperometry was perfomed at -0.65 V vs Ag/AgCl with 3 mM acetic acid (pH = 3.5) and 532 nm CW laser with 43.8 mW/cm². The experiments were performed using squared function at different repetition rates. A) Changes in measured photocurrent (Ai) at different light modulation frequencies (8, 11, 16 and 33 mHz) over different light ON/OFF cycles; and B) Ai versus light modulation frequency.
+
+As one of the key motivations of the paper is to disentangle the mechanisms behind the reported activity under plasmon excitation, I think it would be necessary to show in a more comprehensive way the role of field enhancement and heat.
+We think the additional data and analysis substantiate our claim that hot electrons are involved in the process and are the main culprit for the light-induced response. The off-resonance measurements ruled out the near-field contribution as the direct cause of catalysis. However, they might be implicated in enhancing hot electrons' lifetime.
+The new analysis of the response of the systems to light and the new light modulation experiments further substantiate the hot electron mechanism to the detriment of heat. Still, the heat was found to have some positive effect on the catalysis.
+
+Action taken: we added Figure 4 to the revised version of the manuscript and the text:
+In summary, a photosystem was proposed to confirm the direct involvement of hot electrons in a photocatalytic process, in this case H₂ evolution process. The photosystem effectively mitigates the heat contribution by designing a catalyst that decomposes well below water thermolysis conditions, positioning heat as a mere enhancement rather than the primary cause for the observed H₂ evolution. Off-resonance measurements conclusively eliminate near-field contributions as the direct catalyst of the reaction, although they may still play a role in extending the lifetime of hot electrons. The catalytic response to light modulation exhibits a shape consistent with the desired electron mechanism, contrasting with the detrimental impact of heat. Nevertheless, it was discovered that heat does have some positive influence on catalysis. Unbiased ultrafast spectroscopic measurements confirm charge transfer to respective acceptors. Additionally, in conjunction with NAP-XPS under variable potential, a postulated reaction mechanism highlights the crucial role of cobalt catalyst ligands. These ligands accept plasmon hot electrons and, through CPET steps, reduce and protonate the metal centre, ultimately leading to hydrogen evolution. This study conclusively resolves the longstanding debate within the research community regarding the direct involvement of hot carriers in the photocatalytic process.
+REVIEWERS’ COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+I am now satisfied with the revision.
+
+Reviewer #2 (Remarks to the Author):
+
+I think the authors did a very good job in addressing the second round of questions. The new chronoamperometric results are highly appreciated to clarify the carriers versus heat issue. I think the manuscript is ready for acceptance.

04b173197c545f10945642520e847eff78301d76d085be0937577c7b3426ca9d/preprint/preprint.md

@@ -0,0 +1,219 @@
+Plasmon-ligand-mediated hydrogen evolution with visible light
+
+Jacinto Sa (jacinto.sa@kemi.uu.se)
+Uppsala University   https://orcid.org/0000-0003-2124-9510
+Ananta Dey
+Uppsala University
+Amal Mendalz
+Uppsala University
+Anna Wach
+Paul Scherrer Institut   https://orcid.org/0000-0003-3112-2759
+Robert Vadell
+Uppsala University
+Vitor R. Silveira
+Uppsala University
+Paul Maurice Leidinger
+Paul Scherrer Institut
+Thomas Huthwelker
+Paul Scherrer Institute
+Vitalii Shtender
+Uppsala University
+Zbynek Novotny
+Paul Scherrer Institut
+Luca Artiglia
+Paul Scherrer Institute   https://orcid.org/0000-0003-4683-6447
+
+Article
+
+Keywords:
+
+Posted Date: May 2nd, 2023
+
+DOI: https://doi.org/10.21203/rs.3.rs-2751820/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on January 10th, 2024. See the published version at https://doi.org/10.1038/s41467-024-44752-y.
+Plasmon-ligand-mediated hydrogen evolution with visible light
+
+Ananta Dey1, Amal Mendalz1, Anna Wach2, Robert Bericat Vadell1, Vitor R. Silveira1, Paul Maurice Leidinger2, Thomas Huthwelker2, Vitalii Shtender3, Zbynek Novotny2, Luca Artiglia2, Jacinto Sá1,4*
+
+1 Department of Chemistry-Ångström, Physical Chemistry division, Uppsala University, Box 532, 751 20 Uppsala, Sweden.
+2 Paul Scherrer Institut, CH-5232 Villigen PSI, Switzerland.
+3 Department of Materials Science and Engineering, division of Applied Materials Science, Uppsala University, 75103 Uppsala, Sweden.
+4 Institute of Physical Chemistry, Polish Academy of Sciences, Marcina Kasprzaka 44/52, 01-224 Warsaw, Poland.
+
+*Corresponding author. Email: jacinto.sa@kemi.uu.se
+
+Abstract: Plasmonic systems convert light into electrical charges and heat that mediate catalytic transformations. However, the debate about the involvement of hot carriers in the catalytic process remains shredded in controversy. Here, we demonstrate the direct use of plasmon hot electrons in the hydrogen evolution with visible light. A plasmonic nanohybrid system consisting of NiO/Au/[CoII(phen-NH2)2(H2O)2] (phen-NH2 = 1,10-Phenanthrolin-5-amine) that is unstable at water thermolysis temperatures was consciously assembled, ensuring that the plasmon contribution to the catalytic process is solely from hot carriers. With the combination of photoelectrocatalysis and advanced in situ spectroscopies, one could establish the reaction mechanism, which consisted of electron injection into the phenanthroline-ligands followed by two quick, concerted proton-coupled electron transfer steps resulting in the evolution of hydrogen. Light-driven hydrogen evolution with plasmons provides a sustainable route for producing green hydrogen, which modern society strives to achieve.
+Plasmonic photocatalysis uses electrical charges formed during the plasmon resonance decay triggered by light absorption. A plasmon is a quantized oscillation of the electron density, and its decay can generate hot carriers. Hot carriers in plasmonics refer to the generation of high-energy electrons (and holes) in metal due to the interaction between plasmons and incident light, causing them to become ‘hot’ or have high kinetic energy.\(^1\) These hot carriers can then be used to generate electrical current or to drive chemical reactions. However, hot electrons involvement in catalysis remains disputed, despite reports of their participation in processes such as such as solar to chemical energy reactions,\(^{2-5}\) epoxidations,\(^{6,7}\) dehydrogenations,\(^8\) ammonia electrosynthesis,\(^9\) etc.\(^{10-13}\) The skepticism surrounding their involvement relates to the hot carriers’ ultrafast relaxation (ca. 100 fs),\(^{14}\) and that several examples rationalize their participation as enhancer of the photothermal process. Therefore, the catalytic output is prone to errors since the surface temperature of plasmonic materials is notoriously tricky to measure accurately, thus underestimating the thermal contribution to the catalysis.\(^{15,16}\) Despite the significant progress, it remains challenging to disentangle charge carrier catalysis from photothermal effects.\(^{17-20}\)
+
+The hot electron energy distribution is broad, with a significant fraction of the carriers having energies above the Fermi level of the metal caused by the non-Fermi-Dirac distribution.\(^{21}\) More research is needed to fully understand plasmonics’ hot carrier energy distribution dynamic behaviour,\(^{22,23}\) but the ultrafast relaxation can be partially mitigated via ultrafast charge transfer to suitable acceptors consecutively,\(^{24,25}\) or simultaneously,\(^{26}\) forming this contribution scientific basis to demonstrate the direct involvement of hot carriers in the catalytic process. Moreover, the hot electrons were used to reduce protons to hydrogen. Water can be converted into hydrogen through thermolysis. The exact temperature required for thermal water splitting depends on the specific conditions. Still, typically temperatures in the range of 500-2000°C are required for efficient thermal water splitting,\(^{27}\) a temperature at which the catalytic system presented herein is unstable.
+
+Herein, a plasmonic nanohybrid system consisting of NiO/Au/[Co^{II}(phen-NH_2)_2(H_2O)_2] (phen-NH_2 = 1,10-Phenanthrolin-5-amine) was assembled and tested in hydrogen evolution reaction (HER). NiO acted as a hole acceptor,\(^{28-31}\) and the cobalt complex, a mimic of the HER catalyst reported by Luo et al.\(^{32}\) and the hydrolytic DNA cleavage agent by Sharma et al.,\(^{33}\) as an electron acceptor. The reaction mechanism was monitored by a combination of photoelectrocatalysis, ultrafast spectroscopies, and in situ electrochemistry, followed by near-ambient pressure X-ray photoelectron spectroscopy (NAP-XPS) studies (Figure S1). The results suggest a reaction mediated by the phenanthroline-ligands that accept the electrons from the plasmon and transfer
+them to the cobalt centre in two concerted proton-coupled electron transfers (CEPT) that significantly lowers the energy threshold of the steps as it avoids the formation of higher energy intermediates.\(^{34}\) During the preparation of this work, a study was published with a similar concept, namely a cobalt porphyrin supported on plasmonic that, on illumination, produced H\(_2\).\(^{35}\) Still, there is a clear distinction. In the present contribution, only the Au nanoparticles (Au NPs) are photoactive, contrasting with the published study where the catalyst and Au NPs are photoactive. Thus, their observation might be related to photonic enhancement instead of a plasmonic hot carrier. This contribution also offers more resounding experimental support for the mechanism underpinning the reaction involving the plasmon hot carrier and complex catalyst ligands that are markedly different from what has been published on cobalt systems for HER, including the recent study. The findings distinctly support the involvement of hot electrons in the catalytic process, while the combined spectroscopically approach offers a robust methodology to measure the reaction mechanism on real electrodes.
+
+The as-prepared catalyst has the cobalt centre coordinated into two 1,10-Phenanthrolin-5-amine ligands and a bidentate nitrate group. A second out-sphere nitrate ensures complex neutrality (Figure 1B), consistent with previously reported crystal structures.\(^{33}\) Details on the catalyst, sample preparation, and characterization can be found in supporting information (SI). The optical spectrum of the as-prepared catalyst in dimethylformamide is shown in Figure S2. It displays a strong absorption peak centered at 290 nm with a shoulder at 360 nm, characteristic of phenanthroline complexes.\(^{36}\) Cobalt nitrate complexes are known to have their nitrate exchanged with water,\(^{37}\) which is the solvent used to attach the complex to the Au NPs. Therefore, the complex dissolved in dimethylformamide was titrated with water to evaluate if this occurred. Figure 1A shows the increase of the UV-Vis shoulder located at 360 nm, with an increase in water content, saturating at around 20% water. The exchange was also confirmed by the X-ray photoelectron spectroscopy (XPS) analysis. The N \(I_s\) region in Figure S7, acquired in the vacuum after introducing the electrode in the analysis chamber, displays a sharp peak centered at 398.6 eV. Such a binding energy value can be assigned to pyridinic nitrogen of the phenanthroline ligand.\(^{38}\) Nitrate ligands are typically found at a binding energy of 408 eV, where the collected spectrum shows no features.\(^{39}\) Notably, adding acid to the aqua complex did not change its optical absorption, suggesting that it forms a stable di-aqua complex from the exchange of the bidentate nitrate ligand by water molecules (Figure 1B and 1D).
+Figure 1. Standard characterization of the cobalt catalysts and photosystem. A) Cobalt complex in dimethylformamide titration with water followed by in-situ UV-Vis, with insert showing the effect of acid in the spectrum, it represents with and without acid in water; B) proposed structure of the catalyst in water, which was used to anchor the catalyst to Au NPs; C) UV-Vis spectra of Au NPs region before and after addition of the cobalt complex on thin film, on insert the amino region followed by infrared spectroscopy: i) catalyst before anchoring; ii) catalyst after anchoring it to the Au NPs; D) photosystem structure used for the photo-electrocatalytic H2 evolution.
+
+The attachment of the complex was followed by UV-Vis and infrared spectroscopies (Figure 1C). The UV-Vis of the Au NPs supported on glass shows the characteristic localized surface plasmon resonant (LSPR) peak at 535 nm, consistent with an average particle size of 8 ± 2 nm (determined by dynamic light scattering, Figure S3)).40 The Au LSPR peak shifts to lower energy when the cobalt catalyst is added (Figure 1C), confirming the anchoring and good electronic coupling between both structures.41 Additionally, it is possible to see the complex absorption shoulder located at 370 nm, corroborating the attachment between catalyst and Au NPs. Unfortunately, the glass support (FTO or cover glass) covers the rest of the complex UV-Vis band precluding their measurement. The anchoring was also confirmed by the
+disappearance of the amino bands in the infrared (Figure 1C insert). Before anchoring the complex has two small peaks located between 3300-3450 cm^{-1} associated with N-H bending modes of primary amino groups, which disappear after coordination to the gold surface. The complete disappearance suggests that the catalyst coordinates to the Au NPs via both 1,10-Phenanthrolin-5-amine ligands, as shown schematically in Figure 1D. Note that this was not observed when Au NPs were absent, confirming the selective anchoring of the catalyst to the Au NPs via the amino groups.\(^{42}\)
+
+![Four subfigures showing electrochemical and photo-electrocatalytic studies](page_370_682_1002_627.png)
+
+Figure 2. Electrochemical and photo-electrocatalytic studies. A) Bulk electrolysis of cobalt catalysts in water and the presence of 3 mM acetic acid, using glassy carbon as the working electrode, Pt wire as the counter electrode, Ag/AgCl as reference electrode, 0.1 M LiCl as supporting electrolyte (pH = 5.2) and scan rate 50mV/s; B) Cyclic voltammetry of the NiO/Au and NiO/Au/Co-catalyst in water using Pt wire as a counter electrode, Ag/AgCl as reference electrode and 0.1 M LiCl as supporting electrolyte (pH = 5.2) (scan rate 50 mV/s); C) Effect of light in the chronoamperometry of the complete photosystem applying -0.65 V potential with the insert showing the catalytic wave when the experiments are performed with 3 mM acetic
+acid (pH = 3.5); D) Effect of light in the chronoamperometry of the NiO/Au and NiO/Au/ligand applying - 0.65 V potential in 3 mM acetic acid and 0.1 M LiCl (pH = 3.5).
+
+Bulk electrolysis of the cobalt catalyst in water using 0.1 M LiCl as a supporting electrolyte without acid (Figure 2A) shows no distinguishable reduction peaks before the onset of the catalytic wave, contrasting with what has been proposed elsewhere.\(^{32}\) Addition of acid led to a peak at -1.2V vs Ag/Ag/Cl, suggesting that in the presence of a significant excess of protons, the catalyst follows a stepwise reduction process.\(^{43}\) In contrast, in the absence of protons, it is a concerted single-step two electrons and two protons additions. Unsurprisingly the presence of protons increases the amplitude of the catalytic wave, showing that proton availability is a determining factor when accessing catalytic performance in this system. Therefore, all the catalytic data is acquired in the presence of 3 mM acetic acid.
+
+Figure 2B shows the cyclic voltammetry of the NiO/Au and NiO/Au/Co-catalyst thin films in water. The NiO/Au shows a reduction peak centred at around - 0.49 V vs Ag/AgCl related to the reduction of gold surface oxygen and weak catalytic wave staring at - 0.90 V vs Ag/AgCl. Adding the cobalt catalyst drastically reduced the peak at - 0.49 V vs Ag/AgCl, suggesting that surface functionalization by the catalyst decreases the amount of adsorbed oxygen on Au NPs. The cobalt loading in the complete system was determined to be 11.4 \( \mu \)g/cm\(^2\) by inductively coupled plasma - optical emission spectrometry (ICP-OES), equating to 0.29 wt.% of Co. The low loading creates issues regarding spectroscopy signal-to-noise but ensures that the activity is primarily due to the catalyst and enables the detection of catalyst degradation evidence.
+
+Chronoamperometry data at - 0.65 V vs Ag/Ag/Cl in the absence and presence of light (CW laser 532 nm, selectively exciting only the Au plasmon) is presented in Figure 2C. The photosystem is responsive to the light, increasing the photocurrent by ca. -15\( \mu \)A (-19\( \mu \)A/cm\(^2\)). The increase in photocurrent was shown to be due to the evolution of H\(_2\), as confirmed by online quadrupole mass spectrometry (QMS) analysis (Figure S4). The response was found to be constant during the cycling of light on and off for the duration of the experiment (ca. 360 min). The findings indicate that the photosystem is relatively stable, and the process is catalytic. The significant increase in evolved H\(_2\) relates to the presence of cobalt catalyst since neither the NiO/Au nor NiO/Au/ligand systems produce significant photocurrent (Figure 2D) and had no detectable hydrogen evolution by online QMS analysis. The observed differences are related to heating effects from the illumination and plasmonic decay.
+
+Having established that the photosystem is responsive to light, it is essential to determine if the enhancement is related to plasmon hot electrons. Transient absorption spectroscopy (TAS) was
+performed to evaluate charge transfer in the photosystem. Excitation of the Au NPs LSPR results in a bleach signal and two small winglets on each side of the bleach to broadening of the LSPR peak\(^{44}\) (see the representative spectrum in SI Figure S5). Figure 3A shows the kinetic traces extracted at 490 nm (edge of the positive winglet) after excitation at 550 nm of the Au NPs, NiO/Au and NiO/Au/Co-catalyst systems. The kinetic traces were fitted with a rising edge and a double exponential decay. The rising edge is assigned to the electron-electron (e-e) scattering lifetime, the shorter exponential decay to electron-phonon (e-ph) scattering lifetime and the longer decay to photo-phonon (ph-ph) scattering lifetime.\(^{25,26,45}\) Recently, we demonstrated that charge transfer can be established from changes in the e-ph lifetime. Both electron and hole transfer reduce the e-ph lifetime compared to the plasmon nanoparticles without charge transfer. Hot electrons reduce the e-ph by taking energy from the resonance. Hot holes decrease the e-ph by injecting cold electrons into the resonance, reducing the average electron temperature.\(^{25,26}\)
+
+The Au NPs on glass have e-e lifetime estimated to be 167 ± 49 fs and an e-ph of 5.1 ± 0.4 ps, which is within what has been published previously.\(^{23}\) When attached to NiO (hole acceptor), the Au NPs e-e increased slightly to 198 ± 85 fs with a noticeable decrease in e-ph lifetime to 3.4 ± 1.0 ps, consistent with what is anticipated if holes are transferred from Au NPs to NiO. The system composed of Au/Co-catalyst has an e-e of 148 ± 30 fs and a significantly shorter e-ph lifetime (4.2 ± 0.3 ps) compared with Au NPs alone. Since the catalyst is expected to be the electron acceptor, the reduction in e-ph lifetime suggests that electrons are transferred from Au NPs to the catalyst. The complete photosystem had an e-e of 198 ± 116 fs and the biggest reduction in the e-ph from 5.1 ± 0.4 ps (Au NPs) to 2.6 ± 1.0 ps, suggesting the hot holes and electrons are transferred to the respective acceptors. In sum, the presence of electron and hole acceptors reduced the e-ph, consistent with charge transfer from Au NPs to the acceptors, with the largest e-ph lifetime reduction observed when both electron and hole acceptors are present.
+
+To corroborate the TAS’s findings, transient infrared absorption spectroscopy (TIRAS) spectroscopy studies were performed. Free carriers absorb strongly in the infrared domain due to the formation of a quasi-metallic state.\(^{46}\) The signal is characterized by broad and featureless infrared absorption, often depicted as a background shift in the infrared spectrum.\(^{47}\) A representative TIRAS data map after LSPR excitation at 550 nm is shown in Figure 3B. Kinetic traces extracted between 4645-4700 nm (2150-2130 cm\(^{-1}\)) were fitted with a rising edge and two exponential decay, ascribed to injection and recombination processes lifetime, respectively. NiO/Au shows a rising edge with a 196 ± 104 fs time component, suggesting fast injection of
+the hole. Most of the injected charge recombines within \(417 \pm 117\) fs (ca. 91%). The complete system displayed a similar injection time (\(100 \pm 23\) fs) and increased recombination time (6.6 \(\pm\) 4.2ps, 85% of the signal). However, an increase in signal amplitude is noticeable, suggesting that more charge is transferred when both acceptors are present. Consequently, the complete system has more charge for the catalysis. In both cases, 5-10% of the charge survives past 1 ns, making it useful for catalytic transformations. The TIRAS confirmed that charge is indeed transferred and that one has electrons surviving long enough to perform H\(_2\) evolution.
+
+![Transient spectroscopy data after LSPR excitation at 550 nm. Four panels showing kinetic traces and TIRAS map.](page_246_624_1012_482.png)
+
+Figure 3. Transient spectroscopy data after LSPR excitation at 550 nm. A) Kinetic traces of Au, NiO/Au, NiO/Au/Co-catalyst, extracted at 490 nm from the TAS measurements; B) TIRAS map of the NiO/Au/Co-catalyst; C) TIRAS kinetic trace extracted at 4705 nm for NiO/Au and NiO/Au/Co-catalyst; D) TIRAS kinetic trace extracted at 4705 nm for Au/ligand and Au/Co-catalyst.
+
+Noteworthy is that TIRAS measurements without the NiO, namely with Au/Co-catalyst and Au/ligand, also show a broad featureless infrared absorption (Figure S6), characteristic of free carrier absorption not localized charge. The kinetic traces in Figure 3D are noisy due to low signal and thus challenging to fit. However, the signal qualitatively shows a concise rising
+component (faster than the instrument response function (ca. 100 fs), suggesting a very fast electron injection. In the case of the Au/ligand, the decay is very fast, but when Co is present, the decay is comparatively slow, with the charge surviving past 1 ns. The shape of the TIRAS signal indicates that electrons are injected into the ligands due to the strong coupling between ligand and Au NPs; the charge is delocalised through the aromatic rings and acts as a free carrier. The presence of Co improves hot electrons' lifetime due to some charge stabilization. Still, the metal is not reduced since this would lead to the disappearance of the ‘free carrier’ infrared absorption behavior, which does not happen for at least 1 ns. The TIRAS observation is also consistent with the absence of Co reduction peaks in the electrochemistry. This is a peculiar observation because from the speculated mechanism of the analogous\(^{32}\) and relatable cobalt complexes,\(^{48,49}\) the metal center plays a central role, undergoing two sequential reductions. At the same time, the ligands are often spectators of the catalytic process.
+
+![O 1s and Co 2p XPS spectra under acidic conditions at different potentials](page_355_670_1042_377.png)
+
+Figure 4. In situ NAP-XPS of NiO/Au/Co-catalyst under variable potential in the presence of acid with an X-ray photon energy of \(5000\) eV. The potentials are vs Ag/AgCl reference electrode. A) O 1s signals (the inset shows a magnification of the prominent peaks, highlighting the binding energy shift due to the potential applied); B) Co 2p signals.
+
+To corroborate the peculiar finding, electrochemical experiments combined with in situ near ambient pressure (NAP)-XPS measurements were performed using the dip-and-pull approach\(^{50}\) in the absence (Figure S8) and presence (Figure 4) of acetic acid. The measurements were
+performed with the mesoporous films used for photocatalytic data. This significantly reduces the signal intensity and requires adaptation of the dip-and-pull method to reduce the amount of electrolyte trapped in the porous film.
+
+Briefly, only the lowest segment of the sample was dipped into the electrolyte solution, while the rest formed a liquid film via capillary forces. After some equilibration time (ca. 30 min), the liquid film settled to a point where the photoemission signals of the electrode (O \(1s\), Co \(2p\) and N \(1s\)) were detectable together with the signal of liquid and gas phase water (O \(1s\)). Figures 4A and S8A show the O \(1s\) spectra acquired in situ at different applied potentials. Three prominent peaks, centered at approximately 530.5 eV, 532.5 eV and 534.5 eV, are assigned to lattice oxygen (electrode), liquid water (thin electrolyte film on top of the electrode) and gas phase water, respectively.\(^{51,52}\) Potential control and the availability of a continuous liquid film up to the position monitored by XPS were surveyed by the shifting of the O \(1s\) signal according to the applied potential. Figures 4A and S8A show a shift in the central peak of O \(1s\), centered at 532.5 eV and assigned to liquid water (thin electrolyte layer), as the potential was applied (detected as a positive binding energy shift, proportional to the potential applied to the working electrode), confirming experiment validity and thus access to Co oxidation state at different potentials.
+
+Confident that NAP-XPS experiments reflected the Co oxidation state at different potentials, one can proceed with the analysis of Co \(2p\) region (Figure 4B). Despite the low signal-to-noise, the main features of Co \(2p\) peaks can be detected. The Co \(2p_{3/2}\) prominent peak is centered at around 780 eV, ascribed to Co\(^{2+}\) as expected.\(^{53,54}\) The peak position does not shift with the applied potential, as the working electrode is set to ground potential during the experiment Nevertheless, also a peak shift due to the formation of a different cobalt species is not observed either in the presence or in the absence of acetic acid (Figure S8B). Indeed, the reduction of cobalt from 2+ to the metallic state leads to a shift of the binding energy of the \(2p_{3/2}\) peak by about 2 eV, from 780 to 778 eV.\(^{55,56}\) Figure 4B also shows that at -0.65 V vs Ag/AgCl the cobalt signal decreases in intensity. Such a potential is sufficient for the evolution of H$_2$ since bubbles were detected at the electrode and marked the onset of the catalytic wave (Figure S9). Thus, the decrease in signal-to-noise ratio relates to experimental conditions since the signal statistics are affected by the H$_2$ bubbles formation at the highest potential. Therefore, only Co\(^{2+}\) is present until H$_2$ formation; thus, the reduction peak at lower voltages relates to the reduction of the phenanthroline ligands (Figure S9). Since cobalt reduction is required to evolve hydrogen, not seeing its reduction suggests that the cobalt centre reduction and protonation steps are fast and
+just before the hydrogen evolution, i.e., changes at the cobalt not rate limiting and cannot be detected due to the NAP-XPS temporal resolution.\(^{37}\)
+
+The observation Similar results occur in the absence of acid but the onset potential of H\(_2\) evolution was shifted to 0.8 V vs Ag/AgCl, i.e., about 0.15-0.2 V, as observed in the catalysis (Figure S9). Furthermore, the Co \(2p\) signal intensity detected was lower. The following hypotheses can be postulated, to explain such a behavior: i) the liquid electrolyte film was thicker in this case, attenuating more the signal of the substrate; ii) the surface of the electrode is slightly different in the absence of the acid, suggested by the more prominent electrode-related shoulder (centered at ca. 530.5 eV) in the O \(1s\) spectra (Figure S8).
+
+Figure 5 shows a schematic representation of the hypothetic catalytic cycle. Excitation of Au plasmon results in the formation of hot electrons and holes as part of its resonance decoherence via Landau damping. The hot holes are transferred to the NiO and react at the counter electrode leading to O\(_2\) production. The electrons are transferred to the phenanthroline ligands. Once both ligands are charged, two very fast concerted proton-coupled electron transfer (CPET) steps take place, resulting in the reduction and protonation of the cobalt catalyst and, subsequently, H\(_2\) evolution.
+
+![Schematic representation of the catalytic cycle leading to H2 evolution.](page_682_1042_1087_377.png)
+
+Figure 5. **Schematic representation of the catalytic cycle leading to H\(_2\) evolution.** The NiO was omitted to improve figure legibility.
+In conclusion, the photosystem proposed herein confirmed the involvement of hot electrons in the H$_2$ evolution process, settling the hot debate on their participation in photocatalysis. The proposed photosystem is not stable at water thermolysis conditions, thus precluding temperature (heat) from being the culprit for the observed H$_2$ evolution. Ultrafast spectroscopic measurements confirmed the charge to transfer to the respective acceptors. Moreover, and in conjunction with NAP-XPS under variable potential, a reaction mechanism was postulated in which the cobalt catalyst ligands have a significant role. They accept the electrons coming from Au plasmon and, in two quick CEPT steps, reduce and protonate the metal centre resulting in the evolution of the product.
+
+Acknowledgements: the authors would like to thank the Paul Scherrer Institute for providing access to the Phoenix beamline at the Swiss Light Source.
+
+Funding:
+
+Olle Engkvists stiftelse (OES) (grant no. 210-0007)
+Knut & Alice Wallenberg Foundation (Grant No. 2019-0071)
+Swedish Research Council (grant no. 2019-03597)
+
+Competing interests: Authors declare that they have no competing interests.
+
+Supplementary Materials
+Materials and Methods
+Figs. S1 to S8
+
+References:
+__________________________
+1. Sá, J., Tagliabue, G., Friedli, P., Szlachetko, J., Rittmann-Frank, M. H., Santomauro, F. G., Milne, C. J., Sigg, H. Direct observation of charge separation on Au localized surface plasmon. Energy Environ. Sci. **6**, 3584–3588 (2013).
+2. Mubeen, S., Lee, J., Singh, N., Krämer, S., Stucky, G. D., Moskovits, M. An autonomous photosynthetic device in which all charge carriers derive from surface plasmons. *Nat. Nanotechnol.* **8**, 247–252 (2013).
+3. Aslam, U., Rao, V. G., Chavez, S., Linic, S. Catalytic conversion of solar to chemical energy on plasmonic metal nanostructures. Nat. Catal. **1**, 656–665 (2018).
+4. Li, R., Cheng, W.-H., Richter, M. H., DuChene, J. S., Tian, W., Li, C., Atwater, H. A. Unassisted Highly Selective Gas-Phase CO$_2$ Reduction with a Plasmonic Au/p-GaN Photocatalyst Using H$_2$O as an Electron Donor. ACS Energy Lett. **6**, 1849-1856 (2021).
+5. Kumari, G., Zhang, X., Devasia, D., Heo, J., Jain, P. K. Watching visible light-drive CO$_2$ reduction on a plasmonic nanoparticle catalyst. ACS Nano **12**, 8330-8340 (2018).
+6. Marimuthu, A., Zhang, J., Linic, S. Tuning selectivity in propylene epoxidation by plasmon mediated photo-switching of Cu oxidation state. Science **339**, 1590–1593 (2013).
+7. Linic, S., Aslam, U., Boerigter, C., Morabito, M. Photochemical transformations on plasmonic metal nanoparticles. Nat. Mater. **14**, 567–576 (2015).
+8. Vadai, M., Angell, D. K., Hayee, F., Sytwu, K., Dionne, J. A. In-situ observation of plasmon-controlled photocatalytic dehydrogenation of individual palladium nanoparticles. Nat. Commun. **9**, 4658 (2018).
+9. Contreras, E., Nixon, R., Litts, C., Zhang, W., Alcorn, F. M., Jain, P. K. Plasmon-assisted ammonia synthesis. J. Am. Chem. Soc. **144**, 10743-10751 (2022).
+10. Chen, K., Wang, H. Plasmon.driven photocatalytic molecular transformations on metallic nanostructure surfaces: mechanistic insights gained from plasmon-enhanced Raman spectroscopy. Mol. Syst. Des. Eng. **6**, 250-280 (2021).
+11. Frontiera, R, Guenke, N. L., van Duyne, R. Fano-line resonances arising from long-lived molecule-plasmon interactions in colloidal nanoantennas. Nano Lett. **12**, 5989-5994 (2012.)
+12. Wilson, A. J., Mohan, V., Jain, P. K.. Mechanistic understading of plasmon-enhanced electrochemistry. J. Phys. Chem. C **123**, 29360-29369 (2019).
+13. Chen, K., Wang, H. Plasmon.driven photocatalytic molecular transformations on metallic nanostructure surfaces: mechanistic insights gained from plasmon-enhanced Raman spectroscopy. Mol. Syst. Des. Eng. **6**, 250-280 (2021).
+14. Brongersma, M. L., Halas, N. J., Nordlander, P. Plasmon-induced hot carrier science and technology. Nat. Nanotechnol. **10**, 25–34 (2015).
+15. Sivan, Y., Un, I. W., Dubi, Y. Assistance of metal nanoparticles in photocatalysis – nothing more than a classical heat source. Faraday Discuss. **214**, 215-233 (2019).
+16. Baffou, G., Bordacchini, I., Baldi, A., Quidant, R. Simple experimental procedures to distinguish phototermal from hot-carrier processes in plasmonics. Light Scie. Appl. **9**, 108 (2020).
+17. Zhan, C., Liu, B.-W., Huang, Y.-F., Hu, S., Ren, B., Moskovits, M., Tian, Z.-Q.
+Disentangling charge carrier from photothermal effects in plasmonic metal nanostructures. Nat. Commun. **10**, 2671 (2019).
+
+18. Zhang, X., Li, X., Reish, M. E., Zhang, D., Su, N. Q., Gutiérrez, Y., Moreno, F., Yang, W., Everitt, H. O., Liu, J. Plasmon-Enhanced Catalysis: Distinguishing Thermal and Nonthermal Effects. Nano Lett. **18**, 1714–1723 (2018).
+
+19. Jain, P. K. Taking the heat off of plasmonics chemistry. *J. Phys. Chem. C* **123**, 24347-24351 (2019).
+
+20. Kamarudheen, R., Aalbers, G. J. W., Hamans, R. F., Kamp, L. P. J., Baldi, A.
+Distinguishing amongs all possible activation mechanisms of a plasmon-driven chemical reaction. ACS Energy Lett. **5**, 2605-2613 (2020).
+
+21. Rossi, T. P., Erhart, P., Kuisma, M. Hot-Carrier Generation in Plasmonic Nanoparticles: The Importance of Atomic Structure. ACS Nano **14**, 9963-9971 (2020).
+
+22. Fann, W. S., Strotz, R., Tom, H. W. K., Bokor, J. Electron thermalization in gold. *Phys. Rev. B* **46**, 13592 (1992).
+
+23. Link, S., El-Sayed, M. A. Spectral Properties and Relaxation Dynamics of Surface Plasmon Electronic Oscillations in Gold and Silver Nanodots and Nanorods. *J. Phys. Chem. B* **103**, 8410-8426 (1999).
+
+24. Furube, A., Du, L., Hara, K., Katoh, R., Tachiya, M. Ultrafast Plasmon-Induced Electron Transfer from Gold Nanodots into TiO$_2$ Nanoparticles. *J. Am. Chem. Soc.* **129**, 14852–14853 (2007).
+
+25. Tagliabue, G., DuChene, J. S., Abdellah, M., Habib, A., Gosztola, D. J., Hattori, Y., Cheng, W.-H., Zheng, K., Canton, S. E., Sundararaman, R., Sá, J., Atwater, H. A. Ultrafast Hot-Hole Injection Modifies Hot-Electron Dynamics in Au/p-GaN Heterostructures. *Nat. Mater.* **19**, 1312–1318 (2020).
+
+26. Hattori, Y., Abdellah, M., Meng, J., Zheng, K., Sá, J. Simultaneous hot electron and hole injection upon excitation of gold surface plasmon. *J. Phys. Chem. Lett.* **10**, 3140–3146 (2019).
+
+27. https://www.energy.gov/eere/fuelcells/hydrogen-production-thermochemical-water-splitting (Accessed on 2023-02-20)
+
+28. Odobel, F., Pellegrin, Y. Recent Advances in the Sensitization of Wide-Band-Gap Nanostructured P-Type Semiconductors. Photovoltaic and Photocatalytic Applications. *J. Phys. Chem. Lett.* **4**, 2551-2564 (2013).
+29. Gardner, J. M., Beyler, M., Karnahl, M., Tschierlei, S., Ott, S., Hammarström, L. Light-Driven Electron Transfer between a Photosensitizer and a Proton-Reducing Catalyst Co-Adsorbed to NiO. J. Am. Chem. Soc. **134**, 19322-19325 (2012).
+
+30. Tong, L., Iwase, A., Nattestad, A., Bach, U., Weidelener, M., Gotz, G., Mishra, A., Bäuerle, P., Amal, R., Wallace, G. G., Mozer, A. J. Sustained Solar Hydrogen Generation Using a Dye-Sensitized NiO Photocathode/BiVO$_4$ Tandem Photo-Electrochemical Device. Energy Environ. Sci. **5**, 9472-9475 (2012).
+
+31. Nakamura, K., Oshikiri, T., Ueno, K., Wang, Y., Kamata, Y., Kotake, Y., Misawa, H. Properties of Plasmon-Induced Photoelectric Conversion on a TiO$_2$/NiO p–n Junction with Au Nanoparticles. *J. Phys. Chem. Lett.* **7**, 1004-1009 (2016).
+
+32. Luo, S.-P., Tang, L.-Z., Zhan, S.-Z. A cobalt(II) complex of 2,2-bipyridine, a catalyst for electro- and photo-catalytic hydrogen production in purely aqueous media. *Inorg. Chem. Commun.* **86**, 276-280 (2017).
+
+33. Sharma, S., Toupet, L., Arjmand, F. *De novo* design of a hydrolytic DNA cleavage agent, mono nitratobis(phen)cobalt(II) aqua nitrate complex. *New J. Chem.* **41**, 2883-2886 (2017).
+
+34. Savéant, J. M. Electrochemical approach to proton-coupled electron transfers: recent advances. *Energy Environ. Sci.* **5**, 7718-7731 (2012).
+
+35. Sheng, H., Wang, J., Huang, J., Li, Z., Ren, G., Zhang, L., Yu, L., Zho, M., Li, X., Li, G., Wang, N., Shen, C., Lu, G. Strong synergy between gold nanoparticles and cobalt porphyrin induces highly efficient photocatalytic hydrogen evolution. *Nat. Commun.* **14**, 1528 (2023).
+
+36. Al-Omair, M. A. Biochemical activities and electronic spectra of different cobalt phenanthroline complexes. *Arabian J. Chem.* **12**, 1061-1069 (2019).
+
+37. Katzin, L. I., Gebert, E. Spectrophotometric Investigation of Cobaltous Nitrate in Organic Solvents. *J. Am. Chem. Soc.* **72**, 5455-5463 (1950).
+
+38. Lazar, P., Mach, R., Otyepka, M. Spectrophotometric Investigation of Cobaltous Nitrate in Organic Solvents. *J. Am. Chem. Soc.* **72**, 5455-5463 (1950).
+
+39. Beard, B. C. cellulose nitrate as a binding energy reference in N(1s) XPS studies of nitrogen-containing organic molecules. *Appl. Surf. Scie.* **45**, 221-227 (1990).
+
+40. Silveira, V. R., Bericat-Vadell, R., Sá, J. Photoelectrocatalytic Conversion of Nitrates to Ammonia with Plasmon Hot Electrons. *J. Phys. Chem. C* **127**, 5425-5431 (2023)-
+
+41. Rossi, T. P., Shegai, T., Erhart, P., Antosiewicz, T. J. Strong plasmon-molecule coupling at the nanoscale revealed by first-principles modeling. *Nat. Commun.* **10**, 3336 (2019).
+42. Pavliuk, M. V., Fernandes, A. B., Abdellah, M., Fernandes, D. L. A., Machado, C. O., Rocha, I., Hattori, Y., Paun, C., Bastos, E. L., Sá, J. Nano-hybrid plasmonic photocatalyst for hydrogen production at 20% efficiency. Scie. Rep. 7, 8670 (2017).
+43. Li, C.-B., Bagnall, A. J., Sun, D., Rendon, J., Koepf, M., Gambarelli, S., Mouesca, J.-M., Chavarot-Kerlidou, M., Artero, V. Electrocatalytic reduction of protons to dihydrogen by the cobalt tetraazamacrocyclic complex [Co(N4H)Cl2]+: mechanism and benchmarking of performances. Sustain. Energy Fuels 6, 143-149 (2022).
+44. van Turnhout, L., Hattori, H., Meng, J., Zheng, K., Sá, J. Direct Observation of a Plasmon-Induced Hot Electron Flow in a Multimetallic Nanostructure. Nano Lett. 20, 8220-8228 (2020).
+45. Groeneveld, R. H. M., Sprik, R., Lagendijk, A. Femtosecond spectroscopy of electron-electron and electron-phonon relaxation in Ag and Au. Phys. Rev. B 51, 11433 (1995).
+46. Antila, L. J., Santomauro, F. G., Hammaström, L., Fernandes, D. L. A., Sá, J. Hunting for the elusive shallow traps in TiO2. Chem. Commun. 51, 10914-10916 (2015).
+47. Berger, T., Sterrer, M.; Diwald, O., Knözinger, E., Panayotov, D., Thompson, T. L., Yates Jr, J. T. Light-Induced Charge Separation in Anatase TiO2 Particles. J. Phys. Chem. B 109, 6061-6068 (2005).
+48. Querryiaux, N., Sun, D., Fize, J., Pécaut, J., Field, M. J., Chavarot-Kerlidou, M., Artero, V. Electrocatalytic Hydrogen Evolution with a Cobalt Complex Bearing Pendant Proton Relays: Acid Strength and Applied Potential Govern Mechanism and Stability. J. Am. Chem. Soc. 142, 274-282 (2020).
+49. Eckenhoff, W. T, McNamara, W. R., Du, P., Eisenberg, R. Cobalt complexes as artificial hydrogenases for reductive side of water splitting. Biochim. Biophys. Acta: Bioenergetics 1827, 958-973 (2013).
+50. Novotny, Z., Aegerter, D., Comini, N., Tobler, B., Artiglia, L., Maier, U., Moehl, T., Fabbri, E., Huthwelker, T., Schmidt, T., Ammann, M., van Bokhoven, J. A., Raabe, J., Osterwalder, J. Probing the solid–liquid interface with tender x rays: A new ambient-pressure x-ray photoelectron spectroscopy endstation at the Swiss Light Source. Rev. Scie. Instrum. 91, 023103 (2020).
+51. Axnanda, S., Crumlin, E. J., Mao, B., Rani, S., Chang, R., Karlsson, P. G., Edwards, M. O. M., Lundvist, M., Moberg, R., Ross, P., Hussain, Z., Liu, Z. Using “Tender” X-ray Ambient Pressure X-Ray Photoelectron Spectroscopy as A Direct Probe of Solid-Liquid Interface. Scie. Rep. 5, 9788 (2015).
+52. Favaro, M., Jeong, B., Ross, P. N., Yano, J., Hussain, Z., Liu, Z., Crumlin, E. J.
+Unravelling the electrochemical double layer by direct probing of the solid/liquid interface
+Nat. Commun. **7**, 12695 (2016).
+
+53. Lázaro-Martínez, J. M., Lupano, L. V. L., Piehl, L. L., Rodríguez-castellón, E., Dall’Orto, V. C. New Insights about the Selectivity in the Activation of Hydrogen Peroxide by Cobalt or Copper Hydrogel Heterogeneous Catalysts in the Generation of Reactive Oxygen Species. *J. Phys. Chem. C* **120**, 29332-29347 (2016).
+
+54. Khalil, T. E., Soliman, S. M., Khalil, N. A., El-Faham, A., Foro, S., El-Dissouky, A. Synthesis, structure, X-ray photoelectron spectroscopy (XPS), and antimicrobial, anticancer, and antioxidant activities of Co (III) complexes based on the antihypertensive hydralazine. *Appl. Org. Chem.* **36**, e6565 (2022).
+
+55. Bridge, M. E., Lambert, R. M. Oxygen chemisorption, surface oxidation, and the oxidation of carbon monoxide on cobalt (0001). *Surf. Scie.* **82**, 413-424 (1979).
+
+56. Hyman, M. P., Vohs, J. M. Reaction of ethanol on oxidized and metallic cobalt surfaces. *Surf. Scie.* **605**, 383-389 (2011).
+
+57. Moonshiran, D., Gimbert-Suriñach, C., Guda, A., Picon, A., Lehmann, C. S., Zhang, X., Doumy, G., March, A. M., Benet-Buchholz, J., Soldatov, A., Llobet, A., Southworth, S. H. Tracking the Structural and Electronic Configurations of a Cobalt Proton Reduction Catalyst in Water. *J. Am. Chem. Soc.* **138**, 10586-10596 (2016).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• H2evolutionSI.docx

056cacada9650bd2ff6e41d24cf77d3f922bae4f244b4a05a96d2273dc50b22f/peer_review/peer_review.md

@@ -0,0 +1,705 @@
+Peer Review File
+
+dFLASH; dual FLuorescent transcription factor Activity Sensor for Histone integrated live-cell reporting and high-content screening
+
+Corresponding Author: Dr David Bersten
+
+Parts of this Peer Review File have been redacted as indicated to maintain the confidentiality of unpublished data.
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.
+
+Version 0:
+
+Reviewer comments:
+
+Reviewer #1
+
+(Remarks to the Author)
+Allen et al present work developing fluorescent reporters as an optimised system for drug screening for activators or inhibitors. It is essentially a methods paper and the methodology is clearly described. The benefit of a modular system and having an internal fluorescent control is emphasised. It does seem a useful system but it is less clear whether dFLASH offers a substantial advantage compared to other screening systems (e.g. with luciferase) that are typically done at higher throughput for compound screens. TF fluorescent reporters have been used in various forms for a long time. The screens presented show the potential of the approach but the compound library has not yet yielded drugs that would really be classed as TF inhibitors, which is a current challenge for HIF1.
+
+Specific points:
+- The actual number of responsive elements used in the main reporter systems is not clearly explained. 12X HRE seems excessive and implies that the system may not be as sensitive as claimed. The concern here is that a high number of REs becomes less biologically relevant, and this may be important when considering drug screens. The other reporters seem to only have 5X RE copies. The correct diagrammatical representation of the REs should be included in Figure 2.
+
+-The time taken to active the HIF reporter seems slow, at around 10 hrs. HIF should be stabilised within minutes and HIF targets should increase at the message level within a few hours. Why is this induction slower? This may be a concern for drug screening.
+-None of the dFLASH-HRE experiments seem to have been done in different oxygen gradients. Are similar effects seen? The use of DMOG will inhibit both PHDs and FIH but do specific PHD inhibitors activate the reporter similarly?
+
+-Regarding the drug screens, while there is good concordance with compounds that activate, there is more variability in the inhibitor screens between 24 and 36 hours. As there are already clinically useful drugs to activate HIFs, the challenge (as stated in the manuscript) is to find specific inhibitors. This variability may relate to the stability of Tomato and GFP, which is different from HIF. Therefore, how dynamic is this system? E.g. how long does it take for Tomato levels to return to baseline once DMOG is washed out?
+
+-The dFLASH-PRE seems a more sensitive reporter system that dFLASH-HRE. Accept it is likely beyond the scope of this work but it would be helpful to have some indication from the authors as to the utility of this reporter in chemical screens.
+
+- How was the drug toxicity assessed in the small molecule screens? Were the cells stained for cell death/apoptosis, or was it judged by GFP signal/cell morphology? GFP alone may not be a good indication of toxicity as it is likely to be very stable.
+
+Reviewer #2
+
+(Remarks to the Author)
+The work by Allen et al. presents dual FLuorescent transcription factor Activity Sensor for Histone integrated live-cell reporting (dFLASH), a modular sensor for TF activity that can be readily integrated into cellular genomes. The authors demonstrate that dFLASH platforms can sense the regulation of endogenous Hypoxia Inducible Factor (HIF) and Progesterone receptor (PGR) activities, as well as regulated coactivator recruitment to a synthetic DNA-Binding Domain-Activator Domain fusion proteins. Finally, the authors test the utility of this platform for functional genomics applications by using CRISPRoff to modulate the HIF regulatory pathway, and for drug screening by using high content imaging in a bimodal design to isolate activators and inhibitors of the HIF pathway.
+The manuscript is well written, the experiments are very thoroughly executed and the proposed tool is certainly of great interest for the field. However, the impact of the study is likely to increase by addressing the following questions experimentally:
+
+Major:
+
+Did the authors address the degree of cell death and apoptosis after integration of dFLASH constructs? In this regard, it would also be important to test the dFLASH in non-transformed, non-immortalized and non-cancerous cells. Would the dFLASH tool work in primary cells?
+
+Especially with regard to the studies on the HIF pathway, it would be important to correlate the reporter kinetics after DMOG treatment shown in figure 2 with HIF protein levels and eventually mRNA expression of (a) HIF target gene(s).
+Along these lines and despite the potential issues with fluorescent reporters in low oxygen conditions, how would the dFLASH reporter signal look like in a more physiological hypoxia/reoxygenation experiment?
+
+Minor:
+
+The figure organization could benefit from moving some of the data from the suppl. figures to the main figures. Lots of data that are required to understand the manuscript are shown in the suppl. figures while the main figures are rather spares in panels.
+
+Reviewer #3
+
+(Remarks to the Author)
+dFLASH: Dual Fluorescent transcription factor Activity Sensor for Histone integrated live-cell reporting and high-content screening
+
+Comments to the authors
+
+Allen et al report the development of dFLASH platform that is a modular transcription factor sensor that can be used in live cell reporting for multiple downstream applications such as functional genomics screens, drug discovery and synthetic biology. This system consists of two dual color reporters (an internal control nucEGFP as well as dTomato reporter as a readout). The authors demonstrate the use of this reporter system for readout of endogenous HIF and PGR related pathways. Together these authors developed a system that can be used for a variety of applications. Major strengths of this system are that it can be applied to multiple cell types and the modular design also allows researchers to study various endogenous processes. The dual reporter system also allows readouts through flow cytometry or imaging. Furthermore, they successfully used this system to identify novel inhibitors and activators of the HIF pathway in screens. However, it would useful if the authors could perhaps address more strongly, the ways in which their system is advantageous or different from previous reporter systems. Alternatively, the authors could provide a comparison of previous reporter systems to better communicate the impact or uniqueness of their system. Second, adding more quantifications to some of the imaging and flow cytometry data could also strengthen the results. Overall the data presented in this manuscript is convincing and the applications of the reporter system are strongly demonstrated. Therefore, it could provide an amenable system for reporter-based screening applications to study other endogenous pathways as well.
+
+Additional comments:
+1. Within the introductions and discussions, the authors could discuss previous systems that were built for TF sensing, for example luciferase or reporter systems have been used in the past to enable HF1a pathway analysis (See a). Discussion of these systems alongside the one that the authors have built would be useful in communicating the impact of the developed reporter system. Additionally, the authors could also determine how their system addresses some of the shortcomings of the previous systems and the types of pathways that can be analyzed by the system in order to make the discussion more comprehensive.
+a. An optimized reporter of the transcription factor hypoxia-factor inducible factor 1a reveals complex HIF-1a activation dynamics in single cells https://www.jbc.org/article/S0021-9258(23)00241-7/fulltext
+2. In Figure 1B, the authors are showing how their built reporter system can be used for imaging-based analysis. This figure can be strengthened if they include the pixel intensities of the GFP and RFP channels from select regions of the microscopy images. More specifically it would be helpful to see that the RFP increases and the GFP remains constant. It would also help account for the background fluorescence of the Tomato reporter in using the imaging application. Finally, they could also consider adding black/white image panels.
+3. The schematic in Figure 1B is a little misleading, can the authors include more details or enhance the figure to guide the
+reader? For example, add +DMOG etc. or potentially remove the arrow.
+4. Within the flow plots shown in Figure 2, D-F it would help if the authors quantified the percentage of cells that respond to the given treatments. For example, they can show their gating strategies. They should also comment on the fact that these are not clonal lines and are a heterogeneous population, since they are following up on these experiments with single cells clones.
+5. Could the authors explain why some of their cells do not express dTomato despite antibiotic selection? A two-axis flow plot with gating may be more comprehensive for the main figure (like those shown in Supp Fig2). For example, their flow plots can indicate a range of fluorescence intensities in the tomato channel, how would these numbers change if the authors gated for the highest GFP expressing cells (where the reporter is integrated in euchromatin region). Would that in turn affect the Tomato readout?
+6. Figure 4C/Supplementary Fig 6: The authors can include a single cell flow plot and gating information in the figure to make it more comprehensive for the reader (including numerical/gating details e.g. % of cells).
+7. The authors could clarify the reasoning behind using different promoter versions for CRISPRoffv2.1
+8. It appears that the clonal cell lines in Figure 4 produce multiple dTomato peaks, does that indicate that the line is slightly leaky for expression of dTomato? How does it compare to no fluorescence - doing an overlay would be helpful? Could potentially be a challenge for use in screens if there is indeed background.
+9. In the CRISPRoff experiment, did the authors also include a non-targeting guide? Might be useful to show or include as control
+10. Not essential but the authors can confirm VHL protein knockdown in cells post CRISPRoff induced silencing using a western blot (shows sensitivity of the system).
+11. Discuss how this reporter system can be of better use compared to current systems for screening methods
+12. For Figure 5, the authors could describe the Activation Screen and Inhibitor with more detail or clarity within the text such that it is easy to follow, or describe the workflow and expected results with more clarity.
+13. For Figure 5B/C, the authors could label the top hits from their screen in the plot
+14. For Figure 5B, the squares, circles and triangles can be a little difficult to distinguish, the authors could consider using different colors for clarity.
+
+Version 1:
+
+Reviewer comments:
+
+Reviewer #1
+
+(Remarks to the Author)
+Overall, the authors have not adequately addressed the issues raised during the initial submission, and the manuscript has not been significantly improved. There does not seem to be a much of an advance here.
+
+- Novelty of the HRE reporter. The authors present a table comparing some prior HIF reporters, which is selective and not accurate. Other HRE-fluorescent reporters exist and have been described over the past 20 years (e.g. PMID 11774035, 17270179, 26598532). Some have dual colors, destabilised fluorescent proteins (PEST or adding the ODDD), and have been shown to work in cell lines and in vivo settings. The HRE-HIF-ODD reporter is not constitutively active (minimal SV40 promoter x3HRE). The addition of the internal fluorescent control remains the only novelty of the reporter, but the published NanoFIRE manuscript now referred to in the authors’ response (describing a similar approach to dFLASH but with luciferase) already details the use of the 12XHRE dFLASH fluorescent system.
+
+- The authors state that endogenous enhancers contain 5-6 sites for TFs, but still use 12 HREs. This necessity for 12x HRE is at odds with other prior reporters (e.g. HypoxCR).
+
+- The variability in the inhibitory screens remains a concern. The stability of Tomato is not an advantage as stochastic/basal induction of the reporter will occur (as noted), and dFLASH will not detect turning a TF off. Same argument applies for reoxygenation.
+
+- The assays of cell death or drug toxicity have not been adequately addressed.
+
+- The HRE screen has so far not identified any interesting results.
+
+- The utility of the system in other cell types (SFig. 3D) is not that impressive, with a fairly minimal induction.
+
+Reviewer #2
+
+(Remarks to the Author)
+The authors have adressed all concerns in the revised manuscript. Congratulations!
+
+Reviewer #3
+(Remarks to the Author)
+The authors have successfully addressed all the questions and made modifications to the manuscript accordingly.
+
+Version 3:
+
+Reviewer comments:
+
+Reviewer #4
+
+(Remarks to the Author)
+I was asked primarily to adjudicate the different views of three reviews of this paper and its revision. In my view, the authors have done an excellent job in addressing the comments, and I feel that the comments of Rev 1 are asking too much. Yes, reporters have been developed before including around hypoxia, but the internal control *does* add a lot of value, as does nuclear localization, and the framework appears to be more broadly useful for reporters in general, which to me seem to often be put out into the world in haphazard and poorly controlled ways. I also don’t think it’s fair to knock the authors for citing their own preprint. The Table that the authors provide now (which I assume will be included in the paper either as main or supplement) is useful and transparent, and to my knowledge accurate (and I say this as the developer of one of the methods listed there, which absolutely has limitations that are fairly noted there). Both imaging and non-imaging based reporter methods need better controls, and the internal control is a terrific way to go. I might suggest that the authors also add scQers (Nature Methods 2024) to the table, which has a sequence-based internal control that I think reinforces the broader point about how important this is in any context where one is looking at single cells rather than bulk (in this case by imaging, in that case by sequencing).
+dFLASH response to reviewers
+
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+Allen et al present work developing fluorescent reporters as an optimised system for drug screening for activators or inhibitors. It is essentially a methods paper and the methodology is clearly described. The benefit of a modular system and having an internal fluorescent control is emphasised. It does seem a useful system but it is less clear whether dFLASH offers a substantial advantage compared to other screening systems (e.g. with luciferase) that are typically done at higher throughput for compound screens. TF fluorescent reporters have been used in various forms for a long time.
+
+We thank reviewer #1 and #3 for the request for a clearer explanation of the advantages of the dFLASH system over other analogous approaches.
+
+After discussion with the editor and comparison with analogous fluorescent reporter systems we concluded that no other systems have been developed that can be directly benchmarked in the experimental paradigms outlined here. We have specifically excluded luciferase systems from this as they cannot be applied to live cell high content or FACS based analysis/enrichment as described here. We have developed and published a companion bioluminescent system called NanoFIRE (doi: 10.3390/biom13101545.), that like the dFLASH system robustly reports (20-150x induction, Z' >0.5) on endogenous and synthetic TF pathways in stable cell lines.
+
+The automated nuclear segmentation enabled by the dual nuclear fluorescent proteins as outline by the dFLASH system allows high-throughput normalised high-content imaging not enabled by other systems.
+
+Moreover, we optimised and benchmark the dFLASH system in a variety of ways in the paper including downstream promoter composition and the length of the linker between the enhancer and promoter all which allows a robust reporting platform. Further, we benchmark the utility of the system against two endogenous unrelated TF pathways and one synthetic reporter system using robust screening metrics (Z') which evaluate the screen performance. This is not often observed even in large scale drug screening papers but commonplace in more industry-focused settings.
+
+We agree that specific discussion of the advantages, optimisation and benchmarking of the dFLASH system has not been clearly outlined in the paper and as such we have made some changes to the text. (Page #12 Paragraph#1 line #722-782).
+
+The screens presented show the potential of the approach but the compound library has not yet yielded drugs that would really be classed as TF inhibitors, which is a current challenge for HIF1.
+We agree with the sentiment of Reviewer #1 who outlines that true HIF1 inhibitors are a significant challenge and highly sort after. We describe a proof of principle small scale screen to demonstrate the utility of the system to both identify positive
+and negative regulators in a single experimental paradigm that could act directly or indirectly on the HIF pathway. Extending from this work, the dFLASH system is currently being applied to much larger (>300,000 compound screen) by independent drug screen laboratories reporting similar robust screening metrics presented here. Unfortunately, this in beyond the scope of this paper.
+
+Specific points:
+- The actual number of responsive elements used in the main reporter systems is not clearly explained. 12X HRE seems excessive and implies that the system may not be as sensitive as claimed. The concern here is that a high number of REs becomes less biologically relevant, and this may be important when considering drug screens. The other reporters seem to only have 5X RE copies. The correct diagrammatical representation of the REs should be included in Figure 2.
+
+This has now been corrected this in the schematic for Figure 2. Number of repeats and schematic of a 12xHRE and the 5xPRE & 5xGRE are now displayed for reader clarity, attached below.
+a.
+
+endogenous
+HIF-1α
+Fe2+
+2-OG O2
+PHD/FH
+Hypoxia
+DMOG
+ARNT
+CBP/p300
+PGK/CMV
+EGFP
+5' TshRE 3'
+5' SshRE 3'
+
+b.
+
+Progesterone (P) P-like ligands
+(PKBPS2) PR HSP90
+endogenous
+PR
+CovAct
+PGK/CMV
+EGFP
+5' Prm 3'
+5' SshRE 3'
+
+c.
+
+Dox
+rtTA
+TRE3G
+GAL4DBD-HIFCAD
+HIFCAD
+CAL4DBD
+pMin
+PGK/CMV
+EGFP
+5' SshRE 3'
+
+fih
+
+d.
+
+Ctrl
+DMOG
+55.9
+
+e.
+
+Ctrl
+R5020
+51.5
+
+f.
+
+Ctrl
+DMOG
+40.8
+
+g.
+
+Ctrl
+DMOG
+
+h.
+
+Ctrl
+R5020
+
+i.
+
+Ctrl
+DMOG
+
+j.
+
+Tomato
+DMSO
+DMOG
+
+k.
+
+Tomato
+E1OH
+R5020
+E2
+DMOG
+
+l.
+
+Tomato
+DMSO
+DOX/DMOG
+DOX/DMSO
+DMOG
+Below we address the reviewers comments on the sensitivity of the dFLASH reporter system:
+
+We set out to design a sensitive dual fluorescent reporter that allows use in high-throughput screening via FACS or by high content image segmentation and quantification. We aimed to generate systems which retained endogenous signalling mechanisms but provided robust and consistent signal-to-noise upon activation, as this is a fundamental requirement of screen performance. This remains a challenge however, and is currently being addressed by alternative approaches (see Lampson et al Cell 2024; https://doi.org/10.1016/j.cell.2024.03.022 ). To address this, our 12xHRE response element consists of HIF enhancers from LDHA, ENO1 and PGK1, which are well characterised endogenous HIF target genes, ensuring the endogenous signalling mechanisms of HIF are maintained. By repeating response element 12 times, this provided a reporter system with robust signal to noise which is required for screening applications, but maintained the endogenous signalling mechanism and sensitivity. We have changed the text and methods section to more clearly state this. We found in our Gal4 and PRE controlled systems that 5x response elements (also obtained from endogenous target gene enhances for the PRE) were sufficient for superior dFLASH reporter induction, indicating that a large number of response elements is not a requirement for sensitive reporter induction in these contexts.
+
+However, a number of historical (https://doi.org/10.1002/j.1460-2075.1992.tb05607.x and https://doi.org/10.1016/0022-2836(90)90187-Q) and recent (https://doi.org/10.1016/j.cell.2012.12.027 and https://doi.org/10.1101/2024.02.02.578660) papers demonstrate that gene activation does not scale linearly with the number of enhancer elements increasing sharply and plateauing with up to ~7-8 repeats. This indicates that while synergism is necessary for robust reporting this is unlikely to be increase beyond 7-8 repeats. We also note that on average endogenous enhancers contain 5-6 binding site for TFs, (https://doi.org/10.1038/s41586-020-2528-x) indicating that designs implemented here are inline with those of native enhancers.
+
+The time taken to active the HIF reporter seems slow, at around 10 hrs. HIF should be stabilised within minutes and HIF targets should increase at the message level within a few hours. Why is this induction slower? This may be a concern for drug screening.
+
+The reviewer is correct to outline that the reporter induction begins at ~10hrs post activation. However, while HIF is stabilised within minutes to hours(https://doi.org/10.1016/i.ibc.2023.104599 and http://doi.org/10.1126/science.1059796 ), reporter output is a more complex function of the time taken for accumulation of nuclear HIF, DNA binding and transcriptional activation dynamics as well as the Tomato protein translation and folding dynamics. Indeed, peak HIF protein is often observed 4-6hrs post induction and this has been shown to be cell type dependent with altered induction dynamics.
+
+As many HIF target gene proteins do not appear until ~8-16hrs post induction¹ we view a 10hr lag prior to population Tomato fluorescence is in line with native HIF
+target gene dynamics. However, it is noted that as seen in Supp Video 2 the induction of the monoclonal reporter lines in single cells is heterogenous and likely reflects stochastic mechanisms of gene activation, this is averaged in population dynamics in Figure 3.C.
+
+None of the dFLASH-HRE experiments seem to have been done in different oxygen gradients. Are similar effects seen? The use of DMOG will inhibit both PHDs and FIH but do specific PHD inhibitors activate the reporter similarly?
+
+We have now added new data with the experiments suggested by Reviewer #1. We show similar levels of induction under hypoxic conditions (~1%) (Figure 3f) and using the Prolyl hydroxylase (PHD) specific inhibitor FG-4592 ( Supplementary Figure 6c) we see a difference in activity between DMOG (a pan-2-oxoglutarate dependent dioxygenase inhibitor) and PHD-specific FG-4592. Both upregulate the reporter, however FG-4592 does so less than DMOG. This has been added to the result text on Page 9, Paragraph 2, Line 584-587.
+
+-Regarding the drug screens, while there is good concordance with compounds that activate, there is more variability in the inhibitor screens between 24 and 36 hours. As there are already clinically useful drugs to activate HIFs, the challenge (as stated in the manuscript) is to find specific inhibitors. This variability may relate to the stability of Tomato and GFP, which is different from HIF. Therefore, how dynamic is this system? E.g. how long does it take for Tomato levels to return to baseline once DMOG is washed out?
+
+The bimodal screening methodology was primarily to outline the utility of live-cell drug screens in assaying multiple time points. We acknowledge that the consistency may be improved by single time-point screens. In saying this, the in-screen Z’ remained >0.5 (mean of 0.62 for the DMOG 36-hour screen and 0.61 for the DMOG 24-hour screen across all 20 plates), indicating that the assay was highly consistent at the later time points. Additionally, independent plate controls have now been added to Supp Figure 8.
+
+The dFLASH system was designed such that Tomato is not regulated by O2 dependent proteasomal degradation as the native HIF system is, allowing relatively consistent quantitation of the transcriptional output. The however GFP contains a PEST element such that is has a protein turnover of ~ 2hrs (also discussed below).
+
+-The dFLASH-PRE seems a more sensitive reporter system that dFLASH-HRE. Accept it is likely beyond the scope of this work but it would be helpful to have some indication from the authors as to the utility of this reporter in chemical screens.
+
+Yes, this is indeed an exciting prospect for identification of non-hormonal drugs that might target PGR. To demonstrate the utility of this reporter we have since performed whole genome CRISPR screens identifying (>261 genes with a p-value < 0.001) new regulators of the Progesterone response pathway. We believe this outlines the high-quality of the dFLASH system in a variety of screening approaches,
+but is indeed beyond the scope of this paper. We provide some data from these screens to validate their use to the reviewer but not within the paper. Indeed, we identify PGR as the top hit from this dFLASH loss of function CRISPR screen as well as other known (NCoA2) and many novel regulators.
+
+[REDACTED]
+
+How was the drug toxicity assessed in the small molecule screens? Were the cells stained for cell death/apoptosis, or was it judged by GFP signal/cell morphology? GFP alone may not be a good indication of toxicity as it is likely to be very stable.
+
+The EGFP expressed within dFLASH is destabilised (contains a PEST element) and therefore protein turn-over occurs every ~2 hours. Due to this rapid turn-over, we believe the PEST-EGFP used within dFLASH does provide a better readout of toxicity (i.e. identification of a decrease in expression), compared to if standard EGFP is used, and therefore we use this change in EGFP as a read out of toxicity in the dFLASH system. However, we do agree with the reviewer that thorough assessment of toxicity in high throughput screens is an important consideration, and we are currently considering alternative approaches for future works.
+Reviewer #2 (Remarks to the Author):
+
+The work by Allen et al. presents dual FLuorescent transcription factor Activity Sensor for Histone integrated live-cell reporting (dFLASH), a modular sensor for TF activity that can be readily integrated into cellular genomes. The authors demonstrate that dFLASH platforms can sense the regulation of endogenous Hypoxia Inducible Factor (HIF) and Progesterone receptor (PGR) activities, as well as regulated coactivator recruitment to a synthetic DNA-Binding Domain-Activator Domain fusion proteins. Finally, the authors test the utility of this platform for functional genomics applications by using CRISPRoff to modulate the HIF regulatory pathway, and for drug screening by using high content imaging in a bimodal design to isolate activators and inhibitors of the HIF pathway.
+The manuscript is well written, the experiments are very thoroughly executed and the proposed tool is certainly of great interest for the field. However, the impact of the study is likely to increase by addressing the following questions experimentally:
+
+Major:
+
+Did the authors address the degree of cell death and apoptosis after integration of dFLASH constructs? In this regard, it would also be important to test the dFLASH in non-transformed, non-immortalized and non-cancerous cells. Would the dFLASH tool work in primary cells?
+
+Cell lines were integrated with dFLASH reporters by lentivirus transduction at a low MOI (<0.3) to ensure that each cell contained ~1 integrant per cell as described in the methods. After hygromycin selection all most all of the GFP +ve cells survived. We did not directly assess the degree of cell death and apoptosis after integration; as expected cell death was observed after transduction and subsequent selection due to death of non-transduced cells. However, once a stable line was established, minimal subsequent cell death is observed and cells tolerate the dFLASH reporter well for long periods of culture.
+
+We have subsequently tested a number of non-transformed or primary cells lines with the dFLASH-HRE system and demonstrate significant FG-4592 dependent induction of the reporter in primary mouse granulosa cells. This has been included in Supp Figure3 and discussed on page # 7 paragraph 2, lines 362-370. However, we acknowledge that substantial challenges remain for the application of the dFLASH system in primary / non-immortalised cells. In some contexts that we tested, such as human embryonic stem cells, one, or a combination of the constitutive control EGFP promoter, the destabilised EGFP and viral construct silencing made it difficult to implement the dFLASH reporter. Thus, it is likely that implementation of dFLASH across primary and non-immortalised cells will to require cell-type specific optimisation (such as changing the control promoter or using non-destabilised EGFP), which we are currently attempting in future efforts.
+
+Especially with regard to the studies on the HIF pathway, it would be important to correlate the reporter kinetics after DMOG treatment shown in figure 2 with HIF protein levels and eventually mRNA expression of (a) HIF target gene(s). Along these lines and despite the potential issues with fluorescent reporters in low
+oxygen conditions, how would the dFLASH reporter signal look like in a more physiological hypoxia/reoxygenation experiments
+
+For discussion of dFLASH response to changes in oxygen levels, please see comments to review 1 above.
+
+While O2 (or DMOG) withdrawal may be useful for investigation of cycling O2 dynamics, unlike HIF protein, the Tomato protein is more stable in normoxia or without DMOG. As such the decrease in reporter activity is likely to be a function of transcript levels and protein turnover for the reporter rather than HIF. The high stability of the reporter is advantageous to screening as it allows assaying of reporter output over extended periods after removal of the ligand/stimulation or in processing cells. As addressed in response to Reviewer #1 we have performed experiments in <1% O2, demonstrating (Figure 3f) similar levels of induction to that of DMOG. We have also discussed above the dynamics of the reporter in comparison to native target gene inductions. We have not performed reoxygenation experiments as this would only provide information of the stability of the Tomato mRNA and protein and not inform on specific or generic HIF target genes/proteins.
+
+Minor:
+
+The figure organization could benefit from moving some of the data from the suppl. figures to the main figures. Lots of data that are required to understand the manuscript are shown in the suppl. figures while the main figures are rather spares in panels.
+
+We agree that space limitations and the scale of the paper makes some of the described text difficult to follow due to substantial supplementary data. We have attempted to reorganise the figures to move some supplementary material to the main figures (especially in Figure 3). We hope that this will better allow the reviewer and the reader the access to the information needed to understand and interpret the main findings of the paper.
+
+This includes;
+
+FACS 2D GFP/Tomato dotplots moved from Supp to Figure 2.
+
+Low O2 mcdFLASH-HRE induction, mcdFLASH-PGR dose response curves and drug specificity moved to figure 3 from Supp.
+
+Reviewer #3 (Remarks to the Author):
+
+dFLASH; Dual Fluorescent transcription factor Activity Sensor for Histone integrated live-cell reporting and high-content screening
+
+Comments to the authors
+
+Allen et al report the development of dFLASH platform that is a modular transcription factor sensor that can be used in live cell reporting for multiple downstream
+applications such as functional genomics screens, drug discovery and synthetic biology. This system consists of two dual color reporters (an internal control nucEGFP as well as dTomato reporter as a readout). The authors demonstrate the use of this reporter system for readout of endogenous HIF and PGR related pathways. Together these authors developed a system that can be used for a variety of applications. Major strengths of this system are that it can be applied to multiple cell types and the modular design also allows researchers to study various endogenous processes. The dual reporter system also allows readouts through flow cytometry or imaging. Furthermore, they successfully used this system to identify novel inhibitors and activators of the HIF pathway in screens. However, it would useful if the authors could perhaps address more strongly, the ways in which their system is advantageous or different from previous reporter systems. Alternatively, the authors could provide a comparison of previous reporter systems to better communicate the impact or uniqueness of their system. Second, adding more quantifications to some of the imaging and flow cytometry data could also strengthen the results. Overall the data presented in this manuscript is convincing and the applications of the reporter system are strongly demonstrated. Therefore, it could provide an amenable system for reporter-based screening applications to study other endogenous pathways as well.
+
+We thank the reviewer for the comments and have added some additional quantitation to the flow cytometry experiments in each figure including the % of cells Tomato Induced cells (Figure 2, Figure 3, Supp Figure 5, and Supp Figure 7). The mean fold inductions of populations from imaging or FACS experiment is also described in the monoclonal dFLASH cell lines in Figure 3. We hope that this improves the presentation of the results.
+
+Additional comments:
+1. Within the introductions and discussions, the authors could discuss previous systems that were built for TF sensing, for example luciferase or reporter systems have been used in the past to enable HF1a pathway analysis (See a). Discussion of these systems alongside the one that the authors have built would be useful in communicating the impact of the developed reporter system. Additionally, the authors could also determine how their system addresses some of the shortcomings of the previous systems and the types of pathways that can be analyzed by the system in order to make the discussion more comprehensive.
+a. An optimized reporter of the transcription factor hypoxia-factor inducible factor 1a reveals complex HIF-1a activation dynamics in single cells https://www.jbc.org/article/S0021-9258/23/00241-7/fulltext
+
+We believe this is addressed in response to Reviewer #1’s comments. However, we specifically address the comments regarding the HIF reporters. Reviewer #3 is correct to point out that there are some other approaches taken to report on ‘HIF activity’ (https://doi.org/10.1016/j.cmet.2016.09.015 and https://doi.org/10.1016/j.jbc.2023.104599). We have attached a table to compare and contrast some of these systems. However, the protein stability reporter mentioned as well as the other referenced HIF reporters do not specifically report of HIF enhancer activity. The HIF1a reporter in the recent JBC article only reports on O2 dependent HIF protein abundance (not transcriptional activity). Additionally, the latter paper contains a complex hypoxic enhancer + constitutive (SV40 promoter) driven O2
+destabilised reporter which could decouple O2 dependent enhancer vs post-translational regulation, potentially complicating outputs. In addition, neither are internally controlled at the same genomic loci. Thus, given that the dFLASH system reports specifically on HIF enhancer activity together with the endogenous oxygen-regulation of the HIF transcription factors, and has an internal control, we believe it is thus more reflective of endogenous HIF signalling and better controlled than already available systems.
+
+• See discussion above
+
+2. In Figure 1B, the authors are showing how their built reporter system can be used for imaging-based analysis. This figure can be strengthened if they include the pixel intensities of the GFP and RFP channels from select regions of the microscopy images. More specifically it would be helpful to see that the RFP increases and the GFP remains constant. It would also help account for the background fluorescence of the Tomato reporter in using the imaging application. Finally, they could also consider adding black/white image panels.
+
+Quantitation of the nuclear Tomato/GFP intensities’ are reported in various experiments within the paper as such we have not included mean pixel intensities from images in Figure 1. However, to address reviewers #1 and #3request for individual Tomato and GFP measurements over time we have included individual Tomato and GFP time course data associated with Figure 3 in supplementary Figure 4. We have modified Figure 1 to black and white image panels as requested .
+
+3. The schematic in Figure 1B is a little misleading, can the authors include more details or enhance the figure to guide the reader? For example, add +DMOG etc. or potentially remove the arrow
+
+We have modified Figure 1 in an attempt to more clearly outline the schematic of the dFLASH system. We hope that this is more useful to the reader.
+
+![Schematic diagram of the dFLASH system, including gene constructs, microscopy images, and flow cytometry plots](page_682_1097_682_312.png)
+4. Within the flow plots shown in Figure 2, D-F it would help if the authors quantified the percentage of cells that respond to the given treatments. For example, they can show their gating strategies. They should also comment on the fact that these are not clonal lines and are a heterogeneous population, since they are following up on these experiments with single cells clones.
+
+We have now included quantification for polyclonal populations in dFLASH reporters in Figure 2 and an example of the gating strategy in Extended Data Figure 1. In addition, gating strategy is explained in the Methods section.
+
+    • Quantification and Gating strategy and % cells in FACS plots.
+a.
+
+endogenous
+HIF-1α
+Fe^{2+}
+2-OG O_2
+PHD/FH
+Hypoxia
+DMOG
+ARNT
+
+CBP/p300
+pMin
+Tomato
+EGFP
+
+b.
+
+Progesterone (P) P-like ligands
+FKBP52
+PR
+HSP90
+PR
+CoAc
+pMin
+Tomato
+EGFP
+
+c.
+
+Dox
+rTA
+TRE3G
+GAL4DBD-HIFCAD
+HIFCAD
+GAL4BD
+pMin
+PGK/CMV
+Tomato
+EGFP
+
+d.
+
+Ctrl
+DMOG
+55.9
+
+e.
+
+Ctrl
+R5020
+51.5
+
+f.
+
+Ctrl
+DMOG
+40.8
+
+g.
+
+Ctrl
+DMOG
+
+h.
+
+Ctrl
+R5020
+
+i.
+
+Ctrl
+DMOG
+
+j.
+
+Tomato
+DMSO
+DMOG
+
+k.
+
+Tomato
+EIOH
+R5020
+E2
+DMOG
+
+l.
+
+Tomato
+DMSO
+DOX/DMOG
+DOX/DMSO
+DMOG
+a.
+
+FSC-H/FSC-W    SSC-H/SSC-W    SSC-A/FSC-A    Tomato/GFP
+
+b.
+
+FSC-H/FSC-W    SSC-H/SSC-W    SSC-A/FSC-A    Tomato/GFP
+
+Extended Data 1. Representative plots for gating strategies
+Representative plots of the gating strategy (see Methods) for vehicle (0.1% DMSO, Top row panels) and ligand treated (bottom row) monoclonal (a) HEK293T mcdFLASH-HRE cells and (b) T47D mcdFLASH-PRE cells. Populations are from Figure 3.
+5. Could the authors explain why some of their cells do not express dTomato despite antibiotic selection? A two-axis flow plot with gating may be more comprehensive for the main figure (like those shown in Supp Fig2). For example, their flow plots can indicate a range of fluorescence intensities in the tomato channel, how would these numbers change if the authors gated for the highest GFP expressing cells (where the reporter is integrated in euchromatin region). Would that in turn affect the Tomato readout?
+
+In polyclonal pools the reporter is integrated primarily into gene bodies due to the Lentiviral integrase interaction with LEDGF and recruitment to H3K36me. While these are generally present at actively transcribed genes, the integration site and thus loci dependent effects on enhancer activity may result. In addition, and more likely, stochastic methylation of the enhancer element and not the constitutive promoter may result in differences in the tomato induction. We view this as more likely given that we have observed that polyclonal cell lines that are selected for a short period of time display more homogenous tomato induction (see Supp Figure 1h). Gating based on the highest GFP does not affect the Tomato induction. This can be visualised in the dot plot flow cytometry of GFP and Tomato in the polyclonal pools. As such we have chosen not to include this in the main or supplementary data of the paper.
+
+![Flow cytometry plots showing GFP and Tomato expression in Ctrl and DMOG conditions](page_489_728_484_246.png)
+
+6. Figure 4C/Supplementary Fig 6: The authors can include a single cell flow plot and gating information in the figure to make it more comprehensive for the reader (including numerical/gating details e.g. % of cells
+
+We have now included the % induced Tomato and gating as part of Figure 2, Figure 3/Supplementary Fig 4 and Figure 4/Supplementary Fig 6 (Now Supplementary Figure 7). The main text has been updated to reflect this.
+
+7. The authors could clarify the reasoning behind using different promoter versions for CRISPRoffv2.1
+SFFVp and EF1a are common choices for constitutive expression of transgenes, SFFV is more compact than the full EF1a promoter but can be more prone to silencing in some cell lines (https://doi.org/10.1016/j.cels.2022.11.005). We chose to clone and validate using both of these CRISPRoffv2.1 expressing viruses to provide alternate options for researchers, we have made both of these plasmids are available on addgene #207180 and #207181.
+
+8. It appears that the clonal cell lines in Figure 4 produce multiple dTomato peaks, does that indicate that the line is slightly leaky for expression of dTomato? How does it compare to no fluorescence - doing an overlay would be helpful? Could potentially be a challenge for use in screens if there is indeed background.
+
+Figure 4 depicts clonal HEK293T-mcdFLASH-HRE cells which have been infected with lentivirus expressing a guide RNA targeting the promoter of VHL and selected with puromycin and subsequently the CRISPRoffv2.1 lentivirus and selection with blasticidin prior to Flow cytometry. The resultant is a polyclonal cell line expressing the guide RNA and the CRISPRoffv2.1 in the background of the clonal reporter line. This explains the heterogeneous induction of Tomato (albeit in ~60% of cells by 10 days post CRISPRoffv2.1). This is in line with the most successful implementations of CRISPRoffv2.1 in Nunez et al 2021 (https://doi.org/10.1016/j.cell.2021.03.025) resulting in 60-80% of cells silencing targeted cell surface genes.
+
+As it is unclear where the reviewer is indeed commenting on the background of unstimulated mcdFLASH-HRE in Supplementary Figure 6, we will also address that here. We find that the reporter displays modest background in the absence of DMOG or O2 in some cell lines, although we have found this is much less pronounced in U2OS and Hela cell lines that we have since generated (data not shown). As previously reported for the HIF pathway (https://doi.org/10.1002/jcp.21584 and https://doi.org/10.1038/sj.onc.1204972) higher cell density can lead to non-hypoxic stabilisation of the hypoxic inducible factors and this may also play a role in background basal expression of Tomato in certain experiments. We do not expect this to effect the ability to screen for HIF activity using the dFLASH systems.
+
+9. In the CRISPRoff experiment, did the authors also include a non-targeting guide? Might be useful to show or include as control
+
+As PGR is not expressed in HEK293T cells and has not been described to effect the HIF pathway, we have used a sgRNA targeting the promoter of the PGR gene as a negative control. These experiments demonstrate that guide RNAs targeting the VHL promoter specifically induce the expression of the HEK293T-mcdFLASH-HRE reporter. We have altered the main text to reflect this.
+e.
+
+![Flow cytometry plots showing Tomato induced GFP and VHL knockdown](page_184_120_579_232.png)
+
+10. Not essential but the authors can confirm VHL protein knockdown in cells post CRISPRoff induced silencing using a western blot (shows sensitivity of the system).
+
+VHL is a very well characterised Ub-ligase that is reported to specifically target and degrade HIF1a and HIF2a in normoxia. This is mediated by hydroxylation of two proline residues in the Oxygen dependent degradation domains. Given that it is well-accepted that knockout or mutation of VHL results in stabilisation of HIFα protein and transcriptional output we believe that activation of the mcdFLASH reporter upon CRISPRoffv2.1 knockdown of VHL is sufficient evidence. As such we have not attempted to perform western blots for VHL.
+
+11. Discuss how this reporter system can be of better use compared to current systems for screening methods
+
+Emerging applications in high-throughput pooled or arrayed CRISPR or Drug screens require highly sensitive, consistent, selectable reporters. The dFLASH system provides a modular system that unlike other systems allow single cell normalised quantification of transcriptional outputs in live cells. This allows use in monitoring transcriptional dynamics in single cells, pooled or arrayed CRISPR screens easily engineerable to a variety of cells line and transcription factor response pathways. Unlike luciferase systems, dFLASH allows FACS selectable pools which is commonplace in CRISPR screening. We have rigorously validated the applicability to high throughput screening and describe modular use in specifically reporting on functionally distinct transcription factor pathways.
+
+12. For Figure 5, the authors could describe the Activation Screen and Inhibitor with more detail or clarity within the text such that it is easy to follow, or describe the workflow and expected results with more clarity.
+
+We agree with this feedback and as a result, this section has been reworded and the hit classification has been modified for increased clarity on describing how the dFLASH outputs can be utilised in a HTS manner without substantively changing the conclusions put forward in the earlier version of the manuscript. The workflow is now described with more clarity on Page 9-10, Line 578-625). We have also added a paragraph to the results discussion which now describes how the screen cut offs are
+described, with more clarity and uniformity between time points and agonist/antagonist screening. In addition, all relevant screen details are included in Supplementary Table 5 for quick reference to workflow, results and directs the reader to the appropriate methodology as required (attached below).
+
+<table>
+  <tr>
+    <th>Category</th>
+    <th>Parameter</th>
+    <th>Description</th>
+  </tr>
+  <tr>
+    <td rowspan="6">Assay</td>
+    <td>Type of assay</td>
+    <td>High Content. Live cell screen of a genetic reporter measuring nuclear TF-dependent Tomato expression and constitutive EGFP expression</td>
+  </tr>
+  <tr>
+    <td>Target</td>
+    <td>HIF-1α-dependent Tomato expression</td>
+  </tr>
+  <tr>
+    <td>Primary measurement</td>
+    <td>Ratio of Tomato expression to EGFP expression.</td>
+  </tr>
+  <tr>
+    <td>Key reagents</td>
+    <td>HEK293T mcFLASH-HIF reporter cells and 1mM DMOG in DMSO.</td>
+  </tr>
+  <tr>
+    <td>Assay protocol</td>
+    <td>Assay protocol can be found in detail in the methods in the sections “High Content Imaging (HCI)” and “Bimodal small molecule screen to identify activators or inhibitors of the hypoxic response pathway.”</td>
+  </tr>
+  <tr>
+    <td colspan="2">Additional comments</td>
+    <td><b>Figure 1</b> provides a schematic for the genetic reporter.</td>
+  </tr>
+  <tr>
+    <td rowspan="4">Library</td>
+    <td>Library size</td>
+    <td>1595 compounds supplied at 5mM in 1uL of DMSO that were dried onto the plates.</td>
+  </tr>
+  <tr>
+    <td>Library composition</td>
+    <td>Library was a mixture of synthetic and natural product compounds curated by Prof. Ronald Quinn.</td>
+  </tr>
+  <tr>
+    <td>Source</td>
+    <td>Compounds were sourced from Compounds Australia (<a href="www.compoundsaustralia.com">www.compoundsaustralia.com</a>)</td>
+  </tr>
+  <tr>
+    <td colspan="2">Additional comments</td>
+    <td></td>
+  </tr>
+  <tr>
+    <td rowspan="8">Screen</td>
+    <td>Format</td>
+    <td>96-well plates. Plates supplied were 20 96-well Costar CLS3603 black plates with µclear bottoms.</td>
+  </tr>
+  <tr>
+    <td>Concentration(s) tested</td>
+    <td>For Activation screening, 50µM for each compound was investigated.</td>
+  </tr>
+  <tr>
+    <td>Plate controls</td>
+    <td>For Inhibition screening 25µM was investigated.</td>
+  </tr>
+  <tr>
+    <td>Reagent/ compound dispensing system</td>
+    <td>At each timepoint, compound-free well with equivalent 0.1% DMSO (negative control) and 1mM DMOG (positive control) were included.</td>
+  </tr>
+  <tr>
+    <td>Detection instrument and software</td>
+    <td>Compounds were dispensed into the 96 well format by Compounds Australia. Cells and DMSO or DMOG were added manually. Thermofisher ArrayScan XTi was the imagining instrument.</td>
+  </tr>
+  <tr>
+    <td>Assay validation/QC</td>
+    <td>HCS Studio 3.0 was the primary analysis software for detection and quantification of nuclear fluorescence.</td>
+  </tr>
+  <tr>
+    <td>Correction factors</td>
+    <td>Z' > 0.5 for each screen were confirmed, as was ensuring hits met >2SD parameter for Tomato/GFP in <b>Supplementary Figure 8</b>.</td>
+  </tr>
+  <tr>
+    <td>Normalization</td>
+    <td>N/A</td>
+  </tr>
+  <tr>
+    <td colspan="2">Additional comments</td>
+    <td>Data was Z score normalized.</td>
+  </tr>
+  <tr>
+    <td rowspan="7">Post-HTS analysis</td>
+    <td>Hit criteria</td>
+    <td>Hit criteria is described in method section “Bimodal small molecule screen to identify activators or inhibitors of the hypoxic response pathway”. For activator screens hits had to be >2SD for Tomato/EGFP and >1SD tomato MFI Z score while EGFP expression did not change more than 2SD relative to mean of the compound treated population. For inhibitor this criteria was <2SD for Tomato/EGFP and Tomato MFI, while again EGFP did not change more than 2SD relative to the compound treated population.</td>
+  </tr>
+  <tr>
+    <td>Hit rate</td>
+    <td>36 hour activator screen: 25 compounds (1.4%)<br>24 hour activator screen: 8 compounds (0.5%)<br>Overall activator screen: 3 compound replicated between screens (0.18%)<br>36 hour inhibitor screen: 69 compounds (4.2%)<br>24 hour inhibitor screen: 81 compounds (5.07%)<br>Overall inhibitor screen: 36 compounds replicated between screens (2.25%).</td>
+  </tr>
+  <tr>
+    <td>Additional assay(s)</td>
+    <td>Replicate dFLASH High content assays were done at 24 hours on a subset of hits for activator and inhibitor compounds to confirm their activity (<b>Supplementary Figure 9, 10</b>). Compounds were reordered through Compounds Australia (<a href="www.compoundsaustralia.com">www.compoundsaustralia.com</a>) to confirm identity prior to re-assay and downstream investigations.</td>
+  </tr>
+  <tr>
+    <td>Confirmation of hit purity and structure</td>
+    <td></td>
+  </tr>
+  <tr>
+    <td colspan="2">Additional comments</td>
+    <td></td>
+  </tr>
+</table>
+13. For Figure 5B/C, the authors could label the top hits from their screen in the plot\(
+14. For Figure 5B, the squares, circles and triangles can be a little difficult to distinguish, the authors could consider using different colors for clarity.\)
+
+We agree with the reviewer that these figure panels are difficult to interpret with the current layout. We have therefore aligned all of the data by compound such that it is consistent for each screen and labelled a small number of hits, We have also altered the colouring and symbols used to make it easier to distinguish different compound classifications. Additionally, we added some clarity in the methods as to how theses hits were classified (page 10, paragraph 1 line 606-616, Supplementary Table 5W).
+
+1 Bartoszewski, R. et al. Primary endothelial cell-specific regulation of hypoxia-inducible factor (HIF)-1 and HIF-2 and their target gene expression profiles during hypoxia. FASEB J 33, 7929-7941 (2019). https://doi.org/10.1096/fi.201802650RR
+
+Other changes
+
+Minor grammatical and text changes throughout the manuscript
+
+Figure 3
+
+Z' and fold change calculation corrections
+
+Fold Change = 15.33393
+
+Z' = 0.6241246
+Response the Round 2 Review
+
+Reviewer #1 (Remarks to the Author):
+
+Overall, the authors have not adequately addressed the issues raised during the initial submission, and the manuscript has not been significantly improved. There does not seem to be a much of an advance here.
+
+- Novelty of the HRE reporter. The authors present a table comparing some prior HIF reporters, which is selective and not accurate. Other HRE-fluorescent reporters exist and have been described over the past 20 years (e.g. PMID 11774035, 17270179, 26598532). Some have dual colors, destabilised fluorescent proteins (PEST or adding the ODDD), and have been shown to work in cell lines and in vivo settings. The HRE-HIF-ODD reporter is not constitutively active (minimal SV40 promoter x3HRE). The addition of the internal fluorescent control remains the only novelty of the reporter, but the published NanoFIRE manuscript now referred to in the authors' response (describing a similar approach to dFLASH but with luciferase) already details the use of the 12XHRE dFLASH fluorescent system.
+
+- The authors state that endogenous enhancers contain 5-6 sites for TFs, but still use 12 HREs. This necessity for 12x HRE is at odds with other prior reporters (e.g. HypoxCR).
+
+- The variability in the inhibitory screens remains a concern. The stability of Tomato is not an advantage as stochastic/basal induction of the reporter will occur (as noted), and dFLASH will not detect turning a TF off. Same argument applies for reoxygenation.
+
+- The assays of cell death or drug toxicity have not been adequately addressed.
+
+- The HRE screen has so far not identified any interesting results.
+
+- The utility of the system in other cell types (SFig. 3D) is not that impressive, with a fairly minimal induction.
+
+Reviewer #2 (Remarks to the Author):
+
+The authors have addressed all concerns in the revised manuscript. Congratulations!
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have successfully addressed all the questions and made modifications to the manuscript accordingly.
+
+We submit that our initial reviewer response addressed all of the reviewers' comments with additional experiments and extensive clarification, noting satisfaction from two of the three reviewers. However, reviewer 1 subsequently raised new criticisms that we were not given the opportunity to address, while re-iterated criticisms seem to result from misinterpretations of some of the data and points
+made in the manuscript. Here we seek to clarify concerns of reviewer #1 with this rebuttal and new data demonstrating the performance of the dFLASH system in pooled high-throughput CRISPR screens in two separate cell lines.
+
+We presented what reviewer 1 described as a selective comparison of dFLASH compared to existing platforms because like-for-like empirical comparisons were not possible, largely because in built capabilities in dFLASH are not included in the other systems. This we tried to clearly illustrate in the table which we maintain is comprehensive and accurate. Additionally, reviewer #1 provides no evidence of inaccuracies, which we detail below.
+
+We presented a versatile system, demonstrated by analysing four transcriptional pathways with some associated screening applications. Reviewer 1 focused on nuances of the HIF pathway alone. We disagree that dFLASH does not substantially improve upon previous screening systems as outlined below. We also now provide new data in the manuscript in support of the improved screening potential of the dFLASH system. (Figure R2 or new Figure 5 of the manuscript, Page 9 paragraph 3, line 400-423, Page 10 paragraph 10, line 432-448, Page 22 paragraph 1, line 862-882)
+
+Novelty of the HRE reporter. The authors present a table comparing some prior HIF reporters, which is selective and not accurate. Other HRE-fluorescent reporters exist and have been described over the past 20 years (e.g. PMID 11774035, 17270179, 26598532). Some have dual colors, destabilised fluorescent proteins (PEST or adding the ODDD), and have been shown to work in cell lines and in vivo settings.
+
+We regard our table as complete and accurate and our dFLASH system outperforms the other systems highlighted by reviewer 1 (PMID 11774035, 17270179, 26598532). The improvements in our system include:
+a) incorporating internal controls and selection cassettes
+b) producing nuclear fluorescent proteins that we demonstrate allows automated nuclear segmentation and quantification of transcriptional responses, a feature absent from PMID 17270179 and 1506559
+c) being available to other researchers through Addgene and available for benchmarking.
+d) Unlike the dFLASH reporter system, no other reporter system cited by any reviewer has been shown to function robustly for multiple orthogonal TF response pathways in multiple cell types.
+
+• Importantly, we are unable to find evidence in the literature that there are other internally controlled reporters, conflicting with the statement by reviewer #1. The only HIF reporter with a second readout is detailed in our comparison of reporter systems in the supplementary table and is linked to the cell cycle (PMC4151727). We maintain that this supplementary table comparison of fluorescent reporter systems accurately depicts the advances of our novel, flexible and valuable features of dFLASH.
+
+The HRE-HIF-ODD reporter is not constitutively active (minimal SV40 promoter x3HRE).
+Contrary to reviewer 1’s claim, HIF reporters containing a 3xHRE-SV40 promoter driven oxygen dependent degraded mCherry^{ODD} (ODD-mCherry) as used in Ortman et al Nature Genetics 2019 CRISPR screen, has very recently (Posted September 30, 2024.) been shown to produce constitutive activity and poor Hypoxic response element driven transcriptional responses (
+https://doi.org/10.1101/2024.09.28.615614 , reproduced below as Figure R1). Note that minimal promoter and response elements adopted in this preprint closely resemble the approach that we have taken with dFLASH. We use an optimised minimal promoter with low background which we have now more clearly outlined in the methods and manuscript (Page 4; line 117, Page 45 line 1198). We hope this more clearly explains the high performance of the dFLASH system in signal induced enhancer activation. This is now evident in the new U2OS-dFLASH-HIF cell line data we have now included which demonstrates a >400 fold induction of Tomato (see below and new Figure 5 of the manuscript, Page 9 paragraph 3, line 400-423, Page 10 paragraph 10, line 432-448, Page 22 paragraph 1, line 862-882, ).
+
+[REDACTED]
+
+We have also now additionally benchmarked the performance of the minimal SV40 promoter 3xHRE vs the dFLASH-HRE (2 cell lines) in whole genome CRISPR screening. We demonstrate that the dFLASH system identifies ~>10-15x more hits (Figure R2., p<0.05, and new Figure 5 of the manuscript, Page 9 paragraph 3, line 400-423, Page 10 paragraph 10, line 432-448, Page 22 paragraph 1, line 862-882, Page 47-49; line 1329-1440) in two separate cell types, including known obligate regulators as top hits (HIF1a and ARNT). In order to benchmark the performance of the dFLASH system in pooled CRISPR screens and compare to orthogonal systems, we have included a new figure into the manuscript demonstrating this (new Figure 5 of the manuscript, Page 9 paragraph 3, line 400-423, Page 10 paragraph 10, line 432-448, Page 22 paragraph 1, line 862-882, Page 47-49; line 1329-1440). This not only addresses over the request for benchmarking of the dFLASH system and the questioned suitability to high throughput functional genomics screens, but also the ability to sensitively detect turning off transcription factor activity, as outlined by reviewer #1 below.
+Figure R2. The number of enriched “hit” genes (p<0.01) from HIF1a whole genome wide CRISPR screens from 3xHRE-sv40- mCherryODD vs dFLASH-HRE systems. (this is now new Figure 5 in manuscript, Page 9 paragraph 3, line 400-423, Page 10 paragraph 10, line 432-448, Page 22 paragraph 1, line 862-882, Page 47-49; line 1329-1440)
+
+The addition of the internal fluorescent control remains the only novelty of the reporter, but the published NanoFIRE manuscript now referred to in the authors’ response (describing a similar approach to dFLASH but with luciferase) already details the use of the 12XHRE dFLASH fluorescent system.
+
+Our NanoFIRE manuscript (Oct 2023) references the dFLASH system (reference 13 of paper, Biorxiv 2024) and does not detail any of the utilities demonstrated in this paper.
+
+This related publication has no relation to the advances of the dFLASH system or novelty as claimed by reviewer #1, and the methodology and characterisation of the dFLASH system presented here will be of substantial use to researchers attempting to engineer high throughput screening systems.
+
+- The authors state that endogenous enhancers contain 5-6 sites for TFs, but still use 12 HREs. This necessity for 12x HRE is at odds with other prior reporters (e.g. HypoxCR).
+
+High enhancer numbers are not a requirement of the dFLASH system as shown in our other presented reporters (5x GRE and 5xPRE). The literature demonstrates that there is no necessity to use 12 x response elements or any specific number. Our aim was to generate a robust, sensitive and consistent (high Z’ score) readout of HIF transcriptional response pathways, not specifically to recapitulate native hypoxic
+post-transcriptional cycling. We believe the number of response elements was also sufficiently rebutted in the first review with both synthetic and native examples. The use of multiple, well characterized HIF enhancers (VEGF, LHDA, ENO) as described here is one experimental approach that can be modified according to needs
+
+- The variability in the inhibitory screens remains a concern. The stability of Tomato is not an advantage as stochastic/basal induction of the reporter will occur (as noted), and dFLASH will not detect turning a TF off. Same argument applies for reoxygenation.
+
+We disagree the comments made by the reviewer. The use of an internal control reduces false positives in large screening campaigns and is critical to reduce inevitable false positive hits from interventions that alter basal transcription though cytotoxicity, chromatin effects etc. We have now demonstrated that the dFLASH internal control allowed us to isolate 10 novel inhibitors which were confirmed in biological replicates and dose responses. We addressed in the original rebuttal and in the paper and have now shown further data opposing this comment.
+
+We also note that as described in the rebuttal, dFLASH has been successfully used in a >300,000 HTS where it displayed high Z’ values across screening wells (Z’ >0.75) (Figure R3), where the same strategy of GFP normalization and exclusion of false positives has been employed. We also outline that 2 of the dFLASH systems presented here have been independently validated and used for high throughput screening by Australia’s National Drug Discovery Center (NDCC), Walter and Eliza Hall Research Institute (WEHI).
+
+[REDACTED]
+
+Our results indicate that the use of an oxygen destabilized Tomato is not advantageous as reviewer 1 claims. The levels of destabilised tomato reduce rapidly (5-30 mins T_{1/2} as described by numerous groups) upon reoxygenation. This would reduce the signal of a reporter independent of HIF (or ARNT) diminishing the
+consistency in screening. This is the opposite of what the reviewer asserts. dFLASH is well suited to inhibition of TF activity and we have demonstrated this in genetic CRISPR screens. We have now included new data in the manuscript in Figure R2a or Figure 5a of manuscipt, demonstrating that KO of HIF1A in HEK293T or U2OS almost completely blocks reporter induction (also Figure R2, new Figure 5 of the manuscript, Page 9 paragraph 3, line 400-423, Page 10 paragraph 10, line 432-448, Page 22 paragraph 1, line 862-882, Page 47-49; line 1329-1440). We have also demonstrated that KO of the known HIF negative regulator VHL, leads to complete dFLASH activation under normoxia. Stochastic/basal HIF activity is a normal process and very well described in the literature, we see little evidence of this in the dFLASH system unless cells are grown to confluency. We also reiterate from the original rebuttal that the aim of the dFLASH system was not to investigate hypoxic protein dynamics or mimic the native HIF degradation but was to develop a robust and generic TF screening platform which we believe we have demonstrated.
+
+The assays of cell death or drug toxicity have not been adequately addressed.
+
+Destabilised EGFP is present in dFLASH and its loss, as detected by decreased fluorescence, provided a readout of toxicity that was used to identify non-specific effects on transcription and screen out compounds from the chemical library. While this was deemed sufficient for an initial pass during screening, further analyses of nuclear characteristics provided by high content imaging (eg size and shape, possible due to nuclear segmentation provided by dFLASH) are being developed as more subtle indicators of toxicity. We had previously addressed this in our rebuttal and we would like to reiterate that this is not the main focus of the work presented here. In the previously mentioned >300,000 small molecule screen completed at the Australian Drug Discovery Centre, EGFP nuclei count was successfully used to remove non-specific and cytotoxic compounds, highlighting its use as a tool to assess cell death / drug toxicity in a screening setting.
+
+[REDACTED]
+- The HRE screen has so far not identified any interesting results.
+
+The work here describes a general reporter platform and not new HIF biology or blockbuster drugs, but instead proof of principle of uses of the dFLASH system. As described and shown above, we have identified a substantial number of new regulators of the HIF pathway from CRISPR screens which will make up a separate manuscript. We now include in the manuscript (Figure R5c, d) metadata of screens to outline performance of the dFLASH system in pooled high throughput CRISPR screens, demonstrating the propensity to identify new uncharacterised regulators, and compare them to orthogonal screens.
+
+We also find and describe in this paper 2 new chemical activators, one that works as an iron chelator and several indirect inhibitors. While direct inhibitors of HIF were not detected, this work establishes that the system can efficiently detect regulators of HIF in a high throughput format and that it is suitable for large library screens aimed at discovering direct inhibitors. We reiterate that the aim of this work was to develop a modular, Transcription factor cellular screening platform with a number of high throughput screening applications, which we have demonstrated through application of the dFLASH system to both small molecule and genetic based screens.
+
+- The utility of the system in other cell types (SFig. 3D) is not that impressive, with a fairly minimal induction.
+
+We demonstrate the use of the dFLASH system in a number of cell types, where the majority display robust activation. We have since tested the dFLASH-HRE in Hela and U2OS cells (Figure R3) which further confirm this robust activation. We acknowledge that non-immortalized cells remain a challenge, and a variety of issues may affect the output of the reporters including downstream promoter choice, delivery method etc. These issues are not restricted to the dFLASH system and are shared by many others. We feel that this is a minor aspect of the system which was sufficiently experimentally addressed in the rebuttal and to the satisfaction of 2/3 reviewers. We also reinforce that the dFLASH system was designed with screening applications in mind and not primary cell or in vivo applications, which likely require tailored approaches.
+
+Taken together, we believe that this rebuttal has now addressed all concerns maintained by reviewer #1, including all newly outlined concerns.
+
+Kind Regards,
+
+Dr. David Bersten, on behalf of all authors of the manuscript.
+REVIEWERS’ COMMENTS
+
+Reviewer #4 (Remarks to the Author):
+
+I was asked primarily to adjudicate the different views of three reviews of this paper and its revision. In my view, the authors have done an excellent job in addressing the comments, and I feel that the comments of Rev 1 are asking too much. Yes, reporters have been developed before including around hypoxia, but the internal control *does* add a lot of value, as does nuclear localization, and the framework appears to be more broadly useful for reporters in general, which to me seem to often be put out into the world in haphazard and poorly controlled ways. I also don't think it's fair to knock the authors for citing their own preprint. The Table that the authors provide now (which I assume will be included in the paper either as main or supplement) is useful and transparent, and to my knowledge accurate (and I say this as the developer of one of the methods listed there, which absolutely has limitations that are fairly noted there). Both imaging and non-imaging based reporter methods need better controls, and the internal control is a terrific way to go. I might suggest that the authors also add scQers (NatureMethods 2024) to the table, which has a sequence-based internal control that I think reinforces the broader point about how important this is in any context where one is looking at single cells rather than bulk (in this case by imaging, in that case by sequencing).
+
+We thank reviewer #4 for the comments and balanced review. We have now included the creative scQrs approach to reporter normalisation into our comparison table and now include it as a Supplementary Data 1.

056cacada9650bd2ff6e41d24cf77d3f922bae4f244b4a05a96d2273dc50b22f/preprint/preprint.md

@@ -0,0 +1,751 @@
+dFLASH; dual FLuorescent transcription factor Activity Sensor for Histone integrated live-cell reporting and high-content screening
+
+David Bersten
+david.bersten@adelaide.edu.au
+
+The University of Adelaide
+Timothy Allen
+The University of Adelaide   https://orcid.org/0000-0002-8190-2334
+Alison Roennfeldt
+School of Biological Sciences, University of Adelaide
+Moganalaxmi Reckdharajkumar
+School of Biological Sciences, University of Adelaide   https://orcid.org/0000-0002-9136-2810
+Miaomiao Liu
+Griffith University
+Ronald Quinn
+Griffith University   https://orcid.org/0000-0002-4022-2623
+Darryl Russell
+The University of Adelaide
+Daniel J Peet
+Murray Whitelaw
+University of Adelaide
+
+Article
+
+Keywords:
+
+Posted Date: January 5th, 2024
+
+DOI: https://doi.org/10.21203/rs.3.rs-3732294/v1
+
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+Version of Record: A version of this preprint was published at Nature Communications on April 7th, 2025. See the published version at https://doi.org/10.1038/s41467-025-58488-w.
+dFLASH; dual FLuorescent transcription factor Activity Sensor for Histone integrated live-cell reporting and high-content screening
+
+Authors: Timothy P. Allen1, Alison E. Roennfeldt1,2, Moganalaxmi Reckdharajkumar1, Miaomiao Liu3, Ronald J. Quinn3#, Darryl L. Russell2#, Daniel J. Peet1#, Murray L. Whitelaw1,4# & David C. Bersten1,2*
+
+Affiliations: 1. School of Biological Sciences, University of Adelaide, Adelaide, South Australia, Australia 2. Robinson Research Institute, University of Adelaide, South Australia, Australia. 3. Griffith Institute for Drug Discovery, Griffith University, Brisbane, Australia 4. ASEAN Microbiome Nutrition Centre, National Neuroscience Institute, Singapore 169857, Singapore. *Corresponding author. #labs that contributed to the work
+
+Abstract
+Live-cell reporting of regulated transcription factor (TF) activity has a wide variety of applications in synthetic biology, drug discovery, and functional genomics. As a result, there is high value in the generation of versatile, sensitive, robust systems that can function across a range of cell types and be adapted toward diverse TF classes. Here we present the dual FLuorescent transcription factor Activity Sensor for Histone integrated live-cell reporting (dFLASH), a modular sensor for TF activity that can be readily integrated into cellular genomes. We demonstrate readily modified dFLASH platforms that homogenously, robustly, and specifically sense regulation of endogenous Hypoxia Inducible Factor (HIF) and Progesterone receptor (PGR) activities, as well as regulated coactivator recruitment to a synthetic DNA-Binding Domain-Activator Domain fusion proteins. The dual-colour nuclear fluorescence produced normalised dynamic live-cell TF activity sensing with facile generation of high-content screening lines, strong signal:noise ratios and reproducible screening capabilities (Z' = 0.68-0.74). Finally, we demonstrate the utility of this platform for functional genomics applications by using CRISPRoff to modulate the HIF regulatory pathway, and for drug screening by using high content imaging in a bimodal design to isolate activators and inhibitors of the HIF pathway from a ~1600 natural product library.
+Introduction
+Cells integrate biochemical signals in a variety of ways to mediate effector function and alter gene expression. Transcription factors (TF) sit at the heart of cell signalling and gene regulatory networks, linking environment to genetic output\( ^{1,2} \). TF importance is well illustrated by the consequences of their dysregulation within disease, particularly cancer where TFs drive pathogenic genetic programs\( ^{3-5} \). As a result, there is widespread utility in methods to manipulate and track TF activity in basic biology and medical research, predominantly using TF responsive reporters. Recent examples include enhancer activity screening\( ^6 \) by massively parallel reporter assays, discovery and characterisation of transcription effector domains\( ^7,8 \) and CRISPR-based functional genomic screens that use reporter gene readouts to understand transcriptional regulatory networks\( ^2,9 \). Beyond the use in discovery biology TF reporters are increasingly utilised as sensors and actuators in engineered synthetic biology applications such as diagnostics and cellular therapeutics. For example, synthetic circuits that utilise either endogenous or synthetic TF responses have been exploited to engineer cellular biotherapeutics\( ^{10} \). In particular, the synthetic Notch receptor (SynNotch) in which programable extracellular binding elicits synthetic TF signalling to enhance tumour-specific activation of CAR-T cells, overcome cancer immune suppression, or provide precise tumour target specificity \(^{11-14}\).
+
+Fluorescent reporter systems are now commonplace in many studies linking cell signalling to TF function and are particularly useful to study single cell features of gene expression, such as stochastics and heterogeneity\( ^{15} \), or situations where temporal recordings are required. In addition, pooled CRISPR/Cas9 functional genomic screens rely on the ability to select distinct cell pools from a homogenous reporting parent population. Screens to select functional gene regulatory elements or interrogate chromatin context in gene activation also require robust reporting in polyclonal pools\( ^{16} \). Many of the current genetically encoded reporter approaches, by nature of their design, are constrained to particular reporting methods or applications \(^{9,17}\). For example, high content arrayed platforms are often incompatible with flow cytometry readouts and vice versa. As such there is a need to generate modular, broadly applicable platforms for robust homogenous reporting of transcription factor and molecular signalling pathways.\( ^2 \).
+
+Here we address this by generating a versatile, high-performance sensor of signal regulated TFs. We developed a reporter platform, termed the dual FLuorescent TF Activity Sensor for Histone integrated live-cell reporting (dFLASH), that enables lentiviral mediated genomic integration of a TF responsive reporter coupled with an internal control. The well-defined hypoxic and steroid receptor signalling pathways were targeted to demonstrate that the composition of the modular dFLASH cassette is critical to robust enhancer-driven reporting. dFLASH acts as a dynamic sensor of targeted endogenous pathways as well as synthetic TF chimeras in polyclonal pools by temporal high-content imaging and flow cytometry. Routine isolation of homogenously responding reporter lines enabled robust high content image-based screening (\( Z' = 0.68\text{-}0.74 \)) for signal regulation of endogenous and synthetic TFs, as well as demonstrating utility for functional genomic investigations with CRISPRoff. Array-based temporal high content imaging with a hypoxia response element dFLASH successfully identified novel regulators of the hypoxic response pathway, illustrating
+the suitability of dFLASH for arrayed drug screening applications. This shows the dFLASH platform allows for intricate interrogation of signalling pathways and illustrates its value for functional gene discovery, evaluation of regulatory elements or investigations into chemical manipulation of TF regulation.
+Results
+Design of versatile dFLASH, a dual fluorescent, live cell sensor of TF activity
+To fulfil the need for a modifiable fluorescent sensor cassette that can be integrated into chromatin and enable robust live-cell sensing that is adaptable for any nominated TF, applicable to high content imaging (HCI) and selection of single responding cells from polyclonal pools via image segmentation or flow cytometry (Figure 1c) a lentiviral construct with enhancer regulated expression of Tomato, followed by independent, constitutive expression of d2EGFP as both selectable marker and an internal control was constructed (Figure 1a, b). Three nuclear localisation signals (3xNLS) integrated in each fluorescent protein ensured nuclear enrichment to enable single cell identification by nuclear segmentation, with accompanying image-based quantification of normalised reporter outputs using high content image analysis, or single-cell isolation using FACS in a signal dependent or independent manner. The enhancer insertion cassette upstream of the minimal promoter driving Tomato expression is flanked by restriction sites, enabling alternative enhancer cloning (Figure 1a). The sensor response to endogenous signal-regulated TF pathways was first assessed by inserting a Hypoxia Inducible Factor (HIF) enhancer. HIF-1 is the master regulator of cellular adaption to low oxygen tension and has various roles in several diseases18-20. To mediate its transcriptional program, the HIF-1α subunit heterodimerises with Aryl Hydrocarbon Nuclear Translocator (ARNT), forming an active HIF-1 complex. At normoxia4, HIF-1α is post-translationally downregulated through the action of prolyl hydroxylase (PHD) enzymes and the Von Hippel Lindau (VHL) ubiquitin ligase complex21. Additionally, the C-terminal transactivation domain of HIF-1α undergoes asparaginyl hydroxylation mediated by Factor Inhibiting HIF (FIH), which blocks binding of transcription coactivators CBP/p30022. These hydroxylation processes are repressed during low oxygen conditions, enabling rapid accumulation of active HIF-1α. HIF-1α stabilisation at normoxia4 was artificially triggered by treating cells with the hypoxia mimetic dimethyloxalylglycine (DMOG), which inhibits PHDs and FIH, thereby inducing HIF-1α stabilisation, activity and hypoxic gene expression23. The well characterised regulation and disease relevance of HIF-1α made it an ideal TF target for prototype sensor development.
+
+Optimisation of dFLASH sensors
+Initially, we tested FLASH constructs with repeats of hypoxia response element (HRE) containing enhancers (RCGTG)24 from endogenous target genes (HRE-FLASH), controlling expression of either nuclear mono (m) or tandem dimer (td)Tomato and observed no DMOG induced Tomato expression in stable HEK293T cell lines (mnucTomato or tdnucTomato, Supp Figure 1a,b). Given the HIF response element has been validated previously24, the response to HIF-1α was optimised by altering the reporter design, all of which utilised the smaller mnucTomato (vs tdnucTomato) to contain transgene size. We hypothesised that transgene silencing, chromosomal site-specific effects or promoter enhancer coupling/interference may result in poor signal induced reporter activity observed in initial construct designs. As such we optimised the downstream promoter, the reporter composition and incorporated a 3xNLS d2EGFP internal control from the constitutive promoter to monitor chromosomal effects and transgene silencing.
+Dual FLASH (dFLASH) variants incorporated three variations of the downstream promoter (EF1a, PGK and PGK/CMV) driving 3xNLS EGFP (nucEGFP) and 2A peptide linked hygromycin (detailed in Supp Figure 1c) in combination with alternate reporter transgenes that it expressed mnucTomato alone, or mnucTomato-Herpes Simplex Virus Thymidine Kinase (HSVtk)-2A-Neomycin resistance (Neo). Stable HEK293T and HepG2 HRE-dFLASH cells lines with these backbones were generated by lentiviral transduction and hygromycin selection. The reporter efficacy of dFLASH variant cell lines was subsequently monitored by high content imaging 48 hours after DMOG induction (Supp Figure 1d, e). The downstream composite PGK/CMV or PGK promoters, enabled the strong DMOG induced Tomato or Tomato/GFP expression dramatically outperforming EF1a (Figure 1b and Supp Figure 1d). The composite PGK/CMV provided bright, constitutive nucEGFP expression in both HepG2 and HEK293T cells which was unchanged by DMOG, whereas nucEGFP controlled by the PGK promoter was modestly increased (~2.5 fold) by DMOG (Supp Figure 1e). Substitution of the mnucTomato with the longer mnucTomato-HSVtk-Neo reporter had no effect on DMOG induced reporter induction in EF1a containing HRE-dFLASH cells, still failing to induce tomato expression (Supp Figure 1f). CMV/PGK containing dFLASH sensors maintained DMOG induction when either the mnucTomato or the mnucTomato/HSVtk/Neo reporters were utilised (Supp Figure 1g, h) although mnucTomato without HSVtk and Neo produced lower absolute mnucTomato fluorescence and a smaller percentage of cells responding to DMOG, albeit with lower background. Taken together these findings indicate that certain backbone compositions prevented or enabled robust activation of the enhancer driven cassette, similar to the suppression of an upstream promoter by a downstream, contiguous promoter previously described\(^{25,26}\) suggesting that the 3' EF1a promoter results in poorly functioning multi-cistronic synthetic reporter designs\(^{27}\). Consequently, the PGK/CMV backbone and the mnucTomato/HSVtk/Neo reporter from Supp Figure 1 was chosen as the optimised reporter design (HRE-dFLASH). To confirm that the HRE element was conferring HIF specificity, a no response element dFLASH construct in HEK293T cells treated with DMOG produced no change in either mnucTomato or nucEGFP compared to vehicle-treated populations (Supp Figure 2a). This result, together with the robust induction in response to DMOG (Figure 2D, Supp Figure 1f, 1h), confirms HIF enhancer driven reporter to respond robustly to induction of the HIF pathway (subsequently labelled dFLASH-HIF).
+
+To validate the high inducibility and nucEGFP independence of dFLASH was not specific to the HIF pathway, we generated a Gal4 responsive dFLASH construct (Gal4RE-dFLASH), using Gal4 responsive enhancers\(^{22,28}\). HEK293T cells were transduced with Gal4RE-dFLASH and a dox-inducible expression system to express synthetic Gal4DBDtransactivation domain fusion protein. To evaluate Gal4RE-dFLASH we expressed Gal4DBD fused with a compact VPR (miniVPR), a strong transcriptional activator\(^{29}\) (Supp Figure 2b, 3a-c). We observed ~25% of the polyclonal population was highly responsive to doxycycline treatment (Supp Figure 2b), with a ~14-fold change in Tomato expression relative to nucEGFP by HCl (Supp Figure 3c) demonstrating our optimised dFLASH backbone underpins a versatile reporting platform.
+dFLASH senses functionally distinct TF activation pathways
+
+Following the success in utilising dFLASH to respond to synthetic transcription factor and HIF signalling, we explored the broader applicability of this system to sense other TF activation pathways. We chose the Progesterone Receptor (PGR), a member of the 3-Ketosteroid receptor family that includes the Androgen, Glucocorticoid and Mineralocorticoid receptors, as a functionally distinct TF pathway with dose-dependent responsiveness to progestin steroids to assess the adaptability of dFLASH performance. Keto-steroid receptors act through a well-described mechanism which requires direct ligand binding to initiate homodimerization via their Zinc finger DNA binding domains, followed by binding to palindromic DNA consensus sequences. PGR is the primary target of progesterone (P4, or a structural mimic R5020) and has highly context dependent roles in reproduction depending on tissue type\(^{30,31,32}\). We inserted PGR-target gene enhancer sequences containing the canonical NR3C motif (ACANNNNTGT\(^{31}\)) into dFLASH, conferring specificity to the ketosteroid receptor family to generate PRE-dFLASH (Figure 2b, see Methods).
+
+A chimeric TF system was also established with Gal4DBD fusion proteins to create a synthetic reporter to sense the enzymatic activity of oxygen sensor Factor Inhibiting HIF (FIH). This sensor system termed SynFIH for its ability to synthetically sense FIH activity contained Gal4DBD-HIFCAD fusion protein expressed in a doxycycline-dependent manner, in cells harbouring stably integrated Gal4RE-dFLASH. FIH blocks HIF transactivation through hydroxylation of a conserved asparagine in the HIF-1α C-terminal transactivation domain (HIFCAD), preventing recruitment of the CBP/p300 co-activator complex\(^{22}\). As FIH is a member of the 2-oxoglutarate dioxygenase family, like the PHDs which regulate HIF post-translationally, it is inhibited by DMOG (Figure 2C), allowing induction of SynFIH-dFLASH upon joint Dox and DMOG signalling (Supp Figure 3d,3e). dFLASH-based sensors for PGR and Gal4DBD-HIFCAD generated in the optimised backbone used for dFLASH-HIF (Figure 2a-c). For the PGR sensor we transduced T47D cells with PRE-dFLASH, as these have high endogenous PGR expression, while for the FIH-dependent system we generated HEK293T cells with Gal4RE-dFLASH and the GAL4DBD-HIFCAD system (dFLASH-synFIH).
+
+Stable polyclonal cell populations were treated with their requisite chemical regulators and reporter responses analysed by either flow cytometry or temporal imaging using HCl at 2hr intervals for 38 hours (Figure 2). Flow cytometry revealed all three systems contain a population that strongly induced nucTomato and maintained nucEGFP (Supp Figure 2). In HEK293T cells, ~20% of dFLASH-synFIH and ~50% of dFLASH-HIF population induced Tomato fluorescence substantially relative to untreated controls (Figure 2d, Figure 2f). The ~20% reporter response to inhibition of FIH activity by DMOG (Supp Figure 2e, Figure 2f) is comparable with what was observed for GalRE-dFLASH response to Gal4DBD-miniVPR expression after equivalent selection (Supp Figure 2b). The PGR reporter in T47D cells showed ~50% of the population substantively induced Tomato (Figure 2e, Supp Figure 2d). The presence of considerable responsive populations for FIH, PGR, and HIF sensors, reflected in the histograms of the EGFP positive cells (Figure 2d-f) indicated that isolation of a highly responsive clone or subpopulations can be readily achievable for a range of transcription response types. Importantly, the induction of dFLASH-synFIH by
+Dox/DMOG co-treatment was ablated and displayed high basal Tomato levels in FIH knockout dFLASH-synFIH cells (Supp Figure 3e), indicating that the dFLASH-synFIH specifically senses FIH enzymatic activity.
+
+All dFLASH systems showed consistent signal-dependent increases in reporter activity out to 38 hours by temporal HCl enabling polyclonal populations of dFLASH to track TF activity (Figure 2g-i). PRE-dFLASH was more rapidly responsive to R5020 ligand induction (~6 hours, Figure 2h) than dFLASH-HIF and dFLASH-synFIH to DMOG or Dox/DMOG treatment, respectively (~10 hours, Figure 2g, i). Treatment of PRE-dFLASH with estrogen (E2), which activates the closely related Estrogen Receptor facilitating binding to distinct consensus DNA sites to the PGR, or the hypoxia pathway mimetic DMOG, failed to produce a response on PRE-dFLASH (Figure 2h). This indicates that the PRE enhancer element is selective for the ketosteroid receptor family (also see below), and that enhancer composition facilitates pathway specificity. We also observed a signal-dependent change in EGFP expression by flow cytometry in the T47D PRE-dFLASH reporter cells (Supp Figure 2g) but did not observe a significant change in EGFP expression for HEK293T or HEPG2 dFLASH-HIF (Supp Figure 1c, Supp Figure 2c) or in HEK293T dFLASH-synFIH cells (Supp Figure 2h), with only a small change with Gal4RE-dFLASH with Gal4DBD-miniVPR (Supp Figure 2b). While this change in T47D cells was not detected in the other cellular contexts (see below), it highlights that care needs to be taken in confirming the utility of the constitutive nucEGFP as an internal control in certain scenarios.
+
+Monoclonal dFLASH cell lines confer robust screening potential in live cells
+The observed heterogenous expression of dFLASH within polyclonal cell pools is useful in many assay contexts but reduces efficiency in arrayed high content screening experiments and incompatible with pooled isolation of loss of function regulators. Therefore, monoclonal HEK293T and HepG2 dFLASH-HIF, T47D and BT474 PRE-dFLASH and HEK293T dFLASH-synFIH cell lines were derived to increase reliability of induction, as well as consistency and homogeneity of reporting (Figure 3, Supp Figure 4). The isolated mcdFLASH-synFIH and mcdFLASH-HIF lines also demonstrated constitutive signal insensitive nucEGFP expression (Supp Figure 4a,b,i). While the T47D PRE-mcdFLASH showed a small increase in nucEGFP in response to R5020, this did not preclude the use in normalisation of high content imaging experiments (see below).
+
+No change in EGFP in BT474 PRE-mcdFLASH cells indicates that strong transactivation leading to promoter read through or cell-type specific effects may be at play. Flow cytometry of monoclonal dFLASH cell lines with their cognate ligand inducers (DMOG (Figure 3b), R5020 (Figure 3f) or Dox/DMOG (Figure 3j)) revealed robust homogeneous induction of mnucTomato in all cell lines. Using temporal high content imaging we also found that clonally derived lines displayed similar signal induced kinetics as the polyclonal reporters although displayed higher signal to noise and increased consistency (Figure 3, Supp Figure 4i). Using physiologically relevant concentrations of steroids or steroid analogs (10nM-35nM), the PRE-mcdFLASH lines selectively respond to R5020 (10nM) not E2 (35nM), DHT (10nM), Dexamethasone (Dex, 10nM) or Retinoic acid (RA, 10nM) (Figure 3g, Supp Figure 4i). In addition,
+dose response curves of R5020 mediated Tomato induction indicate that PRE-mcdFLASH line responds to R5020 with an EC_{50} ~200pM, in agreement with orthogonal methods^{33} (Supp Figure 4g, h). This suggests that the PRE-mcdFLASH responds sensitively and selectively to PGR selective agonist R5020, with the potential for high-content screening for modulators of *PGR* activity. As such, we term this line mcdFLASH-PGR from herein, for its specific ability to report on PGR activity at physiological steroid concentrations.
+
+The temporal HCl of populations (**Figure 2** and **Figure 3**) were imaged every 2hrs and do not inherently provide single-cell temporal dynamics of transcriptional responses. Using clonally derived mcdFLASH-PGR or mcdFLASH-HIF lines we also imaged transcriptional responses to R5020 or DMOG, respectively every 15mins (**Supp Video 1 and 2**). High temporal resolution imaging has the potential to monitor transcriptional dynamics in single cells, facilitated by the dual fluorescent nature of dFLASH. Taken together this indicates that clonal lines display improved signal to noise and assay consistency, possibly enabling high content screening experiments.
+
+Typically, high-content screening experiments require high in-plate and across plate consistency, therefore we evaluated mcdFLASH lines (HIF-1α, PGR, FIH) across multiple plates and replicates. System robustness was quantified with the Z' metric^{34} accounting for fold induction and variability between minimal and maximal dFLASH outputs. Signal induced mnucTomato fluorescence across replicates from independent plates was highly consistent (Z' 0.68-0.74) and robust (9.3-11.8 fold, **Figure 3 d, h, l**) the signal induced changes in activity for mcdFLASH-HIF and mcdFLASH-FIH were driven by increased mnucTomato, with minimal changes in nucEGFP (**Figures 3e and 3m**). Despite the changes previously observed in nucEGFP mcdFLASH-PGR in T47D cells provided equivalent reporter to the other systems, (**Figure 3h, i**) as a result, monoclonal mcdFLASH cell lines represent excellent high-throughput screening systems routinely achieving Z' scores > 0.5. Importantly, the induction of the mcdFLASH lines (HEK293T and HepG2 mcdFLASH-HIF, T47D mcdFLASH-PGR and HEK293T mcdFLASH-SynFIH) remained stable over extended passaging (months), enabling protracted large screening applications.
+
+**dFLASH-HIF CRISPR-perturbations of the HIF pathway**
+The robust signal window and high Z' score of mcdFLASH-HIF cell line, coupled with facile analysis by flow cytometry and HCl, indicates that the reporter system is amenable to functional genomic screening. We utilised the recently developed CRISPRoffv2.1 system^{35} to stably repress expression of VHL, which mediates post-translational downregulation of the HIF-1α pathway ^{36,37}. We generated stable mcdFLASH-HIF cells expressing a guide targeting the VHL promoter and subsequently introduced CRISPRoffv2.1 from either a lentivirus driven by an EF1a or SFFV promoter (**Figure 4a, b**). Cells were then analysed by flow cytometry 5- or 10-days post selection to determine if measurable induction of mcdFLASH-HIF reporter was modulated by VHL knockdown under normoxic conditions (**Supp Figure 6**, **Figure 4c, 4d**). As expected, mcdFLASH-HIF/sgVHL cells expressing CRISPRoffv2.1 from either promoter induced the mcdFLASH-HIF reporter in ~35% by 5 days and the majority of cells (~60%) by 10 days as compared to parental cells. Demonstration that mcdFLASH-HIF is responsive to CRISPRi/off perturbations of key regulators of the
+HIF pathway illustrates the potential for the dFLASH platform to provide a readout for CRISPR screens at-scale in a larger format including genome-wide screens.
+
+dFLASH facilitates bimodal screening for small molecule discovery
+Manipulation of the HIF pathway is an attractive target in several disease states, such as in chronic anaemia38 and ischemic disease39 where its promotion of cell adaption and survival during limiting oxygen is desired. Conversely, within certain cancer subtypes40,41 HIF signalling is detrimental and promotes tumorigenesis. Therapeutic agents for activation of HIF-α signalling through targeting HIF-α regulators were initially discovered using in vitro assays. However, clinically effective inhibitors of HIF-1α signalling are yet to be discovered42. The biological roles for HIF-1α and closely related isoform HIF-2α, which share the same canonical control pathway, can be disparate or opposing in different disease contexts requiring isoform selectivity for therapeutic intervention43. To validate that HIF-1α is the sole isoform regulating mcdFLASH-HIF in HEK293T cells44 tandem HA-3xFLAG epitope tags were knocked in to the endogenous HIF-1α and HIF-2α C-termini allowing directly comparison by immunoblot45 and confirmed HIF1a is predominant isoform (Supp Figure 5a). Furthermore, there was no change in DMOG induced mnucTomato expression in HEK293T mcdFLASH-HIF cells when co-treated for up to 72 hours with the selective HIF2a inhibitor PT-2385 (Supp Figure 5b), consistent with the minimal detection of HIF-2α via immunoblot. This confirmed that our HEK293T dFLASH-HIF cell line specifically reports on HIF-1 activity and not HIF-2, indicating that it may be useful for identification of drugs targeting the HIF-1α pathway.
+
+dFLASH-HIF facilitates multiple measurements across different treatment regimens and time points, enabling capture of periodic potentiated and attenuated HIF signalling during a single experiment. Having validated the robust, consistent nature of mcdFLASH-HIF, we exploited its temporal responsiveness for small molecule discovery of activators or inhibitors of HIF-1α signalling in a single, bimodal screening protocol. To test this bimodal design, we utilised a natural product library of 1595 compounds containing structures that were unlikely to have been screened against HIF-1α prior. We first evaluated library compounds for ability to activate the reporter after treatment for 36 hours (Figure 5a) or 24 hours (Figure 5d). The selection of two different screening time points was to minimise any potential toxic effects of compounds at the later time points. Consistency of compound activity between the two screens was assessed by Pearson correlations (Supp Figure 7i, R = 0.79, p < 2.2x10^{-16}). Lead compounds were identified by their ability to increase mnucTomato/nucEGFP (Figure 5b, c) and mnucTomato MFI more than 2SD compared with vehicle controls, with less than 2SD decrease in nucEGFP (21/1595 compounds (1.3%) each expt; Supp Figure 7a, e) and an FDR adjusted P score <0.01 across both screens (3/1595 (0.18%) compounds; Supp Figure 7b, f). After imaging of reporter fluorescence to determine these compound’s ability to activate HIF-1α we then treated the cells with 1mM DMOG and imaged after a further 36-hour (Figure 5c) and 24-hour (Figure 5f) period. Again, consistency of compound activity was assessed by Person correlation (Supp Figure 7j, F, R = 0.62, p < 2.2x10^{-16}). Lead compounds were defined as those exhibiting a decrease in mnucTomato MFI >2SD from DMOG-treated controls in each screen without changing nucEGFP>2SD relative to the DMOG-treated controls (26/1595 compounds (1.3%) (36hr treatment) and
+13/1595 compounds (<1%) (24hr treatment); Supp Figure 7c, g), and decrease in mmucTomato/nucEGFP >2SD with an FDR adjusted P score < 0.01 (3/1595 compounds (0.18%) across both expt; Supp Figure 7d, h).
+
+dFLASH identified novel and known compounds that alter HIF TF activity.
+We confirmed 11 inhibitors and 18 activators of HIF1a activity identified from the pilot screen at three concentrations (Supp Figure 8a, 9a) identifying RQ500235 and RQ200674 (Figure 6a, d) as previously unreported HIF-1α inhibiting or stabilising compounds, respectively. RQ200674 increased reporter activity 2-fold in repeated assays (Figure 6d) and stabilised endogenously tagged HIF-1α at normoxia in HEK293T cells (Supp Figure 8b). Mechanistically, RQ200674 had weak iron chelation activity in an in vitro chelation assay (Figure 6e), suggesting it intersects with the HIF-1α pathway by sequestering iron similar to other reported HIF stabilisers. In the inhibitor compound dataset, Celastrol and Flavokawain B downregulated the reporter at several concentrations (Supp Figure 9b, c). Celastrol is a previously reported HIF-1α inhibitor46-48 and Flavokawain B is a member of the chalcone family which has previously exhibited anti-HIF-1α activity49. RQ500235 was identified as a HIF-1 inhibitor by mcdFLASH-HIF screening. Dose dependent inhibition of mcdFLASH-HIF (Figure 6a) correlated with a dose-dependent decrease in protein expression by immunoblot (Figure 6C). We observed significant (p=0.0139) downregulation of HIF-1α transcript levels (Figure 6D) and were unable to rescue HIF-1α protein loss with proteasomal inhibition (Supp Figure 9d), indicating RQ500235 was decreasing HIF-1α at the RNA level. More broadly however, the identification of these compounds by mcdFLASH-HIF in the bimodal set up demonstrates successful application of the dFLASH platform to small molecule discovery efforts for both gain and loss of TF function.
+Discussion
+We designed and optimised dFLASH to offer a versatile, robust live-cell reporting platform that is applicable across TF families and allows for facile high-throughput applications. We validated dFLASH against three independent signal-responsive TFs, two with endogenous signalling pathways (dFLASH-PRE for Progesterone receptors; dFLASH-HRE for hypoxia induced transcription factors) and a synthetic system for a hybrid protein transcriptional regulator (dFLASH- Gal4RE). Each dFLASH construct produced robustly detected reporter activity by temporal high-content imaging and FACS after signal stimulation for its responsive TF (Figure 2,3). The use of previously validated enhancer elements for HIF\(^{24}\) and synthetic Gal4 DNA binding domains\(^{22,28}\) demonstrated that dFLASH can be adapted toward both endogenous and synthetic pathways displaying highly agonist/activator-specific responses, indicating utility in dissecting and targeting distinct molecular pathways. mcdFLASH lines distinct pathways produced highly consistent (\(Z' = 0.68\)-0.74) signal induced Tomato induction measured by high content imaging suggesting dFLASH is ideally suited to arrayed high-throughput screening (Figure 3). In addition, mcdFLASH lines also displayed homogenous signal induced reporter induction by flow cytometry indicating that pooled high content screening would also be possible.
+
+Indeed, reporter systems like dFLASH have been increasingly applied to functional genomic screens which target specific transcriptional pathways\(^{9,50-52}\). CRISPRoff mediated downregulation of the core HIF protein regulator, VHL produced distinct tomato expressing cell pools (Figure 4), demonstrating genetic perturbations of endogenous TF signalling pathways. The robust induction of the dFLASH-HIF reporter upon VHL knockdown in the majority of cells indicates that whole genome screening would also be successful\(^{9,17,50,53}\).
+
+Using the HIF-1\(\alpha\) specific reporter line, mcdFLASH-HIF, the application of high-content screening was exemplified. This approach was successful in discovering a novel activator and novel inhibitor of the HIF pathway, as well as previously identified inhibitory compounds. This ratified dFLASH as a reporter platform for arrayed-based screening and demonstrates the utility of the linked nucEGFP control for rapid hit bracketing. The novel inhibitor RQ500235 was shown to downregulate HIF-1\(\alpha\) transcript levels, like another HIF-1\(\alpha\) inhibitor PX-478\(^{54,55}\). As PX-478 has demonstrated anti-cancer activity in several cell lines \(^{55,56}\) and preserved \(\beta\)-cell function in diabetic models\(^{54}\), a future similar role may exist for an optimised analogue of RQ500235.
+
+The dFLASH system is characterised by some distinct advantages which may enable more precise dissection of molecular pathways. The ability to control for cell-to-cell fluctuations and to decouple generalised or off-target effects on reporter function may aid the precision necessary for large drug library or genome-wide screening applications\(^{57}\). In addition, dFLASH, unlike many other high-throughput platforms can be used to screen genetic or drug perturbations of temporal transcriptional dynamics or as used here at multiple time points. Also, the results indicate that dFLASH is ideally suited to array-based functional genomics approaches\(^{58}\) allowing for multiplexing with other phenotypic or molecular outputs\(^{59,60,2,61}\).
+The dFLASH approach has some limitations. The fluorescent nature of dFLASH limits the chemical space by which it can screen due to interference from auto-fluorescent compounds. In addition, we acknowledge that fluorescent proteins require O2 for their activity and this limits the use of mnucTomato as a readout of hypoxia. Also, while the backbone design has been optimised for a robust activation of a variety of transcription response pathways, the mechanistic underpinning of this is unclear and could be further improved, providing insights into the sequence and architectural determinants of enhancer activation in chromatin. In addition to the strong effect of the dFLASH downstream promoter on upstream enhancer activity it is clear that either the distance between contiguous promoter/enhancer or the sequence composition of the linker has a functional consequence on enhancer induction.
+
+The incorporation of robust native circuits such as those described here (Hypoxia or Progesterone) has the potential to allow the manipulation or integration of these pathways into synthetic biology circuitry for biotherapeutics. In these cases, it is critical that robust signal to noise is achieved for these circuits to effectively function in biological systems. Further, the use of a synthetic approach to 'sense' FIH enzymatic activity through the HIF-CAD:P300/CBP interaction opens up the possibility that other enzymatic pathways that lack effective *in vivo* activity assay may also be adapted. We also envisage that dFLASH could be adapted to 2-hybrid based screens as a complement to other protein-protein interaction approaches.
+
+The ability to temporally track TF regulated reporters in populations and at the single-cell level enable dFLASH to be used to understand dynamics of transcriptional responses as has been used to dissect mechanisms of synthetic transcriptional repression\(^{7,8}\) or understand notch ligand induced synthetic transcriptional dynamics\(^{62}\). For instance, synthetic reporter circuits have been used to delineate how diverse notch ligands induce different signalling dynamics \(^{62}\). The large dynamic range of the dFLASH-PGR and HIF reporter lines in conjunction with the high proportion of cells induced in polyclonal pools (\textbf{Figure 2}) also suggests dFLASH as a candidate system for forward activity-based enhancer screening. These approaches have been applied to dissect enhancer activity or disease variants with other similar systems such as lentiviral-compatible Massively Parallel Reporter Assays (LentiMPRA)\(^{63,64}\). However, the use of the internal control normalisation provided by dFLASH may be useful in separating chromosomal from enhancer driven effects in forward enhancer screens.
+
+Given dFLASH has robust activity in both pooled and arrayed formats, it offers a flexible platform for investigations. dFLASH can be used to sense endogenous and synthetic transcription factor activity and represents a versatile, stable, live-cell reporter system of a broad range of applications.
+Figure 1. Summary of dFLASH LV-REPORT construction, utility, and validation
+(a) The dFLASH system utilises the lentiviral LV-REPORT construct, consisting of a cis-element multiple cloning site for enhancer insertion, followed by a minimal promoter that drives a transcription factor (TF) dependent cassette that encodes three separate expression markers; a nuclear Tomato fluorophore with a 3x C-terminal nuclear localisation signal (NLS), Herpes Simplex Virus Thymidine Kinase (HSVtK) for negative selection and Neomycin resistance (Neo) for positive selection separated by a 2A self-cleaving peptide (2A). This is followed by a downstream promoter that drives an independent cassette encoding EGFP with a 3x N-terminal NLS, and a Hygromycin resistance selection marker separated by a 2A peptide. (b) This design allows for initial identification of the EGFP fluorophore in nuclei, independent of signal. Expression of the Tomato fluorophore is highly upregulated in a signal-dependent manner. Images shown are monoclonal HEK293T dFLASH-HIF cells. Populations were treated for 48 hours ±DMOG induction of HIF-1α and imaged by HCl. (c) This system can be adapted to a range of different applications. This includes (clockwise) flow cytometry, arrayed screening in a high throughput setting with high content imaging, isolation of highly responsive clones or single cells from a heterogenous population or temporal imaging of pooled or individual cells over time.
+Figure 2. dFLASH provides sensitive readouts to three distinct TF pathways
+(a-c) Three distinct enhancer elements enabling targeting of three different signalling aspects. (a) Hypoxic response elements (HRE) provide a read out for HIF-1α activation; (b) Progesterone response elements (PRE) derived from progesterone receptor target genes facilitate reporting of progestin signaling; (c) Gal4 response elements (GalRE) enable targeting of synthetic transcription factors to dFLASH such as a GAL4DBD-HIFCAD fusion protein that provides a FIH-dependent reporter response. (d-f) Flow cytometry histograms showing Tomato expression following 48 hr treatments of the indicated dFLASH polyclonal reporter cells (d) HEK293T; 1mM DMOG or 0.1% DMSO (Ctrl), (e) T47D; 100nM R5020 or Ethanol (Ctrl), (f) HEK293T; 1μg/mL Doxycycline (Dox) and 1mM DMOG or Dox and 0.1% DMSO (Ctrl). (g-i) Reporter populations as in d-f were temporally imaged for 38 hours using HCl directly after treatment with (g) 0.5mM DMOG or 0.1% DMSO, (4 replicates) (h) 100nM R5020, 35nM E2, 0.5mM DMOG or 0.1% Ethanol (EtOH) (4 replicates), (i) 0.1% DMSO, 1mM DMOG, 100ng/mL Dox and 0.1% DMSO, or 100ng/mL Dox and 1mM DMOG (4 replicates).
+a.
+
+HIF response Pathway
+HEK293T dFLASH-HIF
+
+PGR response Pathway
+T47D dFLASH-PGR
+
+b. Ctrl DMOG ~11 x
+c. DMSO DMOG
+d. Z' = 0.74 FC = 9.3
+e.
+
+Synthetic FIH Sensor
+HEK293T dFLASH-synFIH
+
+j. Ctrl DMOG ~5 x
+k. DMSO DOX/DMOG DOX/DMSO DMSO
+l. Z' = 0.68 FC = 11.3
+m.
+Figure 3. Derivation of robust, screen-ready dFLASH clonal lines
+
+(a) Schematic for derivation and assessment of robustness for clonal lines of (b-e) HEK293T dFLASH-HIF (mcdFLASH-HIF), (f-i) T47D dFLASH-PGR (mcdFLASH-PGR) and (j- m) HEK293T dFLASH-synFIH (mcdFLASH-synFIH) were analysed by flow cytometry, temporal HCl over 38 hours and for inter-plate robustness by mock multi-plate high throughput screening with HCl. (b-e) mcdFLASH-HIF was (b) treated with DMOG for 48 hours and assessed for Tomato induction by flow cytometry relative to vehicle controls with fold change between populations displayed and (c) treated with vehicle or 0.5mM DMOG and imaged every 2 hours for 38 hours by HCl (mean ±sem, 8 replicates). (d-e) mcdFLASH-HIF was treated for 48 hours with 1mM DMOG or vehicle (6 replicates/plate, n = 10 plates) by HCl in a high throughput screening setting (HTS-HCl) for (d) normalised dFLASH expression and (e) Tomato MFI alone. (f – i) T47D mcdFLASH-PGR was (f) assessed after 48 hours of treatment with 100nM R5020 by flow cytometry for Tomato induction and (g) treated with 10nM R5020, 35nM E2, 10nM DHT and vehicle then imaged every 2 hours for 38 hours by temporal HCl for normalised dFLASH expression (mean ±sem, 8 replicates). (h-i) T47D mcdFLASH-PGR was assessed by HTS-HCl at 48 hours (24 replicates/plate, n = 5 plates) for (h) dFLASH normalised expression and (i) Tomato MFI alone. (j) HEK293T dFLASH-synFIH was assessed, with 200ng/mL and or Dox and 1mM DMOG by flow cytometry for dFLASH Tomato induction (k) mcdFLASH-synFIH was treated with 100ng/mL Dox, 1mM DMOG and relevant vehicle controls and assessed for reporter induction by temporal HCl (mean ±sem 4 replicates). (l-m) mcdFLASH-synFIH cells were treated with 200ng/mL Dox (grey), 1mM DMOG (red), vehicle (pink) or Dox and DMOG (orange) and assessed by HTS-HCl after 48 hours (24 replicates/plate, n = 3 plates) for (l) normalised dFLASH expression or (m) Tomato MFI induction between Dox and Dox and DMOG treated populations. Dashed lines represent 3SD from relevant vehicle (+3SD) or requisite ligand treated population (-3SD). Fold change for flow cytometry and HTS-HCl (FC) is displayed. Z’ was calculated from all analysed plates by HTS-HCl. Z’ for all plates analysed was > 0.5.
+Figure 4. Near homogenous activation of mcdFLASH-HIF by CRISPRoff knockdown of VHL.
+
+(a) Clonal (1) mcdFLASH-HIF lines derived post-hygromycin (HygroB) selection were transduced first with the (2) sgRNA vector targeting *VHL* transcriptional start site, followed by puromycin selection (Puro). This pool was subsequently transduced by the (3) CRISPRoffv2.1 virus and selected with blasticidin (BlastS) prior to flow cytometry (on day 5 and 10 post Blasticidin selection). (b) The (1) dFLASH vector with the HRE enhancer was transduced as were 2 variants of the CRISPRoffv2.1 vector with either (3A) EF1α promoter or (3B) SFFV promoter driving the dCas9 expression cassette. (c, d) Flow cytometry for dFLASH-HIF induction in response to the CRISPRoffv2.1 VHL knockdown relative to parental line (Ctrl) with (c) EF1a or (d) SFFV expression constructs after 10 days of selection.
+a. dFLARE-HIF
+Compound Library
+36hrs → Activation Screening → 36hrs 1mM DMOG → Inhibitor Screening
+Hit ● GFP Unchanged ■ GFP High ▼ GFP Low
+
+b.
+
+c.
+
+d. dFLARE-HIF
+Compound Library
+24hrs → Activation Screening → 24hrs 1mM DMOG → Inhibitor Screening
+Hit ● GFP Unchanged ■ GFP High ▼ GFP Low
+
+e.
+
+f.
+Figure 5. Bimodal small molecule screening of the HIF signalling pathway with dFLASH-HIF identifies positive and negative regulators
+(a) HEK293T mcdFLASH-HIF cells were treated with a 1595 compound library and incubated for 36 hours prior to (b) the first round of HCl normalised dFLASH activity. Compounds that changed EGFP >±2SD are shown in grey and excluded as hits. Compounds that increase Tomato/EGFP >2SD from the vehicle controls (dashed line) are highlighted in red. After the activation screen, the compound wells were then treated with 1mM DMOG for 36 hours prior to the second round of HCl. Compounds that decreased dFLASH activity greater than 2SD from DMOG controls (dashed line) are shown in red. Compounds that changed EGFP >±2SD are shown in grey and excluded as hits. Normalised dFLASH output (Z scoring) for all analysed wells. (d-f) The screening protocol of (a-c) was repeated using 24 hr points for HCl.
+Figure 6. Investigating mechanisms for HIF-1α regulation by hit dFLASH-HIF inhibitor RQ500235 and hit activator RQ200674
+
+(a, b) Inhibitor RQ500235 identified from the bimodal screen (a) represses DMOG induced Tomato in dFLASH-HIF cells in a dose dependent manner (n=2, Tom MFI, red; Tom normalised to EGFP, black) and (b) decreases expression of HIF-1α protein as assessed by immunoblot of whole cell extracts from endogenous HA-Flag tagged HIF-1α in HEK293T cells. S.E.= short exposure; L.E.= long exposure. (c) RT-PCR shows HIF-1α transcript is significantly decreased in HEK293T cells treated for 6 hours with RQ500235 (n =3, *p=0.0139). (d) Activator RQ200674 identified from the bimodal screen recapitulated activation of dFLASH-HIF at 50μM in HEK293T cells (n = 2). (e) in vitro iron chelation assay of RQ200674 displays weak chelating activity at 236μM from line of best fit (n = 3) compared to positive control iron chelator and HIF-1α activator, dipyrldyl.
+639
+640    Supplementary Figures
+641
+642
+Supplementary Figure 1. Optimised dFLASH design produces a robust HIF sensor.
+
+(a-b) HEK293T cells with HRE-dFLASH constructs without EGFP and (a) expressing monomeric Tomato or (b) dimeric Tomato were treated -/+ 1mM DMOG for 48 hours and quantified by FACS. Tomato MFI >200AU was used to compare induction (black line). (c-e) HEK293T and HEPG2 cells were transduced with HRE-dFLASH reporters that had different downstream promoters controlling EGFP or Tomato cassette composition and treated for 48 hours -/+ 1mM DMOG prior to HCl. The (d) Tomato/EGFP MFI ratio and (e) EGFP MFI for each backbone variant was then compared (Data from three independent biological replicates). (f) HEK293T cells transduced with reporter constructs containing the downstream PGK/CMV or EF1a promoters were compared for DMOG induction by HCl after 48 hours of -/+ 1mM
+DMOG treatment (Data from three independent biological replicates). Significance was assessed with a Two-Way ANOVA (**** p < 0.001, ns = not significant). (g,h) HEK293T cells with the HRE enhancer and different dFLASH backbone compositions of (g) PGK/CMV dFLASH with Tomato alone as the upstream cassette or (h) dFLASH-HIF were treated for 48-hours -/+ 1mM DMOG prior to EGFP analysis and Tomato induction by FACS.
+a. No Enhancer Control
+HEK293T dFLASH
+
+b. Synthetic TF response
+HEK293T dFLASH-synVPR
+
+c. HEK293T dFLASH-HRE
+
+d. T47D dFLASH-PRE
+
+e. HEK293T dFLASH-synFIH
+
+f.
+
+g.
+
+h.
+
+Supplementary Figure 2. dFLASH provides a TF-responsive, versatile reporter platform in heterogenous cell pools.
+(a-b) HEK293T cells were transduced with (a) dFLASH with no enhancer and treated with 1mM DMOG or 0.1% DMSO (Ctrl) or (b) GalRE-dFLASH and Gal4DBD-miniVPR and treated with H2O (Ctrl) or 1μg/mL Dox for 48 hours prior to FACS. Dot plots of populations’ Tomato and EGFP intensity with or without activating chemicals and histograms comparing EGFP and Tomato MFI between control and treated populations are shown. (c-h) Dot plots and EGFP histograms for control and chemical treated (c, f) dFLASH-HIF, (d, g) dFLASH-PR polyclonal pools (to accompany Figure 2a-c) and (e, h) dFLASH-synFIH.
+Supplementary Figure 3. Synthetic transcription factors drive a strong response from the GalRE-dFLASH reporter and can respond to endogenous signaling pathways.
+
+(a) GAL4DBD-miniVPR is expressed from an independent dox-inducible vector that subsequently binds to GalRE-dFLASH. (b,c) HEK293T GalRE-dFLASH cells were transduced with GAL4DBD-miniVPR expression construct and were treated -/+ doxycycline for 48 hours prior to HCl for (b) Tomato expression (top panels) and EGFP expression (bottom panels). (c) Normalised fluorescence intensity was also quantified for treated populations (n=3, mean ±sem). FC is Fold change between the populations. (d, e) To confirm HEK293T dFLASH-synFIH system was FIH dependent, (d) GalRE-dFLASH and GAL4DBD-HIFCAD vectors were transduced into HEK293T cells with FIH knocked out. (e) FIH KO cells were compared with wildtype HEK293T dFLASH-synFIH (WT) in a 200ng/mL dox background for DMOG-dependent reporter induction by HCl (n=3). (c, e) Significance was assessed by t-test with Welch’s correction (ns = not significant, *** p <0.001, ****p <0.0001).
+HIF response Pathway
+
+HEK293T dFLASH-HRE        HepG2 dFLASH-HRE
+
+a. Ctrl   DMOG
+b. Ctrl   DMOG
+c. Ctrl   DMOG
+
+PGR response Pathway
+
+T47D dFLASH-PRE        BT474 dFLASH-PRE
+
+d. Ctrl   R5020
+e. Ctrl   R5020
+f. Ctrl   R5020
+
+g.
+
+h.
+
+i.
+
+Synthetic FIH Sensor
+HEK293T dFLASH-synFIH
+
+j. Ctrl   DMOG
+k. mcDMOG, mcDMSO, pcDMOG, pcDMSO
+Supplementary Figure 4. Clonal dFLASH cell lines enable improved reporting across different cell types.
+
+(a-c) Flow cytometry of clonal dFLASH-HIF cell lines for (a) HEK293T (see also Figure 3b) and (b,c) HepG2 cells after 48 hours -/+ 0.5mM DMOG. (d-h) dFLASH-PGR functionality was assessed by flow cytometry in (d)T47D (see also Figure 3f) and (e,f) BT474 cells after 48 hours -/+ 100nM R5020. (g,h) T47D dFLASH-PGR cells were treated with increasing concentrations of R5020 (0.01-100nM, 8 replicates per group) and (g) imaged over 38 hours with temporal HCl or (h) imaged at 48 hours to determine sensitivity to R5020. (i) Comparison of inductions of the T47D mcdFLASH-PGR line to different steroids (10nM R5020, 35nM E2, 10nM DHT, 10nM Dex, 10nM RA) by HCl after 48 hours of treatment. (g) and (i) are the mean±sem of normalised Tomato/GFP (within each expt) from n = 3 independent experiments (24 replicates), except Dex and RA (n=2 (16 replicates)). (j, k) Clonally derived HEK293T dFLASH-synFIH cells were (j) analysed by flow cytometry after 48 hours of 200ng/mL Dox -/+ 1mM DMOG (see also Figure 3k) with (k) showing temporal HCl comparisons between monoclonal (mc) and polyclonal (pc) lines (see also Figure 2j).
+a. Clone: 1   2   1   2
+S.E. WB:HA  0   16   0   16   0   16   0   16 <1% O_2
+HIF-1α   HIF-2α
+L.E. WB:HA  150kDa ——   150kDa ——   150kDa ——   150kDa ——
+HIF-1α   HIF-2α
+
+b.
+
+![Western blot and graph showing HIF-1α and HIF-2α levels under hypoxic conditions](page_349_180_749_377.png)
+
+![Line graph comparing Tomato/GFP ratios for Ctrl, DMOG, PT2385, and PT2385 + DMOG over time](page_349_601_749_377.png)
+Supplementary Figure 5. HIF-1α is the predominant isoform that affects the dFLASH reporter in HEK293T cells
+
+(a) Monoclonal HEK293T cells with endogenously HA-Flag tagged HIF-1α or HIF-2α were treated with hypoxia (<1% O₂) for 16 hours prior to anti-HA immunoblotting of whole cell extracts. S.E.= short exposure; L.E.= long exposure. Representative of three independent experiments. (b) mcdFLASH-HIF cells were treated -/+ 1mM DMOG and -/+ 10μM of the HIF-2α antagonist (PT-2385) as indicated and quantified by HCl over 72-hour period.
+a.
+
+1: dFLARE-HRE virus   2:VHL sgRNA virus   2:CRISPRoffv2.1 virus
+
+HygroB   Puro   BlastS
+
+clonal dFLARE-HRE cell line   dFLARE-HRE/CRISPRoffv2.1/sgVHL
+
+5 days   10 days
+
+b. -ve   c. EF1α   d. SFFVp
+
+Supplementary Figure 6. CRISPRoff mediated VHL knockdown induces mcdFLASH-HIF reporter lines.
+(a) HEK293T cells were first transduced with dFLASH-HRE and a clonal reporting line was derived after hygromycin (HygroB) treatment. This line was in turn transduced with the VHL sgRNA vector and selected with puromycin (Puro). This line was then transduced with the CRISPRoffv2.1 vector and selected with blasticidin S (Blast) and populations were subjected to flow cytometry after 5 days or 10 days of selection for analysis of reporter expression. (b-d) dot plots for dFLASH expression from the (b) non-CRISPRoff parental line, (c) EF1a-CRISPRoffv2.1 transduced and (d) SFFVp-CRISPRoffv2.1 populations after 5 or 10 days of blasticidin selection (see also Figure 4).
+36-hour Activator Screen
+
+a.      b.
+
+24-hour Activator Screen
+
+e.      f.
+
+i.      Activator Screen
+R = 0.79, p < 2.2e-16
+
+36-hour Inhibitor Screen
+
+c.      d.
+
+24-hour Inhibitor Screen
+
+g.      h.
+
+j.      Inhibitor Screen
+R = 0.62, p < 2.2e-16
+Supplementary Figure 7. Hit selections and assessment of bimodal screen reproducibility between independent screens for activators and inhibitors of HIF-1α.
+
+Compound-induced dFLASH-HIF reporter activity was used to score hits from the (a-d) 36-hour or the (e-h) 24-hour bimodal screens according to Tomato MFI and adjusted P scores. Lines indicate cut offs for hit criteria with hits shown in red for each metric and dismissed compounds that change EGFP >±2SD shown in grey. (i, j) Pearson correlations of the Tomato/EGFP between the 36-hour and the 24-hour screens for (i) reporter activation (R = 0.62, p < 2.2×10^{-16}) or (j) reporter inhibition (R = 0.62, p < 2.2×10^{-16}) for all 1595 compounds screened. Line indicates line of best fit, grey boundary is 95% confidence interval.
+Supplementary Figure 8. Rescreening of activator hits from 1595 compound small molecule screen reveals RQ200674 causes normoxic stabilisation of HIF-1α.
+
+(a) The 11 top performing hits from the activator screens, including RQ200674 (see also Figure 6d) were rescreened against HEK293T mcdFLASH-HIF at 10μM, 25μM and 50μM. Comparisons between Tomato/GFP and Tomato MFI dFLASH induction shown against vehicle (-ve Ctrl) and 1mM DMOG (+ve Ctrl) treated populations (n=2).
+
+(b) Immunoblot of whole cell extracts from HEK293T cells containing endogenously HA-Flag tagged HIF-1α and treated as indicated with vehicle (0.1% DMSO), 1mM DMOG (+ve Ctrl), or 100μM and 200μM of RQ200674 for 18 hours. Representative of 2 independent experiments.
+a.
+
+![Bar graph showing Tomato/EGFP and Tomato MFI for various compounds](page_120_120_1208_370.png)
+
+b.
+
+![Chemical structure of compound RQ100976](page_120_480_370_120.png)
+
+c.
+
+![Chemical structure of compound RQ000155](page_600_480_370_120.png)
+
+d.
+
+<table>
+  <tr>
+    <th></th>
+    <th>DMOG</th>
+    <th>RQ500235</th>
+    <th>MG132</th>
+  </tr>
+  <tr>
+    <td>WB:HA</td>
+    <td>-</td>
+    <td>-</td>
+    <td>-</td>
+    <td>+</td>
+    <td>+</td>
+    <td>+</td>
+    <td>+</td>
+  </tr>
+  <tr>
+    <td>WB:GAPDH</td>
+    <td>-</td>
+    <td>-</td>
+    <td>-</td>
+    <td>+</td>
+    <td>+</td>
+    <td>+</td>
+    <td>+</td>
+  </tr>
+  <tr>
+    <td>WB:ARNT</td>
+    <td>-</td>
+    <td>-</td>
+    <td>-</td>
+    <td>+</td>
+    <td>+</td>
+    <td>+</td>
+    <td>+</td>
+  </tr>
+</table>
+
+![Western blot images showing HIF-1α, GAPDH, and ARNT bands](page_120_1040_1208_120.png)
+Supplementary Figure 9. Flavokawain B, Celastrol and RQ500235 decrease dFLASH-HIF and proteasomal inhibition doesn’t rescue RQ500235 impact on HIF-1α.
+
+(a-c) The 18 top inhibitory compounds, including (b) Flavokawain B (RQ100976),(c) Celastrol (RQ000155) and RQ500235 (see also Figure 6a) were rescreened against dFLASH-HIF at 10μM, 25μM and 50μM in 1mM DMOG treated 293T dFLASH-HIF cells (24 hours). Comparisons between Tomato/GFP and Tomato MFI dFLASH induction shown against 0.1% DMSO (-ve Ctrl) and 1mM DMOG (+ve Ctrl) treated populations (n=2). (d) Immunoblot of whole cell extracts from HEK293T cells with endogenously HA-Flag tagged HIF-1α following a 12 hr treatment period with with the indicated combinations of 1 mM DMOG (full12 hr), 50μM RQ500235 (final 6 hr) and 10μM MG132 (final 3 hr). Representative of 2 independent experiments.
+Supplementary Movie 1. Single cell temporal dynamics of HEK293T mcdFLASH-HIF cells
+HEK293T mcdFLASH-HIF cells were seeded at 1×10^5 cells/dish in Poly-D-Lysine coated plates overnight prior to imaging with spinning disk confocal microscopy at 40x magnification. Cells were imaged every 15 min for 48 hours for Tomato (Magenta) and EGFP (Green) expression. Time stamps are given in top left.
+
+![Single cell temporal dynamics of HEK293T mcdFLASH-HIF cells](page_246_180_755_563.png)
+Supplementary Movie 2. Single cell temporal dynamics of T47D mcdFLASH-PGR cells
+T47D mcdFLASH-PGR cells were seeded at 5×10^5 cells/dish in Poly-D-Lysine coated plates overnight prior to imaging with spinning disk confocal microscopy at 40x magnification. Cells were imaged every 15 min for 48 hours for Tomato (Magenta) and EGFP (Green) expression. Time stamps are given in top left.
+
+Methods:
+
+Plasmid Construction. cDNAs were amplified using the Phusion polymerase (NEB) and assembled into ClaI/NheI digested pLV410 digested backbone by Gibson assembly31. Sequence verified LV-REPORT plasmid sequences and constructs are listed in Supplementary Table 1. Briefly, the plasmids contained an upstream multiple cloning sites followed by a minimal promoter (derived from the pTRE3G minimal promoter) and then followed by a reporter construct mnucTomato/HSVtk-2a-Neo or other variants). This was then followed by a constitutive promoter (EF1a, PGK or PGK/CMV) driving the expression or hygromycinR cassette with or without a 2a linked d2nucEGFP (Supplementary Figure 1C).
+To improve the performance of our previously reported lentiviral inducible expression systems65, the PGK promoter in Tet-On3G IRES Puro was replaced by digestion with MluI/NheI and insertion of either EF1a-Tet-On3G-2A-puro, EF1a-Tet-On3G-2A-BlastR or EF1a-Tet-On3G-2A-nucTomato using Phusion polymerase (NEB) amplified PCR products from existing plasmids. Plasmids were cloned by Gibson isothermal assembly and propagated in DB3.1 cells (Invitrogen). We also generated a series of constitutive lentiviral plasmids as part of this work pLV-Egl-BlastR (EF1a-Gateway-IRES-BlastR), pLV-Egl-ZeoR (EF1a-Gateway-IRES-ZeoR), pLV-Egl-HygroR (EF1a-Gateway-IRES-HygroR), pLV-SFFVp-gl-BlastR (SFFVp-Gateway-IRES-BlastR), pLV-SV40p-gl-BlastR (SV40p-Gateway-IRES-BlastR). These plasmids were constructed by isothermal assembly of G-Blocks (IDT DNA) or PCR fragments, propagated in ccbD competent cells, sequence verified and deposited with Addgene (Supplementary Table 1).
+
+The Lentiviral backbone expression construct pLV-TET2BLAST-GtwyA was then using to insert expression constructs cloned into pENTR1a by LR Clonase II enzyme recombination (Cat#11791020, Thermo). GAL4DBD-HIFCAD (727-826aa) and the GAL4DBD28 were cloned into pENTR1a by Scal/EcoRV or KpnI/EcoRI respectively. The miniVPR sequence29 was cloned into the pENTR1a-GAL4DBD construct at the EcoRI and NotI sites. The pENTR1a vectors were then Gateway cloned into the pLV-TET2PURO-GtwyA vector. pENTR1a-CRISPRoffv2.1 was generated by inserting an EcoRI/NotI digested CRISPRoffv2.1 (CRISPRoff-v2.1 was a gift from Luke Gilbert, Addgene #167981) into pENTR1a plasmid. pLV-SFFVp-CRISPRofv2.1-IRES-BLAST and pLV-EF1a-CRISPRofv2.1-IRES-BLAST were generated by pENTR1a by LR Clonase II enzyme recombination (Cat#11791020, Thermo). All Lentiviral plasmids were propagated in DH5α without any signs of recombination.
+
+Enhancer element cloning. The 12x HRE enhancer from hypoxic response target genes (PGK1, ENO1 and LDHA) was liberated from pUSTdS-HRE12-mCMV-lacZ24 with XbaI/SpeI and cloned into AvrII digested pLV-REPORT plasmids. Progesterone responsive pLV-REPORT-PRECat PREcat was cloned by isothermal assembly of a G-Block (IDT-DNA) containing enhancer elements from 5 PGR target gene enhancers (Zbtb16, Fkbp5, Slc17a11, Erfnb1, MT2)31 into Ascl/ClaI digested pLV-REPORT(PGK/CMV). Gal4 response elements (5xGRE) were synthesised (IDT DNA) with ClaI/Ascl overhangs and cloned into Cla/Ascl digested pLV-REPORT(PGK/CMV). Sequences are in Supplementary Table 2.
+
+Mammalian cell culture and ligand treatment. HEK293T (ATCC CRL-3216), HEPG2 (ATCC HB-8065) line were grown in Dulbecco’s Modified Eagle Medium (DMEM high glucose) + pH 7.5 HEPES (Gibco), 10% Foetal Bovine Serum (Corning 35-076-CV or Serana FBS-AU-015), 1% penicillin-streptomycin (Invitrogen) and 1% Glutamax (Gibco). T47D (ATCC HTB-133) or BT474 (ATCC HTB-20) were grown in RPMI 1640 (ATCC modified) (A1049101 Gibco) with 10% Foetal Bovine Serum (Fisher Biotech FBS-AU-015) and 1% penicillin-streptomycin66. Cells were maintained at 37°C and at 5% CO2. Clonal lines were isolated by either limiting dilution or FACS single cell isolation into 96 wells trays. Resultant monoclonal populations were evaluated for single colony formation or assessed by HCl or FACS. Ligand treatments were done 24 hours after seeding of cells in requisite plate or vessel. Standard
+concentrations and solvent, unless specified otherwise, are 200ng/mL Doxycycline (Sigma, H2O), 0.5mM or 1mM DMOG (Cayman Scientific, DMSO), 100nM R5020 (Perkin-Elmer NLP004005MG, EtoH), 35nM Estradiol (E2, Sigma E2758, EtOH), 10nM all-trans retinoic acid (RA, Sigma #R2625), 10nM Dihydrotestosterone (DHT, D5027), 10nM Dexamethasone (Dex, Sigma D4902), 10μM PT-2385 (Abcam, DMSO).
+
+Lentiviral Production & stable cell line production. Near confluent HEK293T cells were transfected with either psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259) or pCMV-dR8.2 dvpr (Addgene #8455), pRSV-REV (Addgene; #12253) and pMD2.G along with the Lentivector (described above) and PEI (1μg/μl, polyethyleneimine) (Polysciences, USA), Lipofectamine 2000, or Lipofectamine 3000 at a 3μl:1μg ratio with DNA. Media changed 1-day post-transfection to complete media or Optimem. Virus was harvested 1-2 days post-transfection, then viral media was filtered (0.45μM or 0.22μM, Sartorius) before the target cell population was transduced at a MOI < 1. Cells were incubated with virus for 48 hours prior media being exchanged for antibiotic containing complete media. Standard antibiotic concentrations were 140μg/mL hygromycin (ThermoFisher Scientific #10687010), 1μg/mL Puromycin (Sigma; #P8833) or 10μg/mL Blasticidin S (Sigma; CAT#15205).
+
+Generation of CRISPR knockout or knockdown cell lines. Generation of CRISPR knockout guides and plasmids against FIH has been previously described67. These guides were transfected into HEK293T cells and with PEI at a 3μg:1μg ratio then clonally isolated as above. Knockouts were confirmed with PCR amplification and sanger sequencing coupled with CRISPR-ID68. FIH knockouts were selected via serial dilution and confirmation of knockout by sequencing and T7E1 assay. The VHL sgRNA guides were selected from the Dolcetto CRISPRi library69 with BsmBI compatible overhangs (Supplementary Table 3). These oligos were annealed, phosphorylated then ligated into BsmBI-digested pXPR050 (Addgene#9692), generating XPR-050-VHL. Monoclonal HEK293T LV-REPORT-12xHRE cell lines were transduced with the XPR-050-sgVHL virus, and stable cell lines selected with Puromycin. Subsequently, LV-SFFVp-CRISPRoffv2.1-IRES-BlastR or LV-EF1a-CRISPRoffv2.1-IRES-BlastR virus was infected into HEK293T LV-REPORT-12xHRE/XPR-050-sgVHL stable cells and selected with Blasticidin S (15μg/ml) for 5 days. FACS was used to assess activation of the dFLASH-HRE reporter in parental (dFLASH-HRE/sgVHL) or CRISPRoffv2.1 expressing cells at day 5 or day 10 after Blasticidin S addition.
+
+CRISPR knock-in of tags to endogenous HIF-1α and HIF-2α. CRISPR targeting constructs clones targeting adjacent to the endogenous HIF-1α and HIF-2α stop codons70. Constructs were cloned into px330 by ligating annealed and phosphorylated oligos with BbsI digested px330, using hHIF-1α and hHIF-2α CTD sgRNA (Supplementary Table 3). Knock-in of HA-3xFlag epitopes into the endogenous HIF-1α or HIF-2α loci in HEK293T cells was achieved by transfection with 0.625 μg of pNSEN, 0.625μg of pEFIREs-puro6, 2.5μg of px330-sgHIF-α CTD, and 1.25μg of ssDNA HDR template oligo containing flanking homology to CRISPR targeting site the tag insertion and a PAM mutant into ~0.8×10^6 cells using PEI (3:1). 48 hours after transfection, the medium was removed from cells and replaced with fresh medium supplemented with 2 μg/ml puromycin for 48 hours and the cell medium was changed
+to fresh medium without puromycin. 48 hours later cells were seeded by limiting dilution into 96-well plates at an average of 0.5 cells/well. Correct integration was identified by PCR screening using HIF-1\( \alpha \) and HIF-2\( \alpha \) gDNA screening primers (Supplementary Table 4). Positive colonies reisolated as single colonies by limiting dilution. Isolated HIF-1\( \alpha \) and HIF-2\( \alpha \) tag insertions were confirmed by PCR, sanger sequencing and western blotting.
+
+High Content Imaging (HCl). Cells were routinely seeded at 1x10\(^4\) to 5x10\(^4\) cells per well in black walled clear bottom 96 well plates (Costar Cat#3603), unless otherwise stated. Cell populations were imaged in media at the designated time points at 10x magnification and 2x2 binning using the ArrayScan™ XTI High Content Reader (ThermoFisher). Tomato MFI and EGFP MFI was imaged with an excitation source of 560/25nm and 485/20nm respectively. Individual nuclei were defined by nuclear EGFP expression, nuclear segmentation and confirmed to be single cells by isodata thresholding. Nuclei were excluded from analysis when they couldn’t be accurately separated from neighbouring cells and background objects, cells on image edges and abnormal nuclei were also excluded. EGFP and Tomato intensity was then measured for each individual nucleus from at least 2000 individual nuclei per well. Fixed exposure times were selected based on 10-35% peak target range. Quantification of the images utilised HCS Studio™ 3.0 Cell Analysis Software (ThermoFisher). For assessment of high throughput robustness of each individual reporting line in a high throughput setting (HTS-HCl), replicate 96 well plates were seeded for the HIF (10 plates), PGR (5 plates) and synFIH (3 plates) monoclonal reporter lines and imaged as above at 48 hours. For the HIF line, each plate had 6 replicates per treatment (vehicle or DMOG) per plate. For the PGR, 24 replicates per treatment, either vehicle or R5020 per plate were present with edge wells excluded. 24 replicates per treatment were also used for synFIH, with system robustness assessed between the DOX/DMSO and DOX/DMOG treatment groups. Z' and fold change (FC) for the Tomato/EGFP ratio for each individual plate was then calculated as per \(^{34}\):
+
+\[
+Z' = 1 - \frac{(3\sigma_{c+} - 3\sigma_{c-})}{|\mu_{c+} - \mu_{c-}|}
+\]
+
+Z' for every plate across each system was confirmed to be >0.5. Overall robustness of each system is the average of every individual Z' and FC for each system. For temporal high content imaging, HIF, PGR and synHIF lines were seeded in plates and treated with requisite ligands immediately prior to HCl. Four treatment replicates per plate were used to assess the polyclonal population. 4 treatments per plate were used to assess the synFIH monoclonal (DOX, DMSO, DOX/DMSO, DOX/DMOG), with 100ng/\(\mu\)L Doxycycline utilised, and 8 treatments per plate (vehicle, DMOG or R5020) were used to assess the PGR and HIF monoclonal lines. Plates were humidified and maintained at 37°C, 5% CO\(_2\) throughout the imaging experiment. Plates were then imaged every 2 hours for 40-48 hours. At every timepoint, a minimum 2000 nuclei were resampled from each well population.
+
+T47D mcdFLASH-PGR R5020 Dose response curve EC50 calculation. T47D mcdFLASH-PGR cells were treated with increasing doses of 0.01-100nM R5020 and quantified by HCl after 48hrs. Tomato/GFP values were min/max normalised (\(x'\) =
+\[
+\frac{(x-x_{min})}{(x_{max}-x_{min})}
+\]
+within each experiment (n = 3) and the EC50 constant and curve fitted using the drc R package from \(^{71}\).
+
+Bimodal small molecule screen to identify activators or inhibitors of the hypoxic response pathway. Library of natural and synthetic compounds was supplied by Prof. Ronald Quinn and Compounds Australia, available by request. 5mM of each of the 1595 compounds were spotted in 1\( \mu \)L DMSO into Costar Cat#3603 plates and stored at -80°C prior to screening. Plates were warmed to 37°C prior to cell addition. Monoclonal HIF HEK293T reporter cells were seeded at 0.5x10^4 cells per well across 20 Costar Cat#3603 plates pre-spiked with 5mM of compound in 1uL of DMSO in 100uL. On each plate, 4 wells were treated with matched DMSO amounts to compound wells as were four 1mM DMOG controls. Plates were then imaged using HCl (described above) at 36 hrs or 24 hours for reporter activation. Wells were then treated with 100uL of 2mM DMOG (for 1mM DMOG final, 200uL media final). 4 vehicle and 8 DMOG-treated controls (excluding the initial controls from the activator screen) were used for the inhibitor screen. Cells were imaged again 36 hours (Screen 1) or 24 hours (Screen 2) after treatment with 1mM DMOG in the compound wells. Data was Z scored and control wells were used to establish gating for abnormal expression of Tomato and EGFP fluorophores. For the activator screen, compounds within +/- 2SD EGFP MFI of vehicle wells were counted as having unchanged transcriptional effects. Compounds with Tomato/EGFP ratio greater than +2SD of vehicle controls was counted as a putative hit. For the inhibitor screen, compounds within +/- 2SD EGFP MFI of DMOG controls were counted as having unchanged GFP expression and Compounds with Tomato/EGFP ratio lower than -2SD from the DMOG control were considered putative inhibitors. To correct for false positives within each screen, Z scored compounds were converted to their respective P score and adjusted with a \(^{72}\) correction. Pearson correlations were then used to compare compound expression between screens with the base R package (4.4.0). Putative activators and inhibitors identified in the screens were re-spotted at 1mM, 2.5mM and 5mM in 1\( \mu \)L of DMSO in Costar Cat#3603 96 well trays. Activators were rescreened by HCl after 24 hours against 1x10^4 cells HIF reporter monoclonals in biological duplicate against with vehicle and 1mM DMOG controls in 100\( \mu \)L. Inhibitors were rescreened by HCl after 24 hours in duplicate against 1x10^4 cells HIF reporter monoclonals with 1mM DMOG to compound wells. Final compound concentrations were 10\( \mu \)M, 25\( \mu \)M and 50\( \mu \)M respectively and Tomato MFI and Tomato/EGFP ratio for each compound was assessed. dFLASH Bimodal high throughput screen details can be found in Supplementary Table 5.
+
+Reverse Transcription and Real Time PCR. Cells were seeded in 60mm dishes at 8x10^4 cells per vessel overnight before treatment for 48 hours with 1mM DMOG or 0.1% DMSO. Cells were lysed in Trizol (Invitrogen), and RNA was purified with Qiagen RNAEasy Kit, DNaseI treated and reverse transcribed using M-MLV reverse transcriptase (Promega). cDNA was then diluted for real time PCR. Real-time PCR used primers specific for *HIF-1α*, and human RNA Polymerase 2 (*POLR2A* (Supplementary Table 4). All reactions were done on a StepOne Plus Real-time PCR machine utilising SYBER Green, and data analysed by 'QGene' software. Results are
+normalised to POLR2A expression. RT-qPCR was performed in triplicate and single amplicons were confirmed via melt curves.
+
+Flow cytometry analysis and sorting (FACS). Prior to flow cytometry, cells were trypsinised, washed in complete media and resuspended in resuspended in flow cytometry sort buffer (Ca^{2+}/Mg^{2+}-free PBS, 2%FBS, 25mM HEPES pH 7.0) for cell sorting) prior to cell sorting or flow cytometry analysis buffer (Ca^{2+}/Mg^{2+} free PBS, 2%FBS, 1mM EDTA, 25mM HEPES pH 7.0) for analysis followed by filtration through a 40μM nylon cell strainer (Corning Cat#352340. Cell populations were kept on ice prior to sorting. Flow cytometry was performed either using the BD Biosciences BD LSRFortessa or the BD Biosciences FACS ARIA2 sorter within a biosafety cabinet and aseptic conditions, using an 85μM nozzle. Cell populations were gated by FSC-W/FSC-H, then SSC-W/SSC-H, followed by SSC-A/FSC-A to gate cells. EGFP fluorescence was measured by a 530/30nm detector, and the Tomato fluorescence was determined with the 582/15nm detector. A minimum of 10,000 cells were sorted for all FACS-based analysis. Data is presented as log_{10} intensity for both fluorophores. Tomato induction was gated from the top 1% of the negative control population. Cell counts for histograms are normalised to mode unless stated otherwise. FACS analysis was done on FlowJo™ v10.9.1 software (BD Life Sciences)73.
+
+Time Lapse Spinning Disc Confocal Microscopy. HEK293T mcdFLASH-HIF and T47D mcdFLASH-PGR cells were seeded at 1x10^5 or 5x10^5 cells per dish respectively, onto 50μg/mL poly-D-lysine μ-Dish 35 mm, high Glass Bottom dishes (Ibidi, #81158) in FluoroBrite DMEM (Gibco, A1896701)/10% FBS/ 1% Pens/1% Glutamax/10mM HEPES pH7.9 and incubated overnight at 37°C with 5% CO2 prior imaging. Cells were treatment with either 0.5mM DMOG (mcdFLASH-HIF) or 100nM R05020 (mcdFLASH-PGR) immediately prior to imaging with a CV100 cell voyager spinning disk confocal Tomato (561 nm, 50% laser, 400ms exposure and 20% gain) and EGFP (488 nm, 50% laser, 400ms exposure and 20% gain) fluorescence for 48 hours post treatment with 15min imaging intervals. Images were captured at 40x with an objective lens with a ~30μm Z stack across multiple fields of view. Maximum projected intensity images were exported to Image J for analysis and movie creation.
+
+Cell Lysis and Immunoblotting. Cells were washed in ice-cold PBS and lysates were generated by resuspending cells in either cell lysis buffer (20mM HEPES pH 8.0, 420mM NaCl, 0.5% NP-40, 25% Glycerol, 0.2mM EDTA, 1.5mM MgCl_2, 1mM DTT, 1x Protease Inhibitors (Sigma)) (Supp Figure 4) or urea lysis buffer (6.7M Urea, 10mM Tris-Cl pH 6.8, 10% glycerol, 1% SDS, 1mM DTT) (Figure 6, Supp Figure 8, 9). Quantification of protein levels was done by Bradford Assay (Bio-Rad). Lysates were separated on a 7.5% SDS-PAGE gel and transferred to nitrocellulose via TurboBlot (Bio-Rad). Primary Antibodies used were anti-HIF1α (BD Biosciences #), anti-HA (HA.11, Biolegend #16B12), anti-Tubulin (Serotec #MCA78G), anti-GAPDH (Sigma #G8796), anti-ARNT (Proteintech #14105-1-AP). Primary antibodies were detected using horseradish peroxidase conjugated secondary antibodies (Pierce Bioscience #). Blots were visualised via chemiluminescence and developed with Clarity Western ECL Blotting substrates (Bio-Rad).
+In vitro iron chelation activity assay. Chelation of iron for RQ200674 was measured by a protocol adapted from \( ^{74} \) for use in 96 well plate format. 0.1mM FeSO\(_4\) (50\( \mu \)L) and 50\( \mu \)L of RQ200674, Dipyridyl (positive control) or DMOG solutions were incubated for 1hr at room temperature prior to addition of 100\( \mu \)L of 0.25mM Ferrozine (Sigma) and incubated for a further 10 minutes. Absorbance was measured at 562nM. Chelation activity was quantified as:
+
+\[
+Chelation\ activity = \frac{(A_{control} - A_x)}{A_{control}} \times 100
+\]
+
+Where \( A_{control} \) is absorbance of control reactions without RQ200674, DP or DMOG and \( A_x \) is absorbance of solutions with compound.
+
+Statistical Analysis. All data in graphs were presented as a mean \( \pm \) sem unless otherwise specified. Significance was calculated by a Two-Way ANOVA with Tukey multiple comparison or unpaired t-test with Welches correction where appropriate using Graphpad PRISM (version 9.0.0). All statistical analysis is from three independent biological replicates
+
+Figure Creation. Schematics and diagrams were created with BioRender (BioRender.com) and graphs were made either with ggplot package in R\(^{75}\) and GraphPad PRISM (version 9.0.0).
+
+Data Availability. Source data are provided with this paper. Additional data, including full construct sequences, are available from corresponding authors upon request. Constructs not available on Addgene can be requested from corresponding authors.
+
+Acknowledgements. We thank Nicholas Smith, Alexander Pace, and members of our laboratories for critical feedback and helpful discussions. We also wish to acknowledge Adelaide Microscopy and the AHMS and SAHMRI Flow Cytometry facilities for technical assistance. We acknowledge Compounds Australia (www.compoundsaustralia.com) for their provision of specialized compound management and logistics research services to the project. This work was supported by Australian Government Research Training Scholarships (T.P.A, A.E.R), The Emeritus Professor George Rodgers AO Supplementary Scholarship (T.P.A, A.E.R). The Playford Memorial Trust Thyne Reid Foundation Scholarship (A.E.R). The George Fraser Supplementary Scholarship (A.E.R), The University of Adelaide Biochemistry Trust Fund (D.J.P. and M.L.W) and the Bill and Melinda Gates Foundation Contraceptive Discovery Program [OPP1771844] (D.C.B, D.L.R).
+
+Author contributions. Study was initially conceived by D.C.B and M.L.W. T.P.A, D.C.B., A.E.R designed and performed experiments. T.P.A, D.C.B., M.L.W, M.L. and R.J.Q. performed and analysed the bimodal screening campaign. M.R. and A.E.R. derived FIH KO cell line. T.P.A, D.C.B and M.L.W wrote the manuscript with input from all authors. Work was supervised by D.J.P, D.L.R. & M.L.W.
+
+Source Data. Source data for figures is available with this manuscript.
+
+Competing interests. The authors declare no competing interests.
+Correspondence and requests for materials. Should be addressed to David C. Bersten.
+Supplementary Table 1: Synthetic toolkit for generation of reporter cell lines
+
+<table>
+  <tr>
+    <th>Deposit Name:</th>
+    <th>Availability</th>
+    <th>Purpose</th>
+  </tr>
+  <tr>
+    <th colspan="3">Dual fluorescent reporter constructs:</th>
+  </tr>
+  <tr>
+    <td>pLV-REPORT(EF1a)</td>
+    <td>Addgene #172326</td>
+    <td>Reporter with mnucTomato and EF1a downstream promoter</td>
+  </tr>
+  <tr>
+    <td>pLV-REPORT(EF1a)-TTN</td>
+    <td>Addgene #172327</td>
+    <td>Reporter with mnucTomato-HSVtk-2A-NeoR and EF1a downstream promoter</td>
+  </tr>
+  <tr>
+    <td>pLV-REPORT(PGK)</td>
+    <td>Addgene #172328</td>
+    <td>Reporter with mnucTomato-HSVtk-2A-NeoR and PGK downstream promoter</td>
+  </tr>
+  <tr>
+    <td>pLV-REPORT(PGK/CMV)</td>
+    <td>Addgene #172330</td>
+    <td>Reporter with mnucTomato-HSVtk-2A-NeoR and PGK/CMV downstream promoter</td>
+  </tr>
+  <tr>
+    <td>12xHRE-pLV-Report-EF1a</td>
+    <td>Addgene: #172333</td>
+    <td>Reporter with HRE enhancer</td>
+  </tr>
+  <tr>
+    <td>12xHRE-pLV-REPORT(PGK)</td>
+    <td>Addgene #172334</td>
+    <td>Reporter with HRE enhancer</td>
+  </tr>
+  <tr>
+    <td>12xHRE-pLV-REPORT(PGK/CMV)</td>
+    <td>Addgene #172335</td>
+    <td>Reporter with HRE enhancer</td>
+  </tr>
+  <tr>
+    <td>PREcat-pLV-REPORT(PGK/CMV)</td>
+    <td>By Request</td>
+    <td>Reporter with a PR-responsive concatemer, with enhancers from 5 target genes, containing 6 PR response elements.</td>
+  </tr>
+  <tr>
+    <td>5xGRE-pLV-REPORT(PGK/CMV)</td>
+    <td>Addgene #172336</td>
+    <td>Reporter with GRE enhancer</td>
+  </tr>
+  <tr>
+    <td>12xHRE-pLV-REPORT(EF1a)</td>
+    <td>By Request</td>
+    <td>Reporter with HRE</td>
+  </tr>
+  <tr>
+    <td>12xHRE- pLV-REPORT(EF1a)-tdnucTomato</td>
+    <td>By Request</td>
+    <td>Reporter with tdnucTomato and EF1a downstream promoter</td>
+  </tr>
+  <tr>
+    <th colspan="3">Protein expression constructs:</th>
+  </tr>
+  <tr>
+    <td>pLV-TET2Puro</td>
+    <td>By Request</td>
+    <td>Doxycycline-inducible expression vector</td>
+  </tr>
+  <tr>
+    <td>pLV-TET2BlastR</td>
+    <td>By Request</td>
+    <td>Doxycycline-inducible expression vector</td>
+  </tr>
+  <tr>
+    <td>pLV-TET2nucTomato</td>
+    <td>By Request</td>
+    <td>Doxycycline-inducible expression vector</td>
+  </tr>
+  <tr>
+    <td>pLV-TET2Puro-gal4DBD-miniVPR-HA</td>
+    <td>Addgene #207171</td>
+    <td>Doxycycline-inducible expression vector for GAL4DBD-miniVPR</td>
+  </tr>
+  <tr>
+    <td>pLV-TET2Puro-gal4DBD-HIFCAD</td>
+    <td>Addgene #207173</td>
+    <td>Doxycycline-inducible expression vector for GAL4DBD-HIFCAD (727-826) with Myc tag</td>
+  </tr>
+  <tr>
+    <td>pEF-IRES-puro6 gal4DBD-HIFCAD myc tag</td>
+    <td>Addgene #207171</td>
+    <td>Constitutively expresses GAL4DBD-HIFCAD (727-826) with Myc tag</td>
+  </tr>
+  <tr>
+    <td>pEF-IRES-puro6 gal4DBD-HIFCAD pGalQ linker</td>
+    <td>Addgene #207172</td>
+    <td>Constitutively expresses GAL4DBD-HIFCAD (727-826) with Myc tag</td>
+  </tr>
+  <tr>
+    <td>pENTR1a-CRISPRoffv2.1</td>
+    <td>Addgene #207174</td>
+    <td>Lentiviral expression vector for CRISPRoffv2.1 with BFP tag</td>
+  </tr>
+  <tr>
+    <td>pLV-Egl-NeoR</td>
+    <td>Addgene #207175</td>
+    <td>Gateway-compatible lentiviral expression plasmid with Neomycin resistance</td>
+  </tr>
+  <tr>
+    <td>pLV-Egl-BlasR</td>
+    <td>Addgene #207176</td>
+    <td>Gateway-compatible lentiviral expression plasmid with Blasticidin resistance</td>
+  </tr>
+  <tr>
+    <td>pLV-Egl-HygroR</td>
+    <td>Addgene #207177</td>
+    <td>Gateway-compatible lentiviral expression plasmid with Hygromycin resistance</td>
+  </tr>
+  <tr>
+    <td>pLV-Egl-ZeoR</td>
+    <td>Addgene #207178</td>
+    <td>Gateway-compatible lentiviral expression plasmid with Zeocin resistance</td>
+  </tr>
+</table>
+Supplementary Table 2: Sequences for enhancer cloning
+
+PRECat (G-block)
+gaattacaaaacaattacaaaatttattcagatTGCATGCCCTGCTTACATAAGGAAGTACAGAGTGTA CCAAAACAGCAGACCCAAAAAAAAGCCTGAAAATGTGAGAACCACCAAACACTGTACAGCTTTGATT TCAGGAAGCAAACACTGAGGACGCAAGCCGCTTCTCATGGAAATAATACATCTGTTCGCCACAAGT GACGTTAGCTTCCAGACTGTGCAAGAGTGCACTTCACCCAGTGTGTGTCTATCATGGTCAC ACAGTGTTCCTTTCCGTGGTCACATCTGTGTCCACATTTCCCTTTTGATGGGAACAAAGCAGT CATGTTAGGAAGGGAAGGACACCGGTGTTTAAATCACACAATCCATGGACAGCCGTGGGCATC CAGTAATGCGCTGGAATGAGTCAAGAAGGCTTGCCCAGTTTTCACTAAAGAGCTGCGAGGACA GCCTGTCCCTGTTACAACCACCCACGCTCCGTTGAGGCGCGCCAGCTTTTAGGCGTGACG GTGGGCGCCTATAAAAAGC
+
+5xGRE
+GGTACCAGCTTGATGCCCTGCGAGTCGGAGTACTGTCTCTCCGAGCGGAGTACTGTCTCTCCGA GCGGAGTACTGTCTCTCCGAGCGGAGTACTGTCTCTCCGAGCGGAGTACTGTCTCTCCGAGCGG AGAC
+
+Supplementary Table 3: Index of all sgGuide oligos used
+
+<table>
+  <tr>
+    <th></th>
+    <th>Upper (5'-3')</th>
+    <th>Lower (5'-3')</th>
+  </tr>
+  <tr>
+    <td>VHL Knockdown sgGuide</td>
+    <td>CACCGCCGGGTGGTCTGGATCGCGG</td>
+    <td>AAACCCGCGATCCAGACCACCCCGGC</td>
+  </tr>
+  <tr>
+    <td>hHIF-1α CTD sgRNA</td>
+    <td>CACCGTGAAGAATTACTCAGAGCTT</td>
+    <td>AAACAAGCTCTGAGTAATTCTTCA</td>
+  </tr>
+  <tr>
+    <td>hHIF-2α CTD sgRNA</td>
+    <td>CACCGCCTCCTCAGAGCCCTGGACC</td>
+    <td>AAACGGTCCAGGGCTCTGAGGAGGC</td>
+  </tr>
+</table>
+
+Supplementary Table 4: Primer sets for qPCR and PCR confirmation
+
+<table>
+  <tr>
+    <th></th>
+    <th>Forward (5'-3')</th>
+    <th>Reverse (5'-3')</th>
+  </tr>
+  <tr>
+    <td><i>qPCR HIF-1α</i></td>
+    <td>TATGAGCCAGAAGAACTTTT AGGC</td>
+    <td>CACCTCTTTTTGGCAAGCATCCTG</td>
+  </tr>
+  <tr>
+    <td><i>qPCR PolR2a</i></td>
+    <td>GCACCATCAAGAGAGTGCA G</td>
+    <td>GGGTATTTTGATACCACCCCTCT</td>
+  </tr>
+  <tr>
+    <td>HIF-1α gDNA primers</td>
+    <td>GGCAATCAATGGATGAAAGT GGATT</td>
+    <td>GCTACTGCAATGCAATGGTTTAA AT</td>
+  </tr>
+  <tr>
+    <td>HIF-2α gDNA primers:</td>
+    <td>ACCAACCCCTTCTTTTCAGGCA TGGC</td>
+    <td>GCTTGGTGACCTGGGCAAGTCT GC</td>
+  </tr>
+</table>
+Beitz, A. M., Oakes, C. G. & Galloway, K. E. Synthetic gene circuits as tools for drug discovery. Trends Biotechnol 40, 210-225 (2022). https://doi.org/10.1016/j.tibtech.2021.06.007
+
+Bock, C. et al. High-content CRISPR screening. Nature Reviews Methods Primers 2 (2022). https://doi.org/10.1038/s43586-021-00093-4
+
+Lee, T. I. & Young, R. A. Transcriptional regulation and its misregulation in disease. Cell 152, 1237-1251 (2013). https://doi.org/10.1016/j.cell.2013.02.014
+
+Bersten, D. C., Sullivan, A. E., Peet, D. J. & Whitelaw, M. L. bHLH-PAS proteins in cancer. Nat Rev Cancer 13, 827-841 (2013). https://doi.org/10.1038/nrc3621
+
+Darnell, J. E., Jr. Transcription factors as targets for cancer therapy. Nat Rev Cancer 2, 740-749 (2002). https://doi.org/10.1038/nrc906
+
+Sahu, B. et al. Sequence determinants of human gene regulatory elements. Nat Genet 54, 283-294 (2022). https://doi.org/10.1038/s41588-021-01009-4
+
+Tycko, J. et al. High-Throughput Discovery and Characterization of Human Transcriptional Effectors. Cell 183, 2020-2035 e2016 (2020). https://doi.org/10.1016/j.cell.2020.11.024
+
+DelRosso, N. et al. Large-scale mapping and mutagenesis of human transcriptional effector domains. Nature (2023). https://doi.org/10.1038/s41586-023-05906-y
+
+Ortmann, B. M. et al. The HIF complex recruits the histone methyltransferase SET1B to activate specific hypoxia-inducible genes. Nature Genetics 53, 1022-1035 (2021). https://doi.org/10.1038/s41588-021-00887-y
+
+Tan, X., Letendre, J. H., Collins, J. J. & Wong, W. W. Synthetic biology in the clinic: engineering vaccines, diagnostics, and therapeutics. Cell 184, 881-898 (2021). https://doi.org/10.1016/j.cell.2021.01.017
+
+Choe, J. H. et al. SynNotch-CAR T cells overcome challenges of specificity, heterogeneity, and persistence in treating glioblastoma. Science Translational Medicine 13 (2021).
+
+Allen, G. M. et al. Synthetic cytokine circuits that drive T cells into immune-excluded tumors. Science 378, 1186-+ (2022). https://doi.org/ARTN eaba1624 10.1126/science.aba1624
+
+Hernandez-Lopez, R. A. et al. T cell circuits that sense antigen density with an ultrasensitive threshold. Science 371, 1166-+ (2021). https://doi.org/10.1126/science.abc1855
+
+Roybal, K. T. et al. Engineering T Cells with Customized Therapeutic Response Programs Using Synthetic Notch Receptors. Cell 167, 419-+ (2016). https://doi.org/10.1016/j.cell.2016.09.011
+
+Hasle, N. et al. High-throughput, microscope-based sorting to dissect cellular heterogeneity. Mol Syst Biol 16, e9442 (2020). https://doi.org/10.15252/msb.20209442
+
+Tchasovnikarova, I. A., Marr, S. K., Damle, M. & Kingston, R. E. TRACE generates fluorescent human reporter cell lines to characterize epigenetic pathways. Mol Cell (2021). https://doi.org/10.1016/j.molcel.2021.11.035
+
+Adamson, B. et al. A Multiplexed Single-Cell CRISPR Screening Platform Enables Systematic Dissection of the Unfolded Protein Response. Cell 167, 1867-1882 e1821 (2016). https://doi.org/10.1016/j.cell.2016.11.048
+Singhal, R. & Shah, Y. M. Oxygen battle in the gut: Hypoxia and hypoxia-inducible factors in metabolic and inflammatory responses in the intestine. J Biol Chem 295, 10493-10505 (2020). https://doi.org/10.1074/jbc.REV120.011188
+
+Wegiel, B., Vuerich, M., Daneshmandi, S. & Seth, P. Metabolic Switch in the Tumor Microenvironment Determines Immune Responses to Anti-cancer Therapy. Front Oncol 8, 284 (2018). https://doi.org/10.3389/fonc.2018.00284
+
+Triner, D. & Shah, Y. M. Hypoxia-inducible factors: a central link between inflammation and cancer. J Clin Invest 126, 3689-3698 (2016). https://doi.org/10.1172/JCI84430
+
+Epstien, A. et al. C. elegans EGL-9 and Mammalian Homologs Define a Family of Dioxygenases that Regulate HIF by Prolyl Hydroxylation. Cell 107 (2001).
+
+Lando, D. et al. FIH-1 is an asparaginyl hydroxylase enzyme that regulates the transcriptional activity of hypoxia-inducible factor. Genes Dev 16, 1466-1471 (2002). https://doi.org/10.1101/gad.991402
+
+Tian, Y.-M. et al. Differential Sensitivity of Hypoxia Inducible Factor Hydroxylation Sites to Hypoxia and Hydroxylase Inhibitors. Journal of Biological Chemistry 286, 13041-13051 (2011). https://doi.org/10.1074/jbc.m110.211110
+
+Razorenova, O. V., Ivanov, A. V., Budanov, A. V. & Chumakov, P. M. Virus-based reporter systems for monitoring transcriptional activity of hypoxia-inducible factor 1. Gene 350, 89-98 (2005). https://doi.org/10.1016/j.gene.2005.02.006
+
+Villemure, J. F., Savard, N. & Belmaaza, A. Promoter suppression in cultured mammalian cells can be blocked by the chicken beta-globin chromatin insulator 5'HS4 and matrix/scaffold attachment regions. J Mol Biol 312, 963-974 (2001). https://doi.org/10.1006/jmbi.2001.5015
+
+Emmerman, M. & Temin, H. Comparison of promoter suppression in avian and murine retrovirus vectors. Nucleic Acids Res 14 (1986).
+
+O’Connell, R. W. et al. Ultra-high throughput mapping of genetic design space (Cold Spring Harbor Laboratory, 2023).
+
+Lando, D., Peet, D. J., Dean A. Whelan, Jeffery J. Gorman & Whitelaw, M. L. Asparagine Hydroxylation of the HIF Transactivation Domain: A Hypoxic Switch. Science 295 (2002).
+
+Vora, S. et al. Rational design of a compact CRISPR-Cas9 activator for AAV-mediated delivery. bioRxiv, 298620 (2018). https://doi.org/10.1101/298620
+
+Lydon, J. P. et al. Mice lacking progesterone receptor exhibit pleiotropic reproductive abnormalities. Genes & Development 9, 2266-2278 (1995). https://doi.org/10.1101/gad.9.18.2266
+
+Dinh, D. T. et al. Tissue-specific progesterone receptor-chromatin binding and the regulation of progesterone-dependent gene expression. Scientific Reports 9 (2019). https://doi.org/10.1038/s41598-019-48333-8
+
+Grimm, S. L., Hartig, S. M. & Edwards, D. P. Progesterone Receptor Signaling Mechanisms. J Mol Biol 428, 3831-3849 (2016). https://doi.org/10.1016/j.jmb.2016.06.020
+
+Giannoukos, G., Szapary, D., Smith, C. L., Meeker, J. E. & Simons, S. S., Jr. New antiprogestins with partial agonist activity: potential selective progesterone receptor modulators (SPRMs) and probes for receptor- and coregulator-induced changes in progesterone receptor induction properties. Mol Endocrinol 15, 255-270 (2001). https://doi.org/10.1210/mend.15.2.0596
+Zhang, J.-H., Chung, T. & Oldenburg, K. A Simple Statistical Parameter for Use in Evaluation and Validation of High Throughput Screening Assays. Journal of Biomolecular Screening **4** (1999).
+
+Kampmann, M. CRISPRi and CRISPRa Screens in Mammalian Cells for Precision Biology and Medicine. ACS Chem Biol **13**, 406-416 (2018). https://doi.org/10.1021/acschembio.7b00657
+
+Jaakkola, P. *et al.* Targeting of HIF-a to the von Hippel–Lindau Ubiquitylation Complex by O2-Regulated Prolyl Hydroxylation. Science **292** (2001).
+
+Appelhoff, R. J. *et al.* Differential function of the prolyl hydroxylases PHD1, PHD2, and PHD3 in the regulation of hypoxia-inducible factor. *J Biol Chem* **279**, 38458-38465 (2004). https://doi.org/10.1074/jbc.M406026200
+
+Chen, N. *et al.* Roxadustat Treatment for Anemia in Patients Undergoing Long-Term Dialysis. *N Engl J Med* **381**, 1011-1022 (2019). https://doi.org/10.1056/NEJMoa1901713
+
+Cai, Z., Luo, W., Zhan, H. & Semenza, G. L. Hypoxia-inducible factor 1 is required for remote ischemic preconditioning of the heart. *Proc Natl Acad Sci U S A* **110**, 17462-17467 (2013). https://doi.org/10.1073/pnas.1317158110
+
+Masoud, G. N. & Li, W. HIF-1alpha pathway: role, regulation and intervention for cancer therapy. *Acta Pharm Sin B* **5**, 378-389 (2015). https://doi.org/10.1016/j.apsb.2015.05.007
+
+Semenza, G. L. HIF-1 mediates metabolic responses to intratumoral hypoxia and oncogenic mutations. *J Clin Invest* **123**, 3664-3671 (2013). https://doi.org/10.1172/JCI67230
+
+Semenza, G. L. Pharmacologic Targeting of Hypoxia-Inducible Factors. *Annual Review of Pharmacology and Toxicology* **59**, 379-403 (2019). https://doi.org/10.1146/annurev-pharmtox-010818-021637
+
+Keith, B., Johnson, R. S. & Simon, M. C. HIF1α and HIF2α: sibling rivalry in hypoxic tumor growth and progression. *Nat Rev Cancer* **12**, 9-22 (2012).
+
+Bracken, C. P. *et al.* Cell-specific regulation of hypoxia-inducible factor (HIF)-1alpha and HIF-2alpha stabilization and transactivation in a graded oxygen environment. *J Biol Chem* **281**, 22575-22585 (2006). https://doi.org/10.1074/jbc.M600288200
+
+Ran, F. A. *et al.* Genome engineering using the CRISPR-Cas9 system. *Nature Protocols* **8**, 2281-2308 (2013). https://doi.org/10.1038/nprot.2013.143
+
+Huang, L. *et al.* Inhibitory action of Celastrol on hypoxia-mediated angiogenesis and metastasis via the HIF-1α pathway. *International Journal of Molecular Medicine* **27** (2011). https://doi.org/10.3892/ijmm.2011.600
+
+Ma, J. *et al.* Celastrol inhibits the HIF-1α pathway by inhibition of mTOR/p70S6K/eIF4E and ERK1/2 phosphorylation in human hepatoma cells. *Oncology Reports* **32**, 235-242 (2014). https://doi.org/10.3892/or.2014.3211
+
+Shang, F.-F. *et al.* Design, synthesis of novel celastrol derivatives and study on their antitumor growth through HIF-1α pathway. *European Journal of Medicinal Chemistry* **220**, 113474 (2021). https://doi.org/10.1016/j.ejmech.2021.113474
+
+Srinivasan, B., Johnson, T. E. & Xing, C. Chalcone-based inhibitors against hypoxia-inducible factor 1—Structure activity relationship studies. *Bioorganic &amp; Medicinal Chemistry Letters* **21**, 555-557 (2011). https://doi.org/10.1016/j.bmcl.2010.10.063
+Wan, C. et al. Genome-scale CRISPR-Cas9 screen of Wnt/β-catenin signaling identifies therapeutic targets for colorectal cancer. Science Advances 7, eabf2567 (2021). https://doi.org/10.1126/sciadv.abf2567
+
+Semesta, K. M., Tian, R., Kampmann, M., Von Zastrow, M. & Tsvetanova, N. G. A high-throughput CRISPR interference screen for dissecting functional regulators of GPCR/cAMP signaling. PLOS Genetics 16, e1009103 (2020). https://doi.org/10.1371/journal.pgen.1009103
+
+Adamson, B. et al. A Multiplexed Single-Cell CRISPR Screening Platform Enables Systematic Dissection of the Unfolded Protein Response. Cell 167, 1867-1882.e1821 (2016). https://doi.org/10.1016/j.cell.2016.11.048
+
+Potting, C. et al. Genome-wide CRISPR screen for PARKIN regulators reveals transcriptional repression as a determinant of mitophagy. Proc Natl Acad Sci U S A 115, E180-E189 (2018). https://doi.org/10.1073/pnas.1711023115
+
+Ilegems, E. et al. HIF-1a inhibitor PX-478 preserves pancreatic Beta cell function in diabetes. Science Translational Medicine 14 (2022).
+
+Koh, M. Y. et al. Molecular mechanisms for the activity of PX-478, an antitumor inhibitor of the hypoxia-inducible factor-1α. Molecular Cancer Therapeutics 7, 90-100 (2008). https://doi.org/10.1158/1535-7163.mct-07-0463
+
+Welsh, S., Williams, R., Kirkpatrick, L., Paine-Murrieta, G. & Powis, G. Antitumor activity and pharmacodynamic properties of PX-478, an inhibitor of hypoxia-inducible factor-1A. Molecular Cancer Therapeutics 3 (2004). https://doi.org/https://doi.org/10.1158/1535-7163.233.3.3
+
+Xia, M. et al. Identification of small molecule compounds that inhibit the HIF-1 signaling pathway. Mol Cancer 8, 117 (2009). https://doi.org/10.1186/1476-4598-8-117
+
+Yin, J.-A. et al. Robust and Versatile Arrayed Libraries for Human Genome-Wide CRISPR Activation, Deletion and Silencing. bioRxiv, 2022.2005.2025.493370 (2023). https://doi.org/10.1101/2022.05.25.493370
+
+Feldman, D. et al. Optical Pooled Screens in Human Cells. Cell 179, 787-799.e717 (2019). https://doi.org/10.1016/j.cell.2019.09.016
+
+Feldman, D. et al. Pooled genetic perturbation screens with image-based phenotypes. Nat Protoc 17, 476-512 (2022). https://doi.org/10.1038/s41596-021-00653-8
+
+Yan, X. et al. High-content imaging-based pooled CRISPR screens in mammalian cells. Journal of Cell Biology 220 (2021). https://doi.org/10.1083/jcb.202008158
+
+Nandagopal, N. et al. Dynamic Ligand Discrimination in the Notch Signaling Pathway. Cell 172, 869-880.e819 (2018). https://doi.org/10.1016/j.cell.2018.01.002
+
+Agarwal, V. et al. Massively parallel characterization of transcriptional regulatory elements in three diverse human cell types. bioRxiv (2023). https://doi.org/10.1101/2023.03.05.531189
+
+Gordon, M. G. et al. lentiMPRA and MPRAflow for high-throughput functional characterization of gene regulatory elements. Nat Protoc 15, 2387-2412 (2020). https://doi.org/10.1038/s41596-020-0333-5
+
+Bersten, D. C. et al. Inducible and reversible lentiviral and Recombination Mediated Cassette Exchange (RMCE) systems for controlling gene expression. PLoS One 10, e0116373 (2015). https://doi.org/10.1371/journal.pone.0116373
+Singhal, H. et al. Genomic agonism and phenotypic antagonism between estrogen and progesterone receptors in breast cancer. Sci Adv 2, e1501924 (2016). https://doi.org/10.1126/sciadv.1501924
+
+Chen, D.-Y. et al. Ankyrin Repeat Proteins of Orf Virus Influence the Cellular Hypoxia Response Pathway. Journal of Virology 91, JVI.01430-01416 (2017). https://doi.org/10.1128/jvi.01430-16
+
+Dehairs, J., Talebi, A., Cherifi, Y. & Swinnen, J. V. CRISP-ID: decoding CRISPR mediated indels by Sanger sequencing. Sci Rep 6, 28973 (2016). https://doi.org/10.1038/srep28973
+
+Sanson, K. R. et al. Optimized libraries for CRISPR-Cas9 genetic screens with multiple modalities. Nat Commun 9, 5416 (2018). https://doi.org/10.1038/s41467-018-07901-8
+
+Bersten, D. et al. Core and Flanking bHLH-PAS:DNA interactions mediate specificity and drive obesity (Cold Spring Harbor Laboratory, 2022).
+
+Ritz, C., Baty, F., Streibig, J. C. & Gerhard, D. Dose-Response Analysis Using R. PLOS ONE 10, e0146021 (2016). https://doi.org/10.1371/journal.pone.0146021
+
+Benjamini, Y. & Hochberg, Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society: Series B (Methodological) 57, 289-300 (1995). https://doi.org/https://doi.org/10.1111/j.2517-6161.1995.tb02031.x
+
+Becton, Dickenson & Company. (Ashland, OR, 2021).
+
+Wong, F. C. et al. Antioxidant, Metal Chelating, Anti-glucosidase Activities and Phytochemical Analysis of Selected Tropical Medicinal Plants. Iran J Pharm Res 13, 1409-1415 (2014).
+
+Wickham, H. in Elegant Graphics for Data Analysis VIII, 213 (Springer New York, NY, 2009).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SupplementaryTable5.docx
+• SuppVideo1.avi
+• SuppVideo2.avi
+• NewRS.pdf

06076107a582a7db1a993272bc8d0902287719a4fcc5b06c29c18d0dc685d84a/preprint/preprint.md

@@ -0,0 +1,181 @@
+Pan-cancer copy number variant analysis identifies optimized size thresholds and co-occurrence models for individualized risk-stratification
+
+David Raleigh
+david.raleigh@ucsf.edu
+
+University of California San Francisco   https://orcid.org/0000-0001-9299-8864
+Minh Nguyen
+University of California San Francisco
+William Chen
+UCSF   https://orcid.org/0000-0001-8924-5853
+Naomi Zakimi
+Univeristy of California San Francisco
+Kanish Mirchia
+Univeristy of California San Francisco   https://orcid.org/0000-0002-7371-7059
+Calixto-Hope Lucas
+Johns Hopkins University   https://orcid.org/0000-0002-8347-9592
+
+Brief Communication
+
+Keywords:
+
+Posted Date: January 11th, 2024
+
+DOI: https://doi.org/10.21203/rs.3.rs-3443805/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on July 2nd, 2025. See the published version at https://doi.org/10.1038/s41467-025-61063-y.
+Abstract
+
+Chromosome instability leading to accumulation of copy number gains or losses is a hallmark of cancer. Copy number variant (CNV) signatures are increasingly used for clinical risk-stratification, but size thresholds for defining CNVs are variable and the biological or clinical implications of CNV size heterogeneity or co-occurrence patterns are incompletely understood. Here we analyze CNV and clinical data from 565 meningiomas and 9,885 tumors from The Cancer Genome Atlas (TCGA) to develop tumor- and chromosome-specific CNV size-dependent and co-occurrence models for clinical outcomes. Our results reveal prognostic CNVs with optimized size thresholds and co-occurrence patterns that refine risk-stratification across a diversity of human cancers.
+
+Main
+
+Chromosome instability contributes to the genomic complexity of cancer\(^1\) and is implicated in tumorigenesis, progression, metastasis, and resistance to therapy\(^2-4\). As a marker of chromosome instability, CNV signatures are increasingly used for clinical risk-stratification of diverse cancer types\(^5,6\), and pan-cancer databases such as TCGA\(^7\) have been used to derive prognostic models based on CNVs\(^6,8\). There is no consensus on the optimal size threshold for defining or reporting CNVs, and CNV co-occurrence patterns that may improve risk-stratification models are incompletely understood.
+
+To test the hypothesis that size-dependent CNV models and co-occurrence patterns may improve clinical risk-stratification, CNV size-dependence was investigated in meningiomas, a tumor that is not represented in TCGA datasets but is associated with recurrent CNVs that can be used for risk-stratification\(^9,10\). Loss of chromosomes 1p, 6q, and others distinguish biologically aggressive meningiomas\(^9,10\), but published models have applied inconsistent size thresholds ranging from 5–80% of individual chromosome arms to define meningioma CNVs\(^9-12\). Using a previously described cohort of 565 meningiomas with long-term clinical outcomes data\(^11\), we used DNA methylation arrays to define CNVs ranging from individual CpG loci to entire chromosome arms (Extended Data Fig. 1). Next, we used CNVs ranging from 5–95% of each chromosome arm to generate univariate Cox proportional hazards models for postoperative local freedom from recurrence (LFFR) or overall survival (OS). These analyses revealed “size-dependent” CNVs (Fig. 1a), defined as having a maximum area under the curve (AUC) for 5-year LFFR or OS of at least 0.60 that decreased by at least 5% from the maximum AUC as CNV threshold varied (Supplementary Table 1).
+
+The implications of CNV size-dependence for meningioma risk-stratification were investigated using 2 robust models that rely on CNVs to predict postoperative meningioma LFFR. The first, integrated grade, is based on copy number losses of chromosomes 1p, 3p, 4p/q, 6p/q, 10p/q, 14q, 18p/q, and 19p/q at a uniform threshold of 50% of each chromosome arm plus *CDKN2A* loss and mitotic count from histology\(^9\). The second, integrated score, is based on copy number losses of chromosomes 1p, 6q, and 14q at a uniform threshold of 5% of each chromosome arm plus DNA methylation family\(^13\) and World
+Health Organization (WHO) histological grade10. We tested each model on our cohort of 565 meningiomas using CNV thresholds ranging from 5–95% (Fig. 1b). Integrated grade reached a maximum AUC for 5-year LFFR of 0.78 at a uniform CNV threshold of 20%, and a maximum AUC for OS of 0.77 at a uniform threshold of 30%. Integrated score reached a maximum AUC for LFFR or OS of 0.76 at a uniform CNV threshold of 5%. The performance of each model degraded with varying CNV size thresholds (Fig. 1b), suggesting that CNV size heterogeneity influences risk-stratification for the most common primary intracranial tumor14.
+
+To determine if models based on chromosome-specific CNV size thresholds could improve meningioma risk-stratification, LASSO and elastic net regularized Cox models were trained using optimized CNVs thresholds across the 565 meningiomas in our cohort (Extended Data Fig. 2). Cross-validated AUCs for 5-year LFFR or OS were 0.76 for LASSO models and 0.77–0.78 for elastic net models. CNV size-dependent models identified prognostic chromosome arms that were not included in either integrated grade or integrated score, such as gain of 1q or 17q and loss of 4p, 9p, 10q, or 12q for LFFR, and gain of 1q, 9q, or 10p and loss of 3q, 5p/q, 6p, 9p, 10q, 11p, 13q, 14q, or 18p/q for OS (Fig. 1c), many of which have been previously associated with biologically aggressive meningiomas11. There were numerous areas of focal deletion across chromosome arms with size-dependent CNVs that correlated with decreased expression of genes mapping to these loci from RNA sequencing of 502 meningiomas (Fig. 1d, e and Supplementary Table 2). Ontology analysis of genes mapping to focal CNVs revealed dysregulation of metabolic and hormone signaling pathways (Fig. 1f), both of which have been implicated in meningiomas through mechanisms that are poorly understood15–18.
+
+Prognostic CNVs from integrated grade, integrated score, and size-dependent LASSO or elastic net models (Fig. 1c) tended to co-occur in individual meningiomas (Fig. 2a). Regularized Cox regression models using co-occurrent CNV pairs identified 1p/22q and 9p/14q co-deletion as important predictors of postoperative LFFR or OS, respectively (Extended Data Fig. 3a). These findings remained significant when accounting for the total number of CNVs per meningioma ("CNV burden") on multivariate modeling (Supplementary Table 3), and meningiomas with 1p/22q or 9p/14q co-deletion, as defined using optimized CNV size-thresholds, had significantly worse clinical outcomes than meningiomas with these CNVs in isolation of one another (Fig. 2b).
+
+Chromosome 22q loss is a common early alteration in meningiomas19, but the prognostic significance of this CNV is limited as subsequent genomic alterations lead to divergent meningioma phenotypes, such as immune infiltration or cell cycle misactivation11. Thus, we hypothesized that CNV accumulation in meningiomas may occur sequentially, with some CNVs like loss of chromosome 22q occurring early during tumorigenesis and other CNVs developing later in tumor progression. In support of this hypothesis, hierarchical clustering of meningiomas, binned by CNV burden using optimized size-thresholds, revealed 3 clusters (Fig. 2c, Extended Data Fig. 3b, c). “Early” cluster CNVs, such as loss of 22q, 1p, and 14q, were prevalent regardless of total CNV burden. “Late” cluster CNVs, such as loss of 9p or gain of 1q, were prevalent in samples with higher CNV burden. The third cluster contained uncommon
+CNVs that did not correlate with total CNV burden. Meningioma CNV burden was associated with worse clinical outcomes, suggesting that progressive destabilization and development of late CNVs is associated with worse prognosis (Extended Data Fig. 3d, e).
+
+To test the broader implications of CNV size thresholds and co-occurrence patterns on cancer risk-stratification, SNP array-derived CNV profiles and clinical outcome data were obtained for 9,885 tumors in TCGA\(^7\). Nine cancer types, comprising approximately half of TCGA samples analyzed, were identified with CNV size-dependence, which was again defined using prognostic CNV-based models with a maximum AUC for 5-year local PFS or OS of at least 0.60 that decreased by at least 5% from the maximum AUC as CNV threshold varied (Fig. 3a, Supplementary Table 4). There were areas of focal deletion or amplification on size-dependent CNVs across these 9 cancer types, such as gain of 1q and loss of 17q or 21q that were not identified in size-independent cancers (Supplementary Table 5). Ontology analysis of genes mapping to focal CNVs across these 9 cancer types revealed dysregulation of metabolic, developmental, differentiation, biosynthetic, cytoskeletal, and enzymatic pathways (Extended Data Fig. 4).
+
+As in meningioma, size-dependent CNVs for 2 cancer types, glioblastoma (GBM) and cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), were used as inputs for co-occurrence models (Fig. 3b). In GBM, concurrent 16q loss and 7p gain was associated with worse OS than these CNVs in isolation (Fig. 3c, d). In CESC, concurrent 13q gain and 19p loss, as well as 19p/21q co-deletion, were both significant predictors of OS (Fig. 3c, d). These CNV co-occurrences remained significant predictors for GBM or CESC outcomes in multivariate models that accounted for total CNV burden (Supplementary Table 6). These findings support the clinical relevance of CNV size-dependence and co-occurrence in developing risk-stratification models for human cancer.
+
+In sum, our results demonstrate that CNVs exhibit size-dependence with respect to their prognostic value across multiple cancer types. We find cancer risk-stratification systems using CNVs with chromosome-specific size thresholds and co-occurrence patterns may refine risk-stratification across a diversity of human cancers.
+
+Methods
+
+Inclusion and ethics
+
+This study complied with all relevant ethical regulations and was approved by the UCSF Institutional Review Board (13-12587, 17-22324, 17-23196 and 18-24633). As part of routine clinical practice at UCSF, all patients included in this study signed a waiver of informed consent to contribute deidentified data to research.
+
+Meningioma samples and clinical data
+Meningioma samples were collected from two sites, UCSF and Hong Kong University. Samples from the UCSF cohort (n = 200) were selected from the UCSF Brain Tumor Center Biorepository and Pathology Core in 2017, and comprised all available WHO grade 2 and 3 meningioma frozen samples, WHO grade 1 frozen samples with clinical follow-up of greater than 10 years (n = 40) or those with the longest available clinical follow-up less than 10 years (n = 47). The electronic medical record was reviewed for all patients in late 2018, and paper charts were reviewed in early 2019 for patients treated before the advent of the electronic medical record. The Hong Kong University cohort (n = 365) comprised consecutive meningiomas from patients treated at Hong Kong University from 2000 to 2019 with frozen tissue that was sufficient for DNA methylation profiling. The medical record was reviewed for all patients in late 2019. For both cohorts, meningioma recurrence was defined as new radiographic tumor on magnetic resonance imaging after gross total resection, or progression of residual meningioma on magnetic resonance imaging after subtotal resection.
+
+Meningioma DNA methylation profiling and analysis
+
+DNA methylation profiling was performed as previously described\(^{11}\) using the Illumina Methylation EPIC 850k Beadchip (WG-317-1003, Illumina) according to manufacturer instructions. Pre-processing and β-value calculations were performed using the SeSAMe (v1.12.9) pipeline (BioConductor 3.13) with default settings. All DNA methylation profiling was performed at the Molecular Genomics Core at the University of Southern California. Assignment of meningiomas to DNA methylation groups or DNA methylation subgroups was performed using support vector models (https://william-c-chen.shinyapps.io/MeninMethylClassApp/)\(^{11,20}\).
+
+TCGA CNV and clinical outcomes data
+
+TCGA data was collected from the TCGA PanCanAtlas (https://gdc.cancer.gov/about-data/publications/pancanatlas)\(^{21}\). Copy number information was obtained using the Copy Number dataset (broad.mit.edu_PANCAN_Genome_Wide_SNP_6_whitelisted.seg). Only primary tumor samples were included by filtering TCGA Biospecimen Core Resource (BCR) barcodes for sample numbers containing the “01” designator. Clinical information was obtained from the TCGA-Clinical Data Resource (CDR) Outcome dataset (TCGA-CDR-SupplementalTableS1.xlsx) and was matched to CNV data by BCR barcode.
+
+CNV analysis
+
+CNV profiles were generated from DNA methylation data using the SeSaMe package as previously described\(^{11}\). The “cnSegmentation” command with default settings and the ‘EPIC.5.normal’ dataset as a copy-number normal control were used.
+
+For both meningioma methylation data and TCGA SNP array data, chromosome segments with mean intensity values less than – 0.1 were defined as lost. Mean intensity values greater than 0.15 were defined as gained. CNV profiling excluded sex chromosomes and p arms of acrocentric chromosomes (13p, 14p, 15p, 21p and 22p). CNV threshold analysis for each CNV profile was performed by measuring
+the mean intensity value at intervals of 30000 bases along each chromosome arm and summing nonconsecutive gains and losses. The total number of CNV profiles which met each threshold of gain or loss from 5–95% by 5% increments of the chromosome arm were counted. 5-year AUC for meningioma LFFR and OS, and TCGA PFS and OS, were calculated for each threshold using the survivalROC package (v1.0.3.1) in R, and the optimal threshold for each CNV was chosen based on the highest AUC for each clinical endpoint. Size-dependent CNVs were defined as those with a maximum 5-year AUC of at least 0.6 with another threshold of less than 95% of that maximum AUC.
+
+CNV network plots were constructed using the igraph package (v1.5.1) in R. Plots were constructed using the CNVs selected from regression models, as well as from those identified in the previously published integrated grade\(^9\) and integrated score\(^{10}\) models for meningioma. CNVs were called using their optimal thresholds. In the case of TCGA cancer data, network plots were constructed using size-dependent CNVs and the most important predictors identified in LASSO and Elastic Net Cox regression co-occurrence models. Co-occurrence analysis was limited to pairs of CNVs as sample size was insufficient to analyze the high number of predictors involved when using 3 or more CNVs.
+
+Cluster analysis was performed using CNVs defined with the optimal size-threshold for predicting LFFR. Clustering was done using the factoextra (v1.0.7) and cluster (v2.1.4) packages in R and visualized with the ComplexHeatmap package (v2.15.4).
+
+Survival analysis and modelling
+
+CNV profiles using the optimal threshold for each CNV were used to train regression models on all available meningioma samples, and for all TCGA samples for size-dependent cancer types (BRCA, CESC, GBM, HNSC, LGG, LUAD, OV, PRAD, and UCEC). LASSO and Elastic net regularized Cox regression models were trained with the concordance index (c-index) for each target endpoint, using the glmnet and cv.glmnet functions from the glmnet package (v4.1-8) in R. Elastic net model selection was performed by selecting an optimal alpha value from a range of 0.05 to 0.95 (0.6 for meningioma LFFR, 0.2 for meningioma OS, 0.85 for TCGA PFS, 0.9 for TCGA OS). Model training was performed using 10-fold cross validation. CNV predictors for each model were identified within 1 standard error of the model achieving maximal c-index to reduce over-fitting. A risk metric was calculated for each sample, defined as the product of the regression coefficients and the normalized counts. Model performance was measured with 5-year cross-validation AUC for each model's respective clinical endpoint using the same training dataset with no hold out validation cohort.
+
+Integrated grade\(^9\) was assigned to meningioma samples using CNV calls for each threshold, mitoses per 10 high-power fields, and \(CDKN2A/B\) loss. Integrated score\(^{10}\) was assigned using CNV calls for each threshold, WHO grade, and methylation family\(^{13}\), the latter which had been previously assigned independently by the authors who developed of this system.
+
+Multivariate Cox proportional hazards analysis was performed using the survival package (v3.5-7) in R.
+
+Focal genomic and ontology analysis
+CNV pileup plots demonstrating the proportion of tumors with gains or losses at each position along the chromosome arm were constructed using the ggplot2 (v3.4.3) package in R. Focal regions of loss were selected by selecting loci along the chromosome arm with a higher proportion of samples demonstrating deletion compared to the surrounding regions. Genes present in regions of interest were identified by cross-referencing positions along the chromosome with the Ensembl (release 109)\(^{22}\) database using the biomaRt (v2.54.1) package in R.
+
+Meningioma gene expression analysis was performed using RNA-Seq data as previously described\(^{16}\). Briefly, RNA sequencing was performed on all 200 of the UCSF samples and 302 of the HKU samples meeting quality metrics. For UCSF samples, library preparation was performed using either the TruSeq RNA Library Prep Kit v2 (RS-122-2001, Illumina), sequencing was done on an Illumina HiSeq 4000 to a mean of 42 million reads per sample at the UCSF IHG Genomics Core, Quality control of FASTQ files was performed with FASTQC (v0.11.9), and 50 bp single-end reads were mapped to the human reference genome GRCh38 using HISAT2 (v2.1.0) with default parameters. For HKU samples, library preparation was performed using the TruSeq Standard mRNA Kit (20020595, Illumina) and 150 bp paired-end reads were sequenced on an Illumina NovaSeq 6000 to a mean of 100 million reads per sample at MedGenome Inc. Analysis was performed using a pipeline comprised of FastQC for quality control, and Kallisto for reading pseudo alignment and transcript abundance quantification using the default settings (v0.46.2).
+
+Gene ontology and interaction analysis were performed using Cytoscape. In brief, Gene Set Enrichment Analysis (GSEA, v4.3.2) was performed and gene rank scores were calculated using the formula sign(log\(_2\) fold-change) × −log10(p-value). Pathways were defined using the gene set file Human_GOBP_AllPathways_no_GO_iea_December_01_2022_symbol.gmt, which is maintained by the Bader laboratory. Gene set size was limited to range between 15 and 500, and positive and negative enrichment files were generated using 2000 permutations. The EnrichmentMap App (v3.3.4) in Cytoscape (v3.7.2) was used to visualize the results of pathway analysis. Nodes with FDR q value < 0.05 and p-value < 0.05, and nodes sharing gene overlaps with Jaccard + Overlap Combined (JOC) threshold of 0.375 were connected by blue lines (edges) to generate network maps. Clusters of related pathways were identified and annotated using the AutoAnnotate app (v1.3.5) in Cytoscape that uses a Markov Cluster algorithm to connect pathways by shared keywords in the description of each pathway. The resulting groups of pathways were designated as the consensus pathways in a circle.
+
+Statistics
+
+All experiments were performed with independent biological replicates and repeated, and statistics were derived from biological replicates. Biological replicates are indicated in each figure panel or figure legend. No statistical methods were used to predetermine sample sizes, but sample sizes in this study are similar or larger to those reported in previous publications. Data distribution was assumed to be normal, but this was not formally tested. Investigators were blinded to conditions during clinical data collection and analysis. Bioinformatic analyses were performed blind to clinical features, outcomes or
+molecular characteristics. The clinical samples used in this study were retrospective and nonrandomized with no intervention, and all samples were interrogated equally. Thus, controlling for covariates among clinical samples is not relevant. No data points were excluded from the analyses. Statistical analyses were conducted in R (v4.2.2).
+
+Declarations
+
+Reporting summary
+
+Further information on research design is available in the Nature Research Reporting Summary linked to this article.
+
+Data availability
+
+DNA methylation (n=565) and RNA sequencing (n=502) of the meningiomas analyzed in this manuscript have been deposited in the NCBI Gene Expression Omnibus under the accessions GSE183656 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE183656), GSE101638 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE101638), and GSE212666 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE212666). The publicly available GRCh38 (hg38, https://www.ncbi.nlm.nih.gov/assembly/GCF_000001405.39/), and Kallisto index v10 (https://github.com/pachterlab/kallisto-transcriptome-indices/releases) datasets were used in this study. TCGA data was collected from the publicly available TCGA PanCanAtlas (https://gdc.cancer.gov/about-data/publications/pancanatlas). Copy number information was obtained using the Copy Number dataset (broad.mit.edu_PANCAN_Genome_Wide_SNP_6_whitelisted.seg). Clinical information was obtained from the TCGA-Clinical Data Resource (CDR) Outcome dataset (TCGA-CDR-SupplementalTableS1.xlsx). Source data are provided with this paper.
+
+Code availability
+
+No custom software, tools, or packages were used. The open-source software, tools, and packages used for data analysis in this study are referenced in the methods where applicable and include R (v4.2.2), FASTQC (v0.11.9), HISAT2 (v2.1.0), Kallisto (v0.46.2), SeSAMe (v1.12.9) (BioConductor 3.13), survival R package (v3.5-7), survivalROC R package (v1.0.3.1), biomaRt R package (v2.54.1), glmnet R package (v4.1-8), igraph R package (v1.5.1), factoextra R package (v1.0.7), cluster R package (v2.1.4), ComplexHeatmap R package (v2.15.4), GSEA (v4.3.2), and EnrichmentMap App (v3.3.4) and AutoAnnotate app (v1.3.5) in Cytoscape (v3.7.2).
+
+Acknowledgements
+
+This study was supported by funding from American Brain Tumor Association Jack & Fay Netchin Medical Student Summer Fellowship in memory of Rose Digang to M.P.N., K12 CA260225 and the Chan Zuckerberg Biohub Physician Scientist Fellowship to W.C.C., and R01 CA262311, P50 CA097257, the UCSF Wolfe Meningioma Program Project and the Trenchard Family Charitable Fund, to D.R.R. The
+results shown here are in part based upon data generated by the TCGA Research Network (https://www.cancer.gov/tcga).
+
+Author contributions statement
+
+All authors made substantial contributions to the conception or design of the study; the acquisition, analysis, or interpretation of data; or drafting or revising the manuscript. All authors approved the manuscript. All authors agree to be personally accountable for individual contributions and to ensure that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved and the resolution documented in the literature. M.P.N. and W.C.C. conceived and designed the study and analyzed bioinformatic data with supervision from D.R.R. N.Z. performed gene ontology analyses with supervision from D.R.R. K.M. and C-H.G.L. provided guidance and feedback on study design, analysis, and presentation.
+
+Competing interests statement
+
+The authors declare no competing interests.
+
+References
+
+1. Hanahan, D. Hallmarks of Cancer: New Dimensions. Cancer Discov **12**, 31–46 (2022).
+2. Nguyen, B. *et al.* Genomic characterization of metastatic patterns from prospective clinical sequencing of 25,000 patients. *Cell* **185**, 563-575.e11 (2022).
+3. Lukow, D. A. *et al.* Chromosomal instability accelerates the evolution of resistance to anti-cancer therapies. *Dev Cell* **56**, 2427-2439.e4 (2021).
+4. Bakhoum, S. F. *et al.* Numerical chromosomal instability mediates susceptibility to radiation treatment. *Nat Commun* **6**, 5990 (2015).
+5. Beroukhim, R. *et al.* The landscape of somatic copy-number alteration across human cancers. *Nature* **463**, 899–905 (2010).
+6. Steele, C. D. *et al.* Signatures of copy number alterations in human cancer. *Nature* **606**, 984–991 (2022).
+7. Weinstein, J. N. *et al.* The Cancer Genome Atlas Pan-Cancer analysis project. *Nat Genet* **45**, 1113–1120 (2013).
+8. van Dijk, E. *et al.* Chromosomal copy number heterogeneity predicts survival rates across cancers. *Nat Commun* **12**, 3188 (2021).
+9. Driver, J. *et al.* A molecularly integrated grade for meningioma. *Neuro Oncol* **24**, 796–808 (2022).
+10. Maas, S. L. N. *et al.* Integrated Molecular-Morphologic Meningioma Classification: A Multicenter Retrospective Analysis, Retrospectively and Prospectively Validated. *J Clin Oncol* **39**, 3839–3852 (2021).
+11. Choudhury, A. et al. Meningioma DNA methylation groups identify biological drivers and therapeutic vulnerabilities. Nat Genet **54**, 649–659 (2022).
+12. Youngblood, M. W. et al. Associations of meningioma molecular subgroup and tumor recurrence. Neuro Oncol **23**, 783–794 (2021).
+13. Sahm, F. et al. DNA methylation-based classification and grading system for meningioma: a multicentre, retrospective analysis. Lancet Oncol **18**, 682–694 (2017).
+14. Ostrom, Q. T. et al. CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2015-2019. Neuro Oncol **24**, v1–v95 (2022).
+15. Nassiri, F. et al. A clinically applicable integrative molecular classification of meningiomas. Nature **597**, 119–125 (2021).
+16. Choudhury, A. et al. Hypermitotic meningiomas harbor DNA methylation subgroups with distinct biological and clinical features. Neuro-Oncology **25**, 520–530 (2023).
+17. Miyagishima, D. F., Moliterno, J., Claus, E. & Günel, M. Hormone therapies in meningioma-where are we? J Neurooncol **161**, 297–308 (2023).
+18. Walsh, K. M. et al. Pleiotropic MLLT10 variation confers risk of meningioma and estrogen-mediated cancers. Neurooncol Adv **4**, vdac044 (2022).
+19. Magill, S. T. et al. Multiplatform genomic profiling and magnetic resonance imaging identify mechanisms underlying intratumor heterogeneity in meningioma. Nat Commun **11**, 4803 (2020).
+20. Chang, C.-W. et al. Identification of Human Housekeeping Genes and Tissue-Selective Genes by Microarray Meta-Analysis. PLoS One **6**, e22859 (2011).
+21. Smith, J. C. & Sheltzer, J. M. Genome-wide identification and analysis of prognostic features in human cancers. Cell Reports **38**, (2022).
+22. Cunningham, F. et al. Ensembl 2022. Nucleic Acids Research **50**, D988–D995 (2022).
+
+Figures
+Figure 1
+
+Meningioma risk-stratification models demonstrate CNV size-dependence. a, Heatmaps showing area under the curve for univariate Cox models of LFFR or OS based on individual copy number gains (left, red) or individual copy number losses (right, blue). Models were trained using sequential size thresholds requiring ≥5% to ≥95% of chromosome arms to be gained or lost to define CNVs. Boxes show peak AUCs for “size-dependent” CNVs, defined as having a maximum area under the curve (AUC) for 5-year LFFR or OS of at least 0.60 that decreased by at least 5% from the maximum AUC as CNV threshold was varied. n=565 meningiomas. b, Previously published meningioma risk-stratification models incorporating CNVs (integrated grade based on histology and a ≥50% CNV threshold, or integrated score based on histology,
+DNA methylation profiling, and a ≥5% CNV threshold) show decreasing AUC with varying size-thresholds. n=565 meningiomas. c, CNVs from previously published meningioma risk-stratification models or from newly-derived size-dependent LASSO or elastic net models for meningioma LFFR or OS. n=565 meningiomas. d, CNV profile plots demonstrating focal copy number losses in size-dependent CNVs from LASSO or elastic net models. Chromosomes 14 and 20 are shown as examples of broad/non-focal CNVs. n=565 meningiomas. e, Average RNA sequencing expression of genes mapping to regions of focal copy number loss on size-dependent CNVs from LASSO or elastic net models versus genes mapping to other regions on the same chromosomes. n=502 meningiomas. Error bars show standard error of the mean. Student’s t test, p≤0.0001. f, Network of gene circuits distinguishing genes mapping to regions of focal copy number loss. Nodes represent pathways and edges represent shared genes between pathways (p≤0.05, FDR≤0.05). n=502 meningiomas.
+
+![Network diagrams and heatmaps showing CNV and gene expression data for meningiomas](page_186_370_1207_1092.png)
+Figure 2
+
+Size-dependent CNV co-occurrence is prognostic for meningioma outcomes. a, Network diagrams demonstrating co-occurrence of prognostic size-dependent CNVs from Fig. 1c. b, Kaplan-Meier curves comparing meningioma LFFR or OS according to individual CNVs versus co-occurrent CNV pairs identified as the most important predictors of postoperative outcomes in LASSO Cox models from Extended Data Fig. 3a using optimized thresholds for defining CNVs from Fig. 1a. Log-rank tests. n=565 meningiomas. c, Heatmap showing unsupervised hierarchical clustering of individual CNVs according to the total number of CNVs per meningioma. CNVs were defined using optimal size thresholds for LFFR or OS models from Fig. 1a.
+
+![Network diagrams and Kaplan-Meier curves for meningioma CNVs](page_184_370_1207_1012.png)
+
+Figure 3
+
+Pan-cancer analyses reveal size-dependent CNV co-occurrence risk-stratification models for half of human cancers. a, TCGA SNP-array and clinical outcomes data used in pan-cancer analyses. Cancers with size-dependent prognostic CNVs were defined as having a CNV with a univariate Cox AUC for either PFS or OS of at least 0.60 that dropped by at least 5% from the maximum AUC when varying the size threshold for defining CNVs. b, Network diagrams demonstrating co-occurrence of prognostic size-dependent CNVs for GBM or CESC from Supplementary Table 4. c, LASSO Cox model coefficients using
+size-dependent CNV co-occurrence to predict postoperative OS in GBM or CESC. **d**, Kaplan-Meier curves comparing OS for GBM or CESC with individual CNVs versus co-occurrent CNV pairs identified as the most important predictors of postoperative outcomes in LASSO Cox models. Log-rank tests. n= 571 GBM and 294 CESC.
+
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• NguyenChenNatGenetEDFigv7.docx
+• NguyenChenNatGenetSupplementaryTablesv7.xlsx

084645ecaf0633670ce86e9b62f37cf8c911eec46c0e61646e18f05eb546a1be/preprint/preprint.md

@@ -0,0 +1,375 @@
+PGMG: A Pharmacophore-Guided Deep Learning Approach for Bioactive Molecular Generation
+
+Min Li (limin@mail.csu.edu.cn)
+Central South University    https://orcid.org/0000-0002-0188-1394
+Huimin Zhu
+Central South University
+Renyi Zhou
+Central South University
+Jing Tang
+Research Program in Systems Oncology, Faculty of Medicine, University of Helsinki, Helsinki, Finland
+https://orcid.org/0000-0001-7480-7710
+
+Article
+
+Keywords:
+
+Posted Date: September 15th, 2022
+
+DOI: https://doi.org/10.21203/rs.3.rs-1749921/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+PGMG: A Pharmacophore-Guided Deep Learning Approach for Bioactive Molecular Generation
+
+Huimin Zhu1,†, Renyi Zhou1,‡, Jing Tang2 and Min Li1,*
+
+1 School of Computer Science and Engineering, Central South University, Changsha 410083, China
+2 Faculty of Medicine, University of Helsinki, Helsinki, 00290, Finland
+
+† These two authors contribute equally to the work.
+* Corresponding author, limin@mail.csu.edu.cn
+
+Abstract
+
+The rational design of novel molecules with desired bioactivity is a critical but challenging task in drug discovery, especially when treating a novel target family or understudied targets. Here, we propose PGMG, a pharmacophore-guided deep learning approach for bioactivate molecule generation. Through the guidance of pharmacophore, PGMG provides a flexible strategy to generate bioactive molecules matching given pharmacophore models. PGMG uses a graph neural network to encode pharmacophore features and spatial information and a transformer decoder to generate molecules. A latent variable is introduced to solve the many-to-many mapping between pharmacophores and molecules and improve the diversity of generated molecules. In addition, these generated molecules are of high validity, uniqueness, and novelty. In the case studies, we demonstrate using PGMG in ligand-based and structure-based drug de novo design, as well as in lead optimization scenarios. Overall, the flexibility and effectiveness make PGMG a useful tool to accelerate drug discovery process.
+
+Introduction
+
+The acquisition of biologically active compounds is a vital but challenging step in drug discovery. It has been estimated that the drug-like chemical space is as large as \(10^{60}\) obeying Lipinski’s “Rule of Five”1,2. Hence, it is an extremely difficult task to search for desired molecules in such a huge space. Traditionally, hit compounds which exhibit initial activity on a specific target can be obtained from natural products, designed by medicinal chemists, or acquired by high-throughput screening (HTS)3. These methods consume a lot of human and financial resources, which makes the acquisition of hit compounds inefficient and costly. Recently, some deep generative models have been proposed for the rational design of novel molecules with desired properties, which provide a new perspective for this task.
+
+Among the popular architectures and models for generating molecules from deep neural networks, the variational autoencoders (VAEs)4,5, reinforcement learning (RL)6,7, generative adversarial networks8-10 and auto-regressive models11,12 were successful to design desired molecules at a specified precondition. Regardless what framework were used, most above methods aim at generating molecules with given physicochemical properties, such as the Wildman-Crippen partition coefficient (LogP), synthetic accessibility (SA), molecular weight (molWt), quantitative estimate of drug likeness (QED), and others.
+However, the relationship between molecular and some physicochemical properties such as LogP, QED values are easy to achieve. In drug discovery, the most difficult and the real intrinsic objective is to design molecules that satisfy properties which require wet experimental to measure or require extensive calculations to approximate, such as the activity of the molecule\(^{13}\). These models are less suitable for generating bioactive molecules. For a specified target, these models require a large dataset of known active molecules to fine-tune and thus cannot be applied to a novel target or targets with few active compounds.
+
+Designing molecules using deep generative models with biological activity remains challenging\(^{14-16}\). As mentioned above, one of the main obstacles is the limited data on target-specific molecules, which makes it difficult for models to learn the structure-activity relationship. For a novel target family, the paucity of activity data is even more severe. Besides, the scarcity of activity data affects the strategy of drug design. For example, the choice of ligand-based drug design or structure-based drug design depends on what information can be used, which narrows down the application scenarios of many methods. It is clear that incorporating expert knowledge in the generation process is beneficial to the full utilization of bioactivity data information\(^{17}\). Therefore, combining deep generative models with knowledge in biochemistry to efficiently use the scarce data to design biologically active molecules is a crucial project.
+
+Up to now, some methods that generate bioactive molecules by combining prior knowledge from biochemistry into molecule generation models have been proposed. For example, conditioned GAN can be used to design active-like molecules for desired gene expression signatures\(^{18}\), which provides a new perspective for molecule generation. However, the structure-activity relationship between the biological activity and the molecules generated by such methods is ambiguous. DeLinker\(^{19}\) and SyntaLinker\(^{20}\) adopt fragment-based drug design and retain active fragments while updating linkers to generate active molecules, and DEVELOP\(^{17}\) combines DeLinker with chemical features as constraints to improve the quality of the generated molecules. The fragment-based approaches require explicit knowledge of the active fragments, which lead to a limited chemical space for the model. DeepLigBuilder\(^{21}\) and 3D-Generative-SBDD\(^{22}\) utilize the structure-based drug design strategy and generate molecules based on the binding sites between molecules and proteins in the 3D Euclidean Space. However, these methods are limited when the binding site or the target structure is unknown. There are also some methods that use electronic features in molecule generation. For example, Reduced Graph\(^{23}\) simplifies a SMILES to an acyclic graph of functional group as its input to generation. Shape-based method proposed by Skalic et al\(^{24}\) generate molecules from a 3D representation using a seed ligand with a conditional chemical features. These methods require seed compounds to collect the input electronic features. The above generative models may perform well on specific types of activity data, but their usages are limited because of their assumptions on the data types.
+
+Here, we propose PGMG, a pharmacophore-guided molecule generation approach based on deep learning. PGMG uses pharmacophore models as a bridge to connect different types of activity data and can design bioactive molecules for newly discovered targets when there is no sufficient activity data. A pharmacophore is a set of chemical features and its spatial information that are necessary for a drug to bind to a target and can be constructed by superimposing a small number of active compounds\(^{25}\) or observing the structure of a given target\(^{26}\). Traditional drug design based on pharmacophores has many successful applications\(^{27,28}\), but
+its potential in deep generative models has not been exploited. There are some works\(^{23, 24}\) that use pharmacophore-like information in molecule generation. However, the pharmacophore-like features used in these methods are incomplete and can only be extracted from seed compounds, making it difficult for domain knowledge to be leveraged. In PGMG, we use a complete graph to fully represent a pharmacophore with each node corresponding to a pharmacophore feature and the spatial information encoded as the distance between each node pair. Given the graph as the sole input, PGMG can generate molecules matching the corresponding pharmacophore. This gives PGMG the capability to utilize different types of activity data in a uniform representation and a biologically meaningful way to control the process of the bioactivity molecule design.
+
+Since pharmacophores and molecules have a many-to-many relationship, PGMG introduces latent variables to model such a relationship and boost the variety of generated molecules. Besides, the transformer structure is employed as the backbone to learn implicit rules of SMILES strings to map between latent variables and molecules. We evaluate the PGMG performance comprehensively in molecule generation with goal-directed metrics and drug-like metrics. The results show that PGMG can generate molecules satisfying given pharmacophore models and pharmacokinetic requirements, while maintaining a high level of validity, uniqueness, and novelty. The case studies further demonstrate that PGMG provides an effective strategy for both ligand-based and structure-based drug de novo designs and lead optimization.
+
+Results
+
+Overview of PGMG
+Our proposed PGMG is a pharmacophore-guided molecular generation approach based on deep learning. The overall architecture of PGMG is illustrated in Figure 1.
+
+Given a target pharmacophore, the goal of PGMG is to generate molecules which matches the pharmacophore. Here, we introduce a set of latent variables \( z \) to deal with the many-to-many mapping between pharmacophores and molecules. Thus, a molecule \( x \) can be represented as a unique combination of two complementary encodings including \( c \) representing the given pharmacophore and \( z \) corresponding to how chemical groups are placed within the molecule. From another perspective, the latent variables \( z \) grant PGMG to model multiple modes in the conditional distribution
+
+\[
+P(x|c) = \int_{z \sim P(z|c)} P(x|c,z)P(z|c)dz
+\]
+
+We train two neural networks, an encoder network \( P_\phi(z|c,x) \) to approximate \( P(z|c) \) indirectly and a decoder network \( P_\theta(x|c,z) \) to approximate \( P(x|c,z) \). We embed molecules in SMILES format into dense feature vectors and use Gated GCN\(^{29}\) to embed pharmacophore models. The transformer structure proposed by Vaswani et al.\(^{30}\) is used as the backbone of our model to learn the mapping between pharmacophore and molecular structures.
+Figure 1 | The overall architecture of PGMG. (a) The construction of pharmacophore networks. We use the shortest paths on the molecular graph to approximate the Euclidean distances between two pharmacophore features and construct a fully connected graph to represent a pharmacophore model. (b) The preprocessing of SMILES. We randomize a given canonical SMILES and corrupt it using the infilling scheme. (c) Model structure and pipelines for training and inferencing. c represents the embedding vector sequences for the given pharmacophore model; x represents the embedding sequence of input SMILES; z represents the latent variables for a molecule. Transformer encoder and decoder blocks are stacked with N layers. ⊕ denotes concatenation of two vectors and ⊗ denotes matrix multiplication. The overlap between the training and inferencing process is highlighted in the right panel.
+
+To train PGMG, we need only a number of SMILES strings with no additional information attached. A training sample can be constructed using the SMILES representation of a molecule. First, the chemical features of a molecule are identified using RDKit31 and we randomly select some of them to build a pharmacophore network \( G_p \). As shown in Figure 1a, we approximate the Euclidean distance in the three-dimensional Euclidean space in a pharamacophore using the length of the shortest path between two pharmacophore features on the molecular graph. The analysis of the two distances can be found in Figure S1. Next, a molecule is represented as a randomized SMILES string and then segmented into a token sequence s. We then corrupt s to get the encoder input \( s' \) by using the infilling scheme32 and obtain a training sample (\( G_p, s, s' \)). Since we avoid the use of target-specific active data in the training stage, PGMG bypasses the problem of data scarcity on active molecules.
+When using the trained model to generate molecules, a pharmacophore model is required. The generation process is as follows. Given a pharmacophore model \( c \), a set of latent variables \( z \) is sampled from the prior distribution \( p(z|c) \), which in our case is the standard Gaussian distribution \( N(0, I) \), and molecules are then generated from the conditional distribution \( p(x|z, c) \). There are multiple ways to construct a pharmacophore model using various active data types and this is where the flexibility of the PGMG approach comes in. We employ both ligand-based and structure-based approaches to build pharmacophores and use them to generate active molecules for de novo drug design.
+
+**Performance of PGMG on the unconditional molecule generation task.**
+
+We evaluate our model’s performance on the unconditional molecule generation task by comparing it with other SMILES-based methods including ORGAN\(^9\), VAE\(^4\), SMILES LSTM\(^{33}\), and Syntalinker\(^{20}\). We train PGMG and other SMILES-based models on the ChEMBL dataset\(^{34}\) based on the train-test split used in the GuacaMol benchmark\(^{35}\). Since PGMG is a conditional model, we approximate the unconditional distribution by generating molecules based on randomly sampled pharmacophore features. The molecule generation performance is evaluated by four metrics including validity, novelty, uniqueness, and ratio of available molecules (see Methods for the definition of the metrics). The comparison results of PGMG and four other SMILES-based methods on the four metrics are shown in **Figure 2a**. The results of ORGAN, VAE, SMILES LSTM on validity, novelty and are taken from the GuacaMol benchmark directly.
+
+As shows in **Figure 2a**, PGMG performs better in novelty and the ratio of available molecules, while keeping the same level of validity and uniqueness as the top models. The ratio of available molecules is the ratio of unique novel valid molecules to all generated molecules, and equals product of the previous three metrics, as a composite metric to assess the performance of the model to generate novel molecules. PGMG achieves the highest the ratio of available molecules. Comparing to the second-best method, PGMG improves the ratio of available molecules by 6.3%. Among the SMILES-based methods, SMILES LSTM\(^{33}\) performs the best in uniqueness, while Syntalinker\(^{20}\) performs the best in validity.
+
+In the ablation study, we remove features of PGMG and see how that affects performance. All models are trained using the ZINC\(^{36}\) dataset with the same parameters. Validity, uniqueness, and novelty are evaluated by generating 10 molecules for 1 pharmacophore extracted from each molecule in the test dataset. The match score is evaluated by generating 512×512 molecules for 512 SMILES randomly sampled from the test dataset. The result of our ablation study can be found in **Figure 2b**.
+
+We find when using canonical SMILES to train PGMG (*canonical_SMILES*), the uniqueness increases from 0.98 to 0.99, but the match score decreases from 0.91 to 0.94. A similar result can be found when we change the Gaussian distribution of the latent variable \( z \) to a Dirac delta distribution, denoted as PGMG (*remove_z*). *remove_z* exhibits a huge decrease on the uniqueness (from 0.98 to 0.81) and a certain degree of increase on the match score (from 0.94 to 0.97). If we use random sampling during generation, we can make the uniqueness increase, but it cannot make up for the drop of both the validity and the match score. As we see here, there seems to be a trade-off between the uniqueness and match score.
+We also test PGMG’s performance when replacing the distance between chemical features with a constant number PGMG (remove_dis). The results show a large decrease in both uniqueness (from 0.98 to 0.82) and match score (from 0.94 to 0.60) as expected, which shows that PGMG makes a good use of the spatial information of pharmacophores.
+
+To test whether PGMG catches the distribution of training dataset, we further examine the physicochemical properties of the generated molecules. The distribution of physicochemical properties of the generated molecules and the molecules in the training dataset are compared in Figure 2c. We find that the physicochemical properties distribution such as the topological polar surface area (TPSA), SA, QED, and LogP are close to the training set distribution. This demonstrates that PGMG captures the distribution of molecules in the training dataset well.
+
+<table>
+  <tr>
+    <th></th>
+    <th>Validity</th>
+    <th>Uniqueness</th>
+    <th>Novelty</th>
+    <th>Ratio of available molecules</th>
+    <th>Match score</th>
+  </tr>
+  <tr>
+    <td>PGMG(canonical_SMILES)</td>
+    <td>0.98</td>
+    <td>0.99</td>
+    <td>1.0</td>
+    <td>96.4%</td>
+    <td>0.91</td>
+  </tr>
+  <tr>
+    <td>PGMG(remove_dis)</td>
+    <td>0.99</td>
+    <td>0.82</td>
+    <td>1.0</td>
+    <td>81.2%</td>
+    <td>0.60</td>
+  </tr>
+  <tr>
+    <td>PGMG(remove_z)</td>
+    <td>0.99</td>
+    <td>0.81</td>
+    <td>1.0</td>
+    <td>79.6%</td>
+    <td>0.97</td>
+  </tr>
+  <tr>
+    <td>PGMG(remove_z, random sampling)</td>
+    <td>0.92</td>
+    <td>1.0</td>
+    <td>1.0</td>
+    <td>91.5%</td>
+    <td>0.89</td>
+  </tr>
+  <tr>
+    <td>PGMG</td>
+    <td>0.97</td>
+    <td>0.98</td>
+    <td>1.0</td>
+    <td>94.8%</td>
+    <td>0.94</td>
+  </tr>
+</table>
+
+Figure 2 | Performance of PGMG on the unconditional molecule generation task. (a) Performance of PGMG and SMILES-based models on ChEMBL. (b) Results of the ablation study. (c) Distribution of chemical properties for the ChEMBL training set and the molecules generated by PGMG. The scientific notation at the upper left of the figure indicates the scaling of the vertical coordinates.
+
+PGMG can generate bioactive molecules satisfying given pharmacophores.
+
+We evaluate the extent to which the generated molecules fit the given pharmacophore models and predict binding affinity between protein receptors and molecules generated using PGMG through the molecular docking tool vina37. We use a match score to estimate the matching degree between each molecule-pharmacophore pair (see calculation of match score section of the Supplementary Information for details).
+
+We extract a random pharmacophore model from each molecule in the test dataset. About 220,000 molecules in total are generated from those random pharmacophore models and the match score is calculated between each pair. For comparison, we also calculate the match score between 220,000 random
+molecules from the ChEMBL dataset\(^{34}\) and the selected pharmacophores. The result is shown in **Figure 3a**.
+
+As can be seen from **Figure 3a**, 86.3% of the generated molecules have matching scores concentrated in the range of 0.8-1.0, with 77.9% having a matching score of 1.0. Meanwhile, the matching degrees for the random molecules are centered at 0.45, with only 4.8% having a matching score of 1.0. This result demonstrates PGMG’s ability to generate molecules satisfying the given pharmacophore models.
+
+![Pharmacophore matching test results and the distribution of four target docking scores.](page_120_340_1247_1536.png)
+
+Figure 3 | Pharmacophore matching test results and the distribution of four target docking scores. (a) The match score of random selected molecules and PGMG generated molecules. (b) The distributions of the predicted affinity of top 1000 molecules generated by PGMG over VEGFR2 (PDB: 1YWN), CDK6 (PDB: 2EUF), TGFβ 1 (PDB: 6B8Y),
+BRD4 (PDB: 3MXF), and the affinity for the known bioactivity molecules corresponding to these targets. (c) Distributions of ADMET properties are calculated using ADMETlab 2.0\(^{38}\) of top 1000 molecules generated by PGMG. The threshold of each property according to ADMETlab 2.0 is given as the dashed line. “↑” indicates that the distribution greater than the threshold satisfies the expected property, while “↓” indicates that the part of lower than the threshold satisfies the expected property. TPSA represents the topological polar surface area, optimal: 0–140 (\(\text{\AA}^2\)); MW denotes Molecular Weight, Optimal:100–600; nHA represents the number of hydrogen bond donors, optimal: 0~7; nHD represents the number of hydrogen bond acceptors, optimal: 0~12; SAscore represents the synthetic accessibility score, optimal: 0–6; Madin–Darby Canine Kidney cells (MDCK) measure the uptake efficiency of a drug into the body, optimal: >2 × 10\(^{-6}\) (cm/s); BBB measures the ability of a drug to cross the blood-brain barrier to its molecular targets, qualified value: 0-0.7; F(20%) denotes human oral bioavailability 20% which assess the efficiency of the drug delivery to the systemic circulation, optimal: 0–0.3; CYP2C9 assess drug metabolism reactions, the closer to 1, the more likely it is to be an inhibitor; T12 represents the half-life of the drug, qualified value: 0-0.7 and hERG evaluates whether the molecule is toxic to the heart, qualified value: 0-0.7; ROA measures acute toxicity in mammals, qualified value: 0-0.7. Where a molecule with a property in the optimal range means that the property is optimal, and a molecule with a property in the qualified range means that there is no obvious evidence that the property of the molecule is defective. The scientific notation at the upper left of the figure indicates the scaling of the vertical coordinates.
+
+To further examine the binding activity of molecules generated by PGMG through the guidance of pharmacophores, we obtain pharmacophore models with known target structure from the literature\(^{39-42}\). These targets include VEGFR2, CDK6, TFGβ 1, BRD4. For each pharmacophore model, 10,000 molecules are generated by PGMG. Autodock vina\(^{37}\) is used to calculate the binding affinities of generated molecules. And then, we select the top 1000 molecules with the strongest binding affinity. For comparison, we acquire the known bioactivity molecules for the four targets from CHEMBL, including 13299, 1648, 1885 and 4786 bioactivity molecules, respectively. In Figure 3b, we show the affinity distributions of the top 1000 molecules generated by PGMG and the affinities for the known bioactivity molecules from CHEMBL. The average affinity of the top 1000 molecules generated by PGMG is -10.0 kcal/mol (1YWN), -11.1 kcal/mol (2EUF), -11.0 kcal/mol (6B8Y) and -8.8 kcal/mol (3MXF), and the average affinity of the known bioactivity molecules is -8.0 kcal/mol (1YWN), -9.6 kcal/mol (2EUF), -9.2 kcal/mol (6B8Y) and -7.0 kcal/mol (3MXF) respectively. The distribution of affinities suggests that PGMG can generate desired bioactive molecules.
+
+To evaluate if PGMG is capable of generating drug-like molecules, we further calculate the pharmacokinetics properties (absorption, distribution, metabolism, excretion) and toxicity (ADMET) of the top 1000 molecules. The ADMET distributions of the top 1000 molecules are illustrated in Figure 3c. Most of the molecules generated by PGMG satisfy the pharmacokinetic properties and toxicity constraint for drug candidate according to the standard proposed by ADMETlab 2.0\(^{38}\). And the majority of the generated molecules are predicted with no obvious toxicity to the heart.
+
+Structure-based drug design
+
+Structure-based drug design is a powerful drug design strategy to generate the desired bioactivity molecules using the structure of specific target\(^{43}\). We use four targets (VEGFR2, CDK6, TFGβ 1, BRD4) from the above section with pharmacophore models which are built using ligand-receptor complex as examples to further demonstrate the performance of PGMG in structure-based drug design. It should be noted that the construction of pharmacophore models does not necessarily need any ligand. We choose these
+pharmacophore models for the convenience of the following analyses. We compare several top affinity conformations of the generated molecules with the top affinity conformation of the reference ligand in the crystal complex. Figure 4 shows the binding sites of the four receptors with corresponding molecules. Most of the generated molecules share the same amino acid residues as the reference ligand, which indicates that those generated molecules are capable of fitting into the binding site as well as the reference one.
+
+![A display of the binding sites of the molecules generated by PGMG in structure-based drug design.](page_186_370_1077_1012.png)
+
+Figure 4 | A display of the binding sites of the molecules generated by PGMG in structure-based drug design.
+
+In Figure 4(a-c), the generated molecules of 1YWM have a similar structure with the reference. As for 2EUF and 6B8Y, despite the structural differences between the generated molecules and the reference molecule, the generated molecules (Figure 4 (e-g, i-k)) share some common important functional groups as the reference ligands (Figure 4h, Figure 4l). And interestingly, we find that the structures of the generated molecules may differ from the reference ligand (Figure 4p) in a good way. For example, the molecules generated by PGMG for 3MXF (Figure 4 (m-o)) can bind to D88, P86, and P82 amino acid residues other than N140 (Figure 4p). This finding suggests that PGMG may have the potential in exploring new binding sites. Besides, we exam the drug-likeness using SA and hERG. SA is designed to estimate the ease of synthesis of drug-like molecules, and it’s easy to synthesize when SA is less than 6. The hERG is a toxicity
+metric. Abnormal hERG values for a drug may lead to palpitations, syncope, and even sudden death. This metric measures the probability that a molecule will be toxic. Empirically, over 0.7, the molecule is considered toxic. These generated molecules perform well on SA and hERG. The above results show that PGMG can design molecules that not only fit well into the binding site but also exhibit drug-like quality in the structure-based drug design.
+
+Ligand-based drug design
+Although structure-based drug design is a successful and highly attractive strategy, there are some prerequisites to use this strategy, including a certain target, a high-resolution crystal structure of the target, and some identified interaction sites. However, it is not easy to reach the above prerequisites. Ligand-based drug design is capable of designing drug molecules based on the conformational superposition of known active molecules when the target is unknown or the binding site is unclear. And it has been widely used in drug discovery, such as the search for new drugs for drug resistance. Squalene oxidase is the target for ringworm, superficial skin fungal infections, and other diseases. Butenafine and terbinafine are typical inhibitors for squalene oxidase\(^{44}\). However, these inhibitors are prone to drug resistance and side effects including skin erythema, burning, and itching. Therefore, it is urgent to design novel squalene oxidase inhibitors. Here, we generate 200 molecules using a pharmacophore model constructed from squalene oxidase inhibitors.
+
+![Alignment diagrams of terbinafine, pharmacophore model, and molecules generated by PGMG.](page_186_682_1077_393.png)
+
+Figure 5 Alignment diagrams of terbinafine, pharmacophore model, and molecules generated by PGMG. The different colored spheres represent different pharmacophore features. Aromatic center is red, the positive charge center is yellow, and hydrophobic centers are green. The grey molecules represent terbinafine, and the green molecules represent the molecules generated by PGMG.
+
+As shows in Figure 5, the generated molecules align well to the active conformation of terbinafine which
+is obtained from drugbank^{45}. The listed molecules match well with the desired pharmacophore features, including two hydrophobic centers, a positive charge center, and an aromatic ring center. Here, we notice that PGMG has a good grasp of the equivalence of different substructures under the same pharmacophore feature. It matches the aromatic center with pyrrole, thiophene, and pyrimidine, and the hydrophobic center with aliphatic, cycloalkane, and benzene. This result shows that PGMG can generate diverse molecules while maintaining the important properties of the substructures the same as the known inhibitor.
+
+To further assess the pharmacokinetics and toxicity of the generated molecules, we calculate the TSPA, SA, and hERG of the generated molecules. See the previous section for a detailed SA and hERG description. TSPA is a molecular descriptor measuring drug transport properties such as intestinal absorption and blood-brain barrier (BBB) penetration. The TPSA in the range of 0-140 means optimal. Of the six molecules generated by PGMG, their TSPA, SA, and hERG values are within the rational range. From Figure 5, we can see that PGMG is able to generate molecules that match the pharmacophore model and meet the overall criteria for TSPA, SA, and hERG.
+
+Lead compound optimization
+
+Lead optimization refers to the improvement of one or more properties of a hit compound by chemical modification. The optimization objectives include adjusting the molecular flexibility ratio, improving the pharmacokinetic properties, or enhancing the bioavailability. Here we show how PGMG can help with lead compound optimization using Lavendustin A as a case study. Lavendustin A is an inhibitor of epidermal growth factor receptor (EGFR), while the lipophilicity of Lavendustin A is too poor to cross the cell membrane. It has been shown that improving the lipophilicity of Lavendustin A can lead to nanomolar levels of IC50 inhibition activity at the cellular level^{46}. In this case, we construct Lavendustin A's pharmacophore using Pharao^{47}, then we modify the polar pharmacophore features, and finally use PGMG to generate molecules for the modified pharmacophore to improve the lipophilicity of the generated molecules.
+
+We filter the generated molecules with lipophilicity (LogP > 3.41) to obtain 400 molecules with a higher lipophilicity than Lavendustin A. We calculate Tanimoto similarity using MACCSkeys Fingerprints with RDKit^{31} between the obtained molecules and the 1500 EGFR inhibitors acquired from the ExCAPE database^{48}. **Figure 6** shows some examples of the generated molecules with their closest EGFR inhibitors obtained from the ExCAPE database and their respective Tanimoto similarities. We find that the generated molecules have high similarity to the EGFR target active molecules in the ExCAPE database, which are not included in the training set. And they all own the three pharmacophore features of the aromatic ring, hydrogen-bonded acceptor, and hydrophobic center. Based on the assumption that structurally similar molecules have similar properties, the similarity result demonstrates that molecules generated by PGMG have a probability of inhibiting EGFR. To some extent, the generated molecules gain structural diversity while maintaining the consistency of the pharmacophore. Overall, PGMG can optimize certain properties and maintain the bioactivity of a given lead compound.
+Figure 6 | Display diagram of the molecule generated by PGMG with known inhibitors in the case of Lavendustin A optimization. Molecules generated by PGMG are shown inside the circle and their closest active nearest neighbors are shown outside the circle. The colors indicate the pharmacophore features extracted from Lavendustin A. Red corresponds to the aromatic center, blue represent the hydrogen-bonded acceptor, and green represent the hydrophobic center.
+
+Discussion
+In this work, we develop a pharmacophore-guided deep learning approach for bioactive molecule generation called PGMG. We manage to use pharmacophores as the only constraint during the generation process by (1) encoding both pharmacophore features and spatial information of a given pharmacophore into a complete graph with node and edge attributes and (2) introduce latent variables so that a molecule can be uniquely characterized by a pharmacophore and a set of latent variables to handle the many-to-many relationship of pharmacophores and molecules. Our approach offers some advantages over current molecule generation methods. Firstly, PGMG provides a way to utilize different types of activity data in a uniform representation, allowing it to overcome the problem of data scarcity and be used in various situations. Secondly, pharmacophores are biologically meaningful and can incorporate biochemists’ knowledge, which provides a strong prior and certain interpretability into the generation process. Lastly, after training, PGMG can be directly applied to different targets without further fine-tuning. Besides, it is also worth mentioning that the training scheme itself does not require any activity data to proceed. This training scheme may be useful for other generative methods.
+
+PGMG makes solid progress on the challenging problem of generating desired bioactivate molecules in various scenarios when known active data is scarce. When a target structure is available, PGMG is competent to design a large number of molecules that bind affinity better than the specific-target active
+molecules obtained from the ChEMBL database. Given the pharmacophore for certain targets, the PGMG can also be utilized to design dual or multi-target molecules. Besides, we expect that PGMG can be adopted to prepare chemical libraries to replace those used in HTS campaigns to improve virtual screen efficiency, as it can provide a sufficient number of candidate drug-like molecules for a specified target. Then, this method performs well in ligand-based drug design, which has wide use in drug design when the target structure is absent. The ligand-based case shows that PGMG is able to generate high-quality bioactivity molecules that match the pharmacophore model with structural diversity. This result implies that PGMG can be applied to multiple drug design scenarios such as researching alternative medicine, drug resistance, and scaffold hopping. Finally, the case of lead optimization demonstrates that PGMG can optimize the molecule's properties while maintaining the bioactivity and scaffold diversity of the generated molecules. The results demonstrate that PGMG is a promising approach for structure-based drug design, high-throughput screening, ligand based-drug design, and lead optimization in a real drug discovery setting.
+
+We believe that the de novo drug design is a complicated and situation-specific problem, and computational methods should try to get more assistance from chemists’ experience and judgements. PGMG benefits from this idea a lot. Some limitations of PGMG should be acknowledged. PGMG currently does not support exclusion volume in pharmacophore models and as we focus on the task of generating molecules with desired activity, PGMG does not explicitly constrain the properties of the generated molecules. A future direction of our work is to include the exclusion volume and other features into PGMG and make the generated molecules to be more controllable and malleable. And furthermore, designing multi-conditional generation models to generate active molecules with specified properties is the ultimate goal of drug design, we will continue to work towards this objective.
+
+Methods
+
+Datasets
+We use the ChEMBL 24 dataset containing more than 1.25 million molecules to train PGMG. ChEMBL is a collection of bioactivity data for various targets and compounds from the literature. It contains 13 types of atoms (T = 13): H, B, C, N, O, F, Si, P, S, Cl, Se, Br, and I. Each bond is either a no-bond, single, double, triple or aromatic bond (R = 5).
+
+We also use the ZINC36 molecule dataset from JTVAE49 for our ablation study. It contains 220,000 molecules in the training data, 11 types of atoms (T = 11): H, B, C, N, O, F, P, S, Cl, Br, and I. Each bond is either a no-bond, single, double, triple or aromatic bond (R = 5).
+
+The structure of four targets VEGFR2, CDK6, TFGβ 1, BRD4 are downloaded from PDB50 database.
+
+Representation of Pharmacophores and Molecules
+A pharmacophore model consists of several chemical features and their spatial descriptions and are represented by a fully connected graph with chemical feature types as node attributes and distances as edge weights (a detailed description of the pharmacophore graph and the preparation of the pharmacophore graph is included in SI). We apply a state-of-the-art graph neural network, Gated-GCN29, to embed the graph with consideration of node attributes and edge attributes.
+Molecules are represented in SMILES format. Symbols of stereochemistry like ‘@’ ‘/’ are removed because stereochemistry information does not exist in the graph representation of a pharmacophore and it is not difficult to list all stereoisomers of a molecule. Then the SMILES string is separated into a sequence of tokens corresponding to heavy atoms and structural punctuation marks. For example, the SMILES string \( C(C[NH2-])OC(=O)Cl \) will be split to \( C\ (C\ [NH2-]\ )\ O\ C\ (= O )\ Cl \), where each token will be embedded into a vector.
+
+Encoder and Decoder
+An illustration of the encoder and decoder networks can be found in Figure.1. The encoder and decoder network are adapted from the standard transformer\(^{30}\) architecture with each consisting of several layers of stacked transformer encoder block and transformer decoder block. The difference between the transformer encoder and decoder blocks is that the encoder block uses only self-attention modules and the decoder block uses cross-attention modules to incorporate context in the generation process. Some modifications are made to handle our inputs and to better suit the variational autoencoder structure of PGMG.
+
+We first calculate the latent variables z of molecule x given pharmacophore c by the encoder network. The encoder input is a concatenation of molecule and pharmacophore features. Following BART\(^{32}\), positional and segment encoding is added to the input sequence:
+
+\[
+Input_{encoder} = (E_p'; E'_m)
+\] (2)
+
+\[
+E'_{m_i} = E_{m_i} + SE_m + PE_i
+\] (3)
+
+\[
+E'_{p_j} = E_{p_j} + SE_p
+\] (4)
+
+where \( Input_{encoder} \) is the input representation, \( E_{p_j} \) is the j-th pharmacophore feature vector, \( E_{m_i} \) is the i-th token embedding of molecule features, \( SE_m \) and \( SE_p \) are two segment embedding vectors for molecule features and pharmacophore features, and \( PE_i \) is the positional embedding for i-th token. After several layers of transformer encoder block, the molecule features are averaged by an attention pooling layer to obtain the final molecule representation \( h_x \). \( h_x \) is then fed into two separate sub-networks to compute the mean \( \mu \) and log variance \( \log \Sigma \) of the posterior variational approximation. Latent variables z are then sampled from the Normal distribution \( N(\mu, \Sigma) \).
+
+The decoder network takes the latent variables z and pharmacophore features as input:
+
+\[
+input_{decoder} = (E'_p; E'_z)
+\] (5)
+
+\[
+E'_{z_i} = z_i + SE_z + PE_i
+\] (6)
+
+where \( E'_p \) is the same as described above, \( SE_z \) is the segment embedding for latent variables, and \( PE_i \) is the positional embedding for i-th token. The decoder then uses \( input_{decoder} \) to generate target SMILES autoregressively. Each token is determined on the basis of previously generated tokens and context:
+
+\[
+o_i = (\text{argmax})_o P(o_i|o_{<i},c,z)
+\] (7)
+
+where \( o_i \) is i-th generated token.
+
+Loss Function
+PGMG’s model is trained in an end-to-end manner. The Loss function consists of three different terms, KL Loss, LM Loss, and the mapping loss. The first two terms are the negative evidence lower bound (ELBO) of
+the log likelihood \( \log P_\theta(x|c) \):
+
+\[
+\begin{align*}
+\log P_\theta(x|c_p) &= \log \int P_\theta(x|c,z)P_\phi(z|c)dz \\
+&\geq -KL(P_\phi(z|x,c)||P_\theta(z|c)) + E_{P_\phi(z|x,c)}[\log P_\theta(x|z,c)] \tag{A} \\
+&\approx -KL(P_\phi(z|x,c)||P_\theta(z|c)) + \log P_\theta(x|z,c) \tag{B}
+\end{align*}
+\]
+
+where \( KL \) denotes the Kullback-Leibler divergence and we assume \( P_\theta(z|c) \) the prior distribution of \( z \) to be a standard gaussian \( N(0,I) \). We call the left part of (A) KL Loss and it serves as a regulation term to mitigate the gap between the true prior distribution of \( z \) and the posterior distribution and to make the latent space of \( z \) smoother. The expectation term on the right part of (A) is estimated through sampling, and it is optimized using Monte Carlo estimation with one data point for each sample\(^{51}\). This gives us the right part of (B). Since \( m \) takes form of the SMILES string, we call it the language modeling loss (LM Loss).
+
+The third part of PGMG’s loss function is the mapping loss. It evaluates the model’s performance in predicting the mapping between heavy atoms and pharmacophore elements. We use the mapping loss as a regulation term to help alleviate the problem of posterior collapse. The mapping score of the \( i^{th} \) pharmacophore \( p_i \) and the \( j^{th} \) output token \( o_j \) is calculated as
+
+\[
+mapping_{score_{ij}} = \sigma \left( g(W_pE_{p_i}) \odot g(W_oE_{o_j}) \right)
+\]
+
+where \( E_{p_i} \) and \( E_{o_j} \) are the embedding vectors of \( p_i \) and \( o_j \) respectively, \( W_p \) and \( W_o \) are two learnable matrices to project two different embeddings into the same space, \( \odot \) is the dot product, \( \sigma \) is the sigmoid function, and \( g \) is the ReLU function. The calculation of mapping scores can be vectorized as
+
+\[
+mapping_{score} = \sigma \left( g(W_pE_p) \; g(W_oE_o) \right)
+\]
+
+Since SMILES format contains tokens other than atom symbols, we mask them when calculating the mapping loss. The mapping loss is then calculated as the cross-entropy of the masked scores and labels. An illustration of the masked mapping score and label is given in Supplementary Figure S3.
+
+**Training details and model parameter settings**
+During training, we inject noise into the input to make training more robust by using the infilling scheme. Some random subsequences in every input sequence are replaced with a single *mask* token. Teacher forcing technique is applied to the generation process during training, by which we replace the previously generated tokens with the ground truth to produce the next token. Aside from the mapping loss introduced before, another approach we use to alleviate posterior collapse is KL annealing\(^{52}\), where an increasing coefficient is used to control the size of KL Loss.
+
+We use the same model parameters in both ChEMBL and ZINC datasets. The hidden dimension is 384. The transformer encoder blocks and transformer decoder blocks are stacked 8 times. We use an 8-head attention and the feed-forward dimension is 1024. We use Adam optimizer to train the model with a 3e-4 learning rate and a 1e-6 weight decay rate. Cosine learning rate annealing is applied with a cycle length of 4 epochs. We use the gradient clipping technique and set the maximum gradient to be 5. Since the ChEMBL dataset contains a lot more molecules compared to the ZINC dataset, it requires less training epochs to reach
+a similar validation performance. Thus, the number of training epochs for the former is 32 and 48 for the latter.
+
+Evaluation
+Firstly, four different metrics on 2D level, validity, uniqueness, novelty, and ratio of available molecules are employed to evaluate the ability of the PGMG to generate novel molecules. Validity is the percentage of chemically valid molecules with generated SMILES. Uniqueness measures how many valid molecules are non-repetitive. Novelty refers to the percentage of generated chemically valid molecules not present in the training set. And the ratio of available molecules is the proportion of novel molecules in all generated results. These metrics are calculated as follows:
+
+\[
+validity = \frac{\#Number\ of\ chemically\ valid\ SMILES}{\#of\ generated\ SMILES}
+\] (11)
+
+\[
+uniqueness = \frac{\#of\ non-duplicate,\ valid\ SMILES}{\#of\ valid\ SMILES}
+\] (12)
+
+\[
+novelty = \frac{\#of\ novelty\ molecules\ not\ in\ training\ set}{\#of\ unique\ molecules}
+\] (13)
+
+\[
+ratio\ of\ available\ molecules = \frac{\#of\ novel\ molecules}{\#of\ generated\ SMILES}
+\] (14)
+
+Secondly, goal-directed metrics are evaluated by the match score, which indicates the match degree of the generated molecules to the specified pharmacophore (see calculation of match score section of the Supplementary Information for details). We further evaluate the binding activity of the generated molecules to the target using affinity calculated by Autodock vina^{37}. Finally, we use ADMETlab 2.0^{38} to predict the ADMET properties of the generated molecules and to assess the drug-like potential of the generated molecules.
+
+Acknowledgements
+
+This work is financially supported by the National Natural Science Foundation of China under Grants (No. 61832019 to M.L.), Hunan Provincial Science and Technology Program (2019CB1007) [M.L.], and European Research Council (No. 716063 to J.T.)
+
+Author Contributions
+
+M.L. and J.T. guided the research and provided the experimental platform. H.Z., R.Z, and M.L conceived the initial idea and started the project. H.Z collected and preprocessed the data and R.Z designed the model. R.Z performed the generation experiments and H.Z performed the case studies. H.Z, R.Z, J.T., and M.L wrote the paper.
+
+Declaration of Interests
+
+The authors declare no competing interests.
+
+Data availability
+The data that support the findings of this study are available at https://github.com/CSUBioGroup/PGMG.
+
+Code availability
+The code used to generate results shown in this study is available at https://github.com/CSUBioGroup/PGMG.
+
+reference
+
+1. Lipinski, C.A., Lombardo, F., Dominy, B.W. & Feeney, P.J. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Advanced drug delivery reviews 23, 3-25 (1997).
+2. Bohacek, R.S., McMartin, C. & Guida, W.C. The art and practice of structure - based drug design: a molecular modeling perspective. Medicinal research reviews 16, 3-50 (1996).
+3. Goodnow Jr; R.A. Hit and lead identification: Integrated technology-based approaches. Drug Discovery Today: Technologies 3, 367-375 (2006).
+4. Gómez-Bombarelli, R. et al. Automatic chemical design using a data-driven continuous representation of molecules. ACS central science 4, 268-276 (2018).
+5. Jin, W., Barzilay, R. & Jaakkola, T. Multi-objective molecule generation using interpretable substructures. International conference on machine learning. PMLR, 4849-4859 (2020).
+6. Zhou, Z., Kearnes, S., Li, L., Zare, R.N. & Riley, P. Optimization of molecules via deep reinforcement learning. Scientific reports 9, 1-10 (2019).
+7. Wang, J. et al. Multi-constraint molecular generation based on conditional transformer, knowledge distillation and reinforcement learning. Nature Machine Intelligence 3, 914-922 (2021).
+8. Maziarka, Ł. et al. Mol-CycleGAN: a generative model for molecular optimization. Journal of Cheminformatics 12, 1-18 (2020).
+9. Guimaraes, G.L., Sanchez-Lengeling, B., Outeiral, C., Farias, P.L.C. & Aspuru-Guzik, A. Objective-reinforced generative adversarial networks (ORGAN) for sequence generation models. Preprint at http://arxiv.org/abs/1705.10843 (2017).
+10. Fu, T., Xiao, C. & Sun, J. Core: Automatic molecule optimization using copy & refine strategy. Proceedings of the AAAI Conference on Artificial Intelligence 34, 638-645 (2020).
+11. Mahmood, O., Mansimov, E., Bonneau, R. & Cho, K. Masked graph modeling for molecule generation. Nature communications 12, 1-12 (2021).
+12. Amabilino, S., Pogány, P., Pickett, S.D. & Green, D.V. Guidelines for recurrent neural network transfer learning-based molecular generation of focused libraries. Journal of Chemical Information and Modeling 60, 5699-5713 (2020).
+13. Tripp, A., Chen, W. & Hernández-Lobato, J.M. An evaluation framework for the objective functions of de novo drug design benchmarks. ICLR2022 Machine Learning for Drug Discovery (2022).
+14. Tkatchenko, A. Machine learning for chemical discovery. Nature communications 11,(2020).
+15. Elton, D.C., Boukouvalas, Z., Fuge, M.D. & Chung, P.W. Deep learning for molecular design—a review of the state of the art. Molecular Systems Design & Engineering 4, 828-849 (2019).
+16. Walters, W.P. & Murcko, M. Assessing the impact of generative AI on medicinal chemistry. Nature biotechnology 38, 143-145 (2020).
+17. Imrie, F., Hadfield, T.E., Bradley, A.R. & Deane, C.M. Deep generative design with 3D pharmacophoric constraints. Chemical science 12, 14577-14589 (2021).
+18. Méndez-Lucio, O., Baillif, B., Clevert, D.-A., Rouquié, D. & Wichard, J. De novo generation of hit-like molecules from gene expression signatures using artificial intelligence. Nature communications 11, 1-10 (2020).
+19. Imrie, F., Bradley, A.R., van der Schaar, M. & Deane, C.M. Deep generative models for 3D linker design. Journal
+of chemical information and modeling **60**, 1983-1995 (2020).
+20. Yang, Y. et al. SyntaLinker: automatic fragment linking with deep conditional transformer neural networks. *Chemical science* **11**, 8312-8322 (2020).
+21. Li, Y., Pei, J. & Lai, L. Structure-based de novo drug design using 3D deep generative models. *Chemical science* **12**, 13664-13675 (2021).
+22. Luo, S., Guan, J., Ma, J. & Peng, J. A 3D Generative Model for Structure-Based Drug Design. *Advances in Neural Information Processing Systems* **34**, 6229-6239 (2021).
+23. Pogány, P., Arad, N., Genway, S. & Pickett, S.D. De novo molecule design by translating from reduced graphs to SMILES. *Journal of chemical information and modeling* **59**, 1136-1146 (2018).
+24. Skalic, M., Jiménez, J., Sabbadin, D. & De Fabritiis, G. Shape-based generative modeling for de novo drug design. *Journal of chemical information and modeling* **59**, 1205-1214 (2019).
+25. Schneidman-Duhovny, D., Dror, O., Inbar, Y., Nussinov, R. & Wolfson, HJ. PharmaGist: a webserver for ligand-based pharmacophore detection. *Nucleic acids research* **36**, W223-W228 (2008).
+26. Wang, X. et al. PharmMapper 2017 update: a web server for potential drug target identification with a comprehensive target pharmacophore database. *Nucleic acids research* **45**, W356-W360 (2017).
+27. Ma, Z. et al. Pharmacophore hybridisation and nanoscale assembly to discover self-delivering lysosomotropic new-chemical entities for cancer therapy. *Nature communications* **11**, 1-12 (2020).
+28. Meslamani, J. et al. Protein-ligand-based pharmacophores: generation and utility assessment in computational ligand profiling. *Journal of chemical information and modeling* **52**, 943-955 (2012).
+29. Bresson, X. & Laurent, T. Residual Gated Graph ConvNets. *Preprint at https://arxiv.org/abs/1711.07553* (2017).
+30. Vaswani, A. et al. Attention is all you need. *Advances in neural information processing systems* **30** (2017).
+31. Landrum, G. A. RDKit: Open-source cheminformatics. http://www.rdkit.org.
+32. Lewis, M., Liu, Y., Goyal, N., Ghazvininejad, M. & Zettlemoyer, L. BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension. *Preprint at http://arxiv.org/abs/1910.13461* (2019).
+33. Segler, M.H., Kogej, T., Tyrchan, C. & Waller, M.P. Generating focused molecule libraries for drug discovery with recurrent neural networks. *ACS central science* **4**, 120-131 (2018).
+34. Mendez, D. et al. ChEMBL: towards direct deposition of bioassay data. *Nucleic acids research* **47**, D930-D940 (2019).
+35. Brown, N., Fiscato, M., Segler, M.H. & Vaucher, A.C. GuacaMol: benchmarking models for de novo molecular design. *Journal of chemical information and modeling* **59**, 1096-1108 (2019).
+36. Sterling, T. & Irwin, JJ. ZINC 15-ligand discovery for everyone. *Journal of chemical information and modeling* **55**, 2324-2337 (2015).
+37. Trott, O. & Olson, AJ. AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. *Journal of computational chemistry* **31**, 455-461 (2010).
+38. Xiong, G. et al. ADMETlab 2.0: an integrated online platform for accurate and comprehensive predictions of ADMET properties. *Nucleic Acids Research* **49**, W5-W14 (2021).
+39. Lee, K. et al. Pharmacophore modeling and virtual screening studies for new VEGFR-2 kinase inhibitors. *European journal of medicinal chemistry* **45**, 5420-5427 (2010).
+40. Shawky, A.M., Ibrahim, N.A., Abourehab, M.A., Abdalla, AN. & Gouda, A.M. Pharmacophore-based virtual screening, synthesis, biological evaluation, and molecular docking study of novel pyrrolizines bearing urea/thiourea moieties with potential cytotoxicity and CDK inhibitory activities. *Journal of enzyme inhibition and medicinal chemistry* **36**, 15-33 (2021).
+41. Jiang, J., Zhou, H., Jiang, Q., Sun, L. & Deng, P. Novel transforming growth factor-beta receptor 1 antagonists through a pharmacophore-based virtual screening approach. *Molecules* **23**, 2824 (2018).
+42. Yan, G. et al. Pharmacophore - based virtual screening, molecular docking, molecular dynamics simulation, and biological evaluation for the discovery of novel BRD 4 inhibitors. *Chemical Biology & Drug Design* **91**, 478-490 (2018).
+43. Pei, J., Yin, N., Ma, X. & Lai, L. Systems biology brings new dimensions for structure-based drug design. Journal of the American Chemical Society **136**, 11556-11565 (2014).
+44. Kermani, F. et al. In vitro activities of antifungal drugs against a large collection of Trichophyton tonsurans isolated from wrestlers. Mycoses **63**, 1321-1330 (2020).
+45. Wishart, D.S. et al. DrugBank 5.0: a major update to the DrugBank database for 2018. Nucleic acids research **46**, D1074-D1082 (2018).
+46. Nussbaumer, P. et al. Novel antiproliferative agents derived from lavendustin A. Journal of medicinal chemistry **37**, 4079-4084 (1994).
+47. Taminau, J., Thijis, G. & De Winter, H. Pharao: pharmacophore alignment and optimization. Journal of Molecular Graphics and Modelling **27**, 161-169 (2008).
+48. Sun, J. et al. ExCAPE-DB: an integrated large scale dataset facilitating Big Data analysis in chemogenomics. Journal of cheminformatics **9**, 1-9 (2017).
+49. Jin W, Barzilay R, Jaakkola T. Junction Tree Variational Autoencoder for Molecular Graph Generation. Proceedings of the 35th International Conference on Machine Learning. PMLR,2323-2332(2018).
+50. Burley, S.K. et al. Protein Data Bank (PDB): the single global macromolecular structure archive. Protein Crystallography **1607**, 627-641 (2017).
+51. Kingma, D.P. & Welling, M. Auto-Encoding Variational Bayes. Preprint at http://arxiv.org/abs/1312.6114 (2014).
+52. Bowman, S.R. et al. Generating sentences from a continuous space. Preprint at http://arxiv.org/abs/1511.06349 (2015).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SupplementaryinformationPGMGAPharmacophoreGuidedDeepLearningApproachforBioactiveMolecularGeneration.docx

08a32d4f18a4d90852d0a1f5f66103659aefcc1d8e292575b94fd513a2b10b92/peer_review/peer_review.md

@@ -0,0 +1,354 @@
+A draft UAE-based Arab pangenome reference
+
+Corresponding Author: Dr Mohammed Uddin
+
+This manuscript has been previously reviewed at another journal. This document only contains information relating to versions considered at Nature Communications. Mentions of prior referee reports have been redacted.
+
+Parts of this Peer Review File have been redacted as indicated to remove third-party material.
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.
+
+Version 0:
+
+Reviewer comments:
+
+Reviewer #1
+
+(Remarks to the Author)
+The authors have done a good job of addressing my major concerns. I'm very glad to see the assemblies are now being openly shared. Given that the assemblies will be made openly available, I would only request that the publication be released with the GenBank accessions of the assemblies.
+
+There are a few rough edges in the text - e.g., I noted CPC was not disambiguated when it was first used, and in one place, the text talks about 100 plus reference diploid Arab genome assemblies, when, in fact, it is talking about 100 plus haplotypes. These small issues should be carefully checked.
+
+Altogether though, I think this paper represents a new and significant resource that will be of great value for the community.
+
+(Remarks on code availability)
+
+Reviewer #2
+
+(Remarks to the Author)
+Reviewer comments to the authors
+
+Genomic references from underrepresented populations including the Middle East are valuable to advance genomic research regionally and in the world. The current study aims to establish a pangenome reference for 'Arab populations' based on de novo assembled genomes form 53 individuals recruited from the United Arab Emirates (UAE). The authors use established methods to generate and process data, producing de novo assemblies which they incorporate to a pangenome graph and highlight some features from it. In principle the idea is valuable, however the implementation shows several issues that persist despite the multiple revisions. This includes lack of novel insights, weakness/inconsistency in population structure and other analyses, over statements on novelty, as well as ambiguity on ethical processes for data sharing.
+Notably, there is lack of purposeful study design to ensure the selection of representative ancestries that would be needed to cover wide Arab populations beyond UAE. As a result, this draft pan genome should be named UAE pan genome instead of Arab pangenome. The below elaborates on the main points.
+
+Samples and study design
+
+• The samples in this study were not targeted to be representative based on genetic ancestry. The 53 subjects were recruited from the UAE, whose population represent less than 1.9% of all Arab populations. The authors added a few expats from the region but still recruited form the UAE instead of engaging collaborators from other counties who know better about the local diversity. For example, two Moroccan nationals were but is well known that Morocco is predominantly native Berber (not Arab). See for example HPRC protocol that was established for sample selection to ensure added value in terms capturing diversity.
+• In previous submissions authors indicated data was under restricted access due to IRB. Since authors now say informed written consent has been obtained, they should share these documents to ensure the protection of human subjects in this study.
+• Authors mention study subjects were randomly selected from clinics in Dubai. Amongst the 53 individuals at least a few would have been expected to have some common complex disorder phenotypes. However none are which is contradictory.
+• Data availably does not mention anything about HiC and raw PacBio and ONT read data.
+
+Population structure
+
+Presents to be severely undeveloped:
+• The admixture analysis that was originally done had multiple inconsistencies with what has been previously established in the literature, and that persists here. Now, the admixture plots are omitted entirely from main and supplementary figures.
+• The PCA shown in Figure 1e is still problematic showing East Asian clustered next to Americans. Notably, the UAE samples shown in red are shifted to the right in comparison to the reference Arab samples shown in pink. This reflects the lack of a priori proper selection of samples to ensure wide representation.
+
+Data generation
+
+• Given that various data types have been generated for the subjects in this study, there is no attempt to check for sample swaps and contamination
+
+Assembly metrics
+
+• There is mention of exclusion of “non-human eukaryotic pathogen genomes” in addition to 109 contigs from ChrM and exclusion of 4 contigs from other chromosomes based on sequence identity. What is the total length of what was excluded? From which individual assemblies? And what is the reason for the presence of pathogenic sequences? What is the possibility that some of what was labelled as mis-joining could be a genuine chromosomal rearrangement?
+• In Suppl Table 11, authors report #Ns, #indels and mismatches per 100kb. Should report the total numbers instead.
+• In Suppl Table 11, the numbers of misassembled contigs relative to CHM13 as a fraction from number of contigs is on average 56%. This is too high.
+• Authors make the statements “All samples surpassed the 40 Mb N50 of the HPRC reference genome”. However the N50 they report include the misassembled conigs shown in Supl Table 11.
+• QV for base pair quality was it calculated using the long read data used for the assemblies? If yes, what bias this could constitute?
+• Unclear how the coverage was computed exactly.
+• The reported average size of 50/77 Mb of contigs that did not align to CHM13/GRCh38 does that include the misassembled contigs mentioned previously? Same question for the results reported regarding phasing quality etc
+
+Gene duplication analysis
+
+• This section shows several inconsistencies. Results were generated showing multiple copies of genes across the assemblies (Suppl Table 14a), but authors only discuss the duplications (Fig 2g). There is large variance in the number of genes with additional copies relative to GRCH38, ranging from 24 to 206 genes. The number of additional copies itself ranges between 1 and 46. Some samples have unrealistically high number of impacted genes such as APR_023_hap2 which has 206 genes with additional multiple copies relative to GRCH38, which is 4 times higher than the average, while hap 1 of the same sample has only 43 genes! Ultimately by looking at the trio, it is odd to see that the child has much lower number of genes with multiple copies in comparison to the father and the mother.
+• In view of the above, I wonder how significant are the results of the comparison with HPRC.
+
+Variant analysis
+
+• No mention of what reference the variant calls were made against and if any filtering was applied including the low-quality regions of the assemblies.
+• The authors report a high proportion of novel variants, including for SNVs and Indels where it is 15-19% per genome. This is odd because for this type of variation there is genomic completeness in the references used, and samples from Middle East have already been extensively sampled in the literature. The authors do not attempt to check in GNOMAD and other Middle Eastern cohorts such as Mineral et al. 2021 and Scott et al. 2016.
+• For the SV calls, there is no attempt to QC the calls which are more prone to mis-assembly errors
+
+Novel euchromatic sequences
+
+• Authors report their assemblies contain 112 Mbs of novel sequence relative to CHM13, other pan genomes, and 1.4 kb of Mt DNA. This amounts to ~ 4% of the human genome which seems excessive and may not as striking. How much does that compare to the inverse situation i.e the amount of novel sequences in HPRC samples relative to the samples in this study?
+
+Complex structural variation
+
+• Strange the criteria used here for naming this category of SVs. Authors should just refer to SVs larger than certain length and frequent above a certain threshold.
+• Again novelty claimed without assessing relevant datasets beyond HPRC+CPC.
+
+Mitochondrial pangeome
+
+• No mention which reference variants were called against
+• The Mt assemblies are based on PacBio only. No mention of what particular QC was performed for these genomes nor what mis-assemblies were found
+
+Performance gain from pangenome-aided
+
+• Higher calling rate using Pan genomes has been shown before so that is not specific to the APR graph.
+• The mapping rate is between the APR and HPRC is almost identical. Unclear what is the added value here
+
+Discussion
+
+To be revised based on the comments above.
+
+(Remarks on code availability)
+
+Version 1:
+
+Reviewer comments:
+
+Reviewer #2
+
+(Remarks to the Author)
+
+(Remarks on code availability)
+To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
+Reviewer’s Comments:
+
+Reviewer #1 (Remarks to the Author)
+
+The authors have done a good job of addressing my major concerns. I'm very glad to see the assemblies are now being openly shared. Given that the assemblies will be made openly available, I would only request that the publication be released with the GenBank accessions of the assemblies.
+
+There are a few rough edges in the text - e.g., I noted CPC was not disambiguated when it was first used, and in one place, the text talks about 100 plus reference diploid Arab genome assemblies, when, in fact, it is talking about 100 plus haplotypes. These small issues should be carefully checked.
+
+Altogether though, I think this paper represents a new and significant resource that will be of great value for the community.(Remarks on code availability)
+
+Response: We thank the reviewer for their thoughtful feedback and constructive suggestions. We are pleased to have addressed the major concerns raised. As requested, we have included the GenBank accession numbers for the assemblies in the supplementary table (Supplementary Table 35). We have carefully reviewed the final manuscript to correct any typographical errors. The abbreviation "CPC" is clarified as "Chinese Pangome Consortium" at its first mention in the main section. Furthermore, the reference to "106 assemblies" has been corrected to "106 haplotypes".
+
+Reviewer #2 (Remarks to the Author)
+
+Reviewer comments to the authors
+
+Genomic references from underrepresented populations including the Middle East are valuable to advance genomic research regionally and in the world. The current study aims to establish a pangenome reference for ‘Arab populations’ based on de novo assembled genomes form 53 individuals recruited from the United Arab Emirates (UAE). The authors use established methods to generate and process data, producing de novo assemblies which they incorporate to a pangenome graph and highlight some features from it. In principle the idea is valuable, however the implementation shows several issues that persist despite the multiple revisions. This includes lack of novel insights, weakness/inconsistency in population structure and other analyses, over statements on novelty, as well as ambiguity on ethical processes for data sharing. Notably, there is lack of purposeful study design to ensure the selection of representative ancestries that would be needed to cover wide Arab populations beyond UAE. As a result, this draft pan genome should be named UAE pan genome instead of Arab pangenome. The below elaborates on the main points.
+Samples and study design
+
+• The samples in this study were not targeted to be representative based on genetic ancestry. The 53 subjects were recruited from the UAE, whose population represent less than 1.9% of all Arab populations. The authors added a few expats from the region but still recruited form the UAE instead of engaging collaborators from other counties who know better about the local diversity. For example, two Moroccan nationals were but is well known that Morocco is predominantly native Berber (not Arab). See for example HPRC protocol that was established for sample selection to ensure added value in terms capturing diversity.
+
+Response: We thank the reviewer for raising this important point. However, we respectfully contend that the draft pangenome we constructed covers a significant gap in the global genomic literature despite the relatively small sample size. This is not very different from previous important work in other populations. For example, HPRC developed a draft global pangenome using 47 samples, including only 2% from Europe and just one sample from South Asia, a region with a population exceeding 2 billion. Similarly, the Chinese Pangenome includes one or two samples from most of its geographically distinct populations. While the sample sizes for these pangenomes remain small, the assemblies successfully capture common sequence diversity across a wide range of ethnicities and global populations. Likewise, the Arab Pangenome includes samples from eight Arab countries, representing most of the major populations within the MENA region.
+
+The point regarding the two Moroccan nationals suggests a conflation of genetics, ethnicity and language. The population structure of Moroccans (Berber/Arab) has been published extensively, and they show the same ancestral components (Arauna et al., 2017), “the present analysis of additional Berber samples reinforces the idea of no strong genetic distinction between Arabs and most Berber groups.”). Second, in contrary to the statement, Morocco is not predominately “native Berber (not Arab)”, as two thirds of the population identify as Arab. Consequently, both Arab and Beber groups would benefit from our pangenome data, which is crucial as there is no high-quality pangenome including Moroccan samples published so far. We present below an ADMIXTURE plot of our Moroccan samples with published Moroccan samples genotyped on the Human Origins Array from the Allen Ancient DNA Resource (AADR) (Figure 1). As shown, the ADMIXTURE plots are very similar, illustrating that our samples capture the variation found in Moroccans.
+Figure 1: Population genetic ancestry inference for the Moroccan samples of the APR cohort. The ADMIXTURE plot shows ancestry components at K=7, K=8, and K=9 for two Moroccan samples from the Arab Pangenoome Project alongside Moroccan samples from AADR dataset. Each bar represents an individual sample, with the proportions of ancestral components represented by different colors.
+
+We would also like to emphasize that the ancestral ancient populations that gave rise to modern-day North African and Middle Eastern groups are similar, despite the geographic range. All modern-day Middle East samples (Egypt, Levant, Arabia etc) are a mixture of four ancient populations (Almarri et al., 2021), with current-day structure/cline resulting due slight differences of these source populations. Consequently, due to the ancestral history shared by all these groups, we believe that the pangenome will be relevant to all Middle Eastern and Arab populations.
+
+• In previous submissions authors indicated data was under restricted access due to IRB. Since authors now say informed written consent has been obtained, they should share these documents to ensure the protection of human subjects in this study.
+
+Response: We thank the reviewer for raising this point regarding data accessibility and consent. To clarify, at no point did we mention that the data was under restricted access. Our original intention was always to make the data publicly available upon the publication of the manuscript.
+However, following the first review, we expedited this process and made the raw data publicly accessible at the review stage.
+
+For reference, we have enclosed our earlier communication from the first rebuttal, which clearly outlines our actions to ensure data availability. Additionally, we confirm that copies of the informed consent documents will be provided to the editor to ensure transparency and compliance with ethical standards. The relevant part of the consent form reads:
+
+“As part of this study, my de-identified data (i.e., data that does not include my name or any personal identifiers) will be uploaded to public research repositories. This data will be accessible to other researchers for future scientific studies aimed at advancing our understanding of genetics and related fields. My privacy will be protected, and no personally identifiable information will be shared.”
+
+[REDACTED]
+
+Snapshot to our previous response.
+• Authors mention study subjects were randomly selected from clinics in Dubai. Amongst the 53 individuals at least a few would have been expected to have some common complex disorder phenotypes. However none are which is contradictory.
+
+Response: We would like to clarify that random sampling was never mentioned in our study. Instead, our study employed convenience sampling with specific criteria to exclude individuals with chronic diseases. This approach ensured a cohort of apparently healthy individuals to establish a baseline for the Arab pangenome, as detailed in the methods and supplementary methods section (Page 4).
+
+• Data availably does not mention anything about HiC and raw PacBio and ONT read data.
+
+Response: All raw sequencing data, including Hi-C, PacBio, and ONT reads, have been uploaded to the global Sequence Read Archive (SRA) under the accession number PRJNA1108179.
+
+The current data availability statement in the manuscript reads as follows:
+“We have submitted our sequencing data (including PacBio, ONT) for all the samples to the global Sequence Read Archive (SRA) repository that can be openly accessed and downloaded under the accession no PRJNA1108179.”
+
+We revised this statement to include Hi-C data for completeness:
+“We have submitted raw sequencing data (including PacBio, ONT, Hi-C) for all the samples to the global Sequence Read Archive (SRA) repository that can be openly accessed and downloaded under the accession no PRJNA1108179.”
+
+Additionally, details about the assemblies, pangenome, and VCF files can be accessed via the links provided in the data availability section.
+
+Population structure
+
+Persists to be severely undeveloped:
+• The admixture analysis that was originally done had multiple inconsistencies with what has been previously established in the literature, and that persists here. Now, the admixture plots are omitted entirely from main and supplementary figures.
+
+Response: We had updated the ADMIXTURE analysis and included the results in our previous submission, presented as Extended Data Fig. 1 (attached below). This analysis, consistent with the literature, highlights the APR cohort's genetic diversity in the context of global populations. Using linkage disequilibrium (LD) pruning (detailed in the methods section), we filtered SNPs with MAF > 0.05 and identified K=8 as the optimal value based on cross-validation.
+Extended Data Fig. 1: Population genetic ancestry inference of the 53 APR samples using ADMIXTURE.
+Assuming ancestry components K ranging from 3 to 9, each plot shows independent ancestry fraction with its own color-coding representation labeled with short vertical lines. Samples included are from South Asian, European, East Asian, Native American, Oceanian, African and Arab ethnicities.
+
+• The PCA shown in Figure 1e is still problematic showing East Asian clustered next to Americans. Notably, the UAE samples shown in red are shifted to the right in comparison to the reference Arab samples shown in pink. This reflects the lack of a priori proper selection of samples to ensure wide representation.
+
+Response: We believe the PCA is correct, as East Asians will cluster near Americans. These samples are from the HGDP dataset which has been published in multiple studies (Bergstrom et al., 2020; Jakobsson et al., 2008; J. Z. Li et al., 2008) where the same patterns in the PCA emerge (Figure 2). East Asians and Americans will differentiate in subsequent principal components. We note that we have added other Middle Eastern samples to this PCA as we elaborate in the methods section. Please see below figures from the previous studies that support our statements above.
+[REDACTED]
+
+Figure 2.a: Principal component analysis (PCA) plot showing global population clustering (Figure 1E from Jakobsson et al., 2008, Nature). The PCA illustrates East Asians positioned near Americans, consistent with genetic proximity in subsequent principal components. b: PCA plot from (Figure S3B from Li et al., 2008, Science). The plot highlights similar clustering patterns, further validating the observed relationships among global populations, including East Asians and Americans.
+
+Data generation
+
+• Given that various data types have been generated for the subjects in this study, there is no attempt to check for sample swaps and contamination
+
+Response: We appreciate your comment regarding sample validation. To ensure data integrity, we performed several quality checks, including manual verification of sample identities, sex chromosome checks to confirm consistency with metadata, and fineSTRUCTURE analysis to detect potential sample swaps or duplications. Additionally, relatedness analysis was conducted using identity-by-descent, and no anomalies were found from the analysis. Furthermore, we checked the number of variants in each sample as an additional quality control measure. Significant discrepancies in the number of variants would indicate potential sample contaminations or mix ups, but no such inconsistencies were observed. These measures provided confidence in the accuracy and contamination-free nature of the dataset.
+
+Assembly metrics
+
+• There is mention of exclusion of “non-human eukaryotic pathogen genomes” in addition to 109 contigs from ChrM and exclusion of 4 contigs from other chromosomes based on sequence identify. What is the total length of what was excluded? From which individual assemblies? And what is the reason for the presence of pathogenic sequences? What is the possibility that some of what was labelled as mis-joining could be a genuine chromosomal rearrangement?
+
+Response: Thank you for your detailed comment. Non-human eukaryotic pathogen genomes were excluded to ensure the integrity of the assembly. Kraken, a widely used tool for read classification, was employed to identify and remove non-target reads that could have been naturally present in the individuals who donated samples or potentially introduced during DNA
+extraction or sequencing (Cornet & Baurain, 2022; Q. Li, Yan, Lam, & Luo, 2022; Sherman et al., 2019). For example, in a previous African pangenome study based on short-read data, 29 individuals were found to have contigs related to malaria infections and 1 with human beta herpesvirus, which were excluded after screening (Sherman et al., 2019 Nature Genetics) On average, 7 HiFi reads (range: 0–15; representing ~0.00001% of total reads) with an average size of 61.8kb and 187 ONT reads (range: 11–1797; representing ~0.0003% of total reads) with an average size of 2.81 Mb per sample were removed. This represents a negligible fraction of the total sequencing data, and it is common to observe such levels of non-human eukaryotic reads in ONT data.
+
+Regarding mitochondrial contigs, 109 mitochondrial contigs (average length: 16.29 kb) were removed and used to build a mitochondrial pangenome, which was separately analyzed. Out of 10,997 total contigs across 53 assemblies, 4 contigs classified as misjoins were excluded. These were identified in APR031 hap1 (104.81 Mb), APR022 hap1 (98.41 Mb), APR003 hap1 (100.09 Mb), and APR043 hap2 (5.97 Mb). The total genome length of the respective samples was 2.95 Gb, 2.91 Gb, 2.95 Gb, and 3.04 Gb. Resolving misjoins is one of the most challenging aspects of genome assembly due to the involvement of rDNA and other complex repeat elements that can lead to inaccuracies in splitting the misjoined contigs.
+
+• In Suppl Table 11, authors report #Ns, #indels and #mismatches per 100kb. Should report the total numbers instead.
+
+Response: We report these metrics normalized per 100kb, as provided by QUAST, to allow comparisons with other studies that follow the same approach (see Table 1 below, where HPRC reported the same metric with values comparable to ours). This standardized format is widely used in assembly evaluations (Mikheenko, Prjibelski, Saveliev, Antipov, & Gurevich, 2018; Muñoz-Barrera et al., 2024; Zhang, Jia, & Wei, 2019) and provides a consistent framework for assessing assembly quality.
+
+• In Suppl Table 11, the numbers of misassembled contigs relative to CHM13 as a fraction from number of contigs is on average 56%. This is too high.
+
+Response: Thank you for your comment. We acknowledge the concerns regarding the reported fraction of misassembled contigs. To address this, we analyzed the Human Pangenome Reference Consortium (HPRC) assemblies using QUAST. While HPRC did not report the number of misassembled contigs in their manuscript, our analysis of their publicly available assemblies (https://s3-us-west-2.amazonaws.com/human-pangenomics/index.html?prefix=working/HPRC/HG00438/assemblies/), showed that the fraction of misassembled contigs in their dataset is comparable to the APR. Specifically, HPRC samples had an average of 147 misassembled contigs out of 371 total contigs, similar to the trends observed in the APR dataset (average 56 misassembled contigs out of 103 total contigs). For additional context, Table 1 presents the QUAST report for 10 HPRC samples, highlighting the similarities in assembly metrics.
+
+Table 1: QUAST report of 10 HPRC samples
+<table>
+  <tr>
+    <th>Sample</th>
+    <th>Total contigs</th>
+    <th># misassembled contigs</th>
+    <th>N50 (bp)</th>
+    <th># indels per 100 kbp</th>
+    <th># mismatches per 100kb</th>
+    <th>Unaligned length (bp)</th>
+  </tr>
+  <tr><td>HG00438_Maternal</td><td>259</td><td>129</td><td>54936949</td><td>25.7</td><td>149.49</td><td>24947097</td></tr>
+  <tr><td>HG00438_Paternal</td><td>278</td><td>115</td><td>48061544</td><td>26.37</td><td>149.32</td><td>29891555</td></tr>
+  <tr><td>HG00735_Maternal</td><td>251</td><td>111</td><td>56474489</td><td>25.13</td><td>141.18</td><td>27588990</td></tr>
+  <tr><td>HG00735_Paternal</td><td>321</td><td>132</td><td>53422923</td><td>25.16</td><td>141.82</td><td>28779735</td></tr>
+  <tr><td>HG00741_Maternal</td><td>307</td><td>128</td><td>41001116</td><td>26</td><td>137.58</td><td>31296475</td></tr>
+  <tr><td>HG00741_Paternal</td><td>311</td><td>118</td><td>51040418</td><td>25.16</td><td>141.68</td><td>34453559</td></tr>
+  <tr><td>HG01123_Maternal</td><td>374</td><td>158</td><td>54362305</td><td>24.56</td><td>136.8</td><td>25535378</td></tr>
+  <tr><td>HG01123_Paternal</td><td>571</td><td>176</td><td>44719827</td><td>24.88</td><td>136.34</td><td>30607908</td></tr>
+  <tr><td>HG01175_Maternal</td><td>322</td><td>144</td><td>36535860</td><td>25.42</td><td>138.32</td><td>30713555</td></tr>
+  <tr><td>HG01175_Paternal</td><td>395</td><td>156</td><td>34803293</td><td>25.94</td><td>149.13</td><td>31661659</td></tr>
+  <tr><td>HG01358_Maternal</td><td>335</td><td>166</td><td>48753445</td><td>24.71</td><td>138.74</td><td>27266293</td></tr>
+  <tr><td>HG01358_Paternal</td><td>442</td><td>169</td><td>52599478</td><td>25.58</td><td>142.62</td><td>28700297</td></tr>
+  <tr><td>HG01361_Maternal</td><td>308</td><td>134</td><td>45122217</td><td>25.09</td><td>133.34</td><td>30143609</td></tr>
+  <tr><td>HG01361_Paternal</td><td>378</td><td>117</td><td>47178056</td><td>24.72</td><td>138.46</td><td>26966319</td></tr>
+  <tr><td>HG01891_Maternal</td><td>352</td><td>143</td><td>81112077</td><td>30.69</td><td>178.82</td><td>26657225</td></tr>
+  <tr><td>HG01891_Paternal</td><td>499</td><td>151</td><td>57096483</td><td>31.46</td><td>170.8</td><td>36542338</td></tr>
+  <tr><td>HG01952_Maternal</td><td>326</td><td>126</td><td>54639450</td><td>25.44</td><td>142.05</td><td>28299280</td></tr>
+  <tr><td>HG01952_Paternal</td><td>486</td><td>153</td><td>44250376</td><td>25.77</td><td>142.58</td><td>31846874</td></tr>
+  <tr><td>HG02148_Maternal</td><td>425</td><td>217</td><td>39938933</td><td>24.83</td><td>136.8</td><td>22557821</td></tr>
+  <tr><td>HG02148_Paternal</td><td>493</td><td>192</td><td>41874143</td><td>25.18</td><td>133.84</td><td>25156041</td></tr>
+</table>
+
+Moreover, it is important to consider that differences flagged as misassemblies by QUAST may represent true structural variations, such as rearrangements, large indels, gene duplication events, and variations in repeat copy numbers (Chawla, Kumar, & Shankar, 2016). These variations are an inherent feature of de novo genome assemblies, especially when dealing with complex and repetitive genomic regions.
+
+• Authors make the statements “All samples surpassed the 40 Mb N50 of the HPRC reference genome”. However the N50 they report include the misassembled conigs shown in Supl Table 11.
+
+Response: Table 1 demonstrates that the HPRC also reported N50 values within the 40 Mb range, which includes misassembled contigs. In our study, we have generated over 2,011 GB of sequencing data with reads exceeding 100 kb using Oxford Nanopore’s ultra-long read protocol. This significant amount of ultra-long read data has been pivotal in achieving our large N50 values. Moreover, the Japanese T2T group, as presented in the HPRC meeting, reported similar N50 values following a comparable methodology. We have ensured consistency in N50 computation by adhering to similar methods used by the HPRC, underscoring the robustness of our results.
+• QV for base pair quality was it calculated using the long read data used for the assemblies? If yes, what bias this could constitute?
+
+Response: Thank you for your comment. Multiple studies have demonstrated the use of Yak k-mer analysis on long-read data to estimate the base accuracy (QV) of contigs. This approach is becoming the new standard for assembly quality assessment, as long-read technologies provide high accuracy for assembly without requiring a hybrid approach. For instance, Benedict Paten’s lab (Kolmogorov et al., 2023) recently applied Yak k-mer on long-read data for QV estimation, and no significant bias was observed in their analyses. Similarly, we utilized multiple long-read (PacBio 33x and ONT 54x ULK coverage) and Hi-C (65x coverage) data for building assembly and computed QV using PacBio reads. The Q scores achieved (33.1) using PCR free high fidelity PacBio long reads are comparable to or as reliable as those obtained using short reads, hence anticipated bias will be extremely minimum.
+
+• Unclear how the coverage was computed exactly.
+
+Response: Similar to HPRC and CPC, we computed coverage based on T2T-CHM13 and GRCh38 (excluding alt).
+
+‘Haploid assemblies with an X chromosome had an average total length of 3.02 Gb, representing 99.1% of the T2T-CHM13 (3.06 Gb). Conversely, haploid assemblies with a Y chromosome averaged a total length of 2.98 Gb, highlighting the inherent size difference between the sex chromosomes.’
+
+• The reported average size of 50/77 Mb of contigs that did not align to CHM13/GRC38 does that include the misassembled contigs mentioned previously? Same question for the results reported regarding phasing quality etc
+
+Response: The unaligned length represents the total length of all segments from the assembled sequences that could not be aligned to the reference genomes (T2T-CHM13/GRC38). This unaligned fraction likely includes individual-specific variations, insertion of new sequences, structural variations, gene duplication events or novel genomic regions that are absent or different in the reference genome.
+
+Similar unmapped sequence data has been reported by the HPRC (refer to Table 1), further supporting the presence of novel regions or structural variations. Additionally, the unaligned sequences may encompass complex genomic rearrangements or repeats (i.e. rDNA), which remain challenging to map against reference genomes. This underscores the importance of comprehensive assemblies in capturing population-specific and individual-level genomic diversity.
+
+Gene duplication analysis
+
+• This section shows several inconsistencies. Results were generated showing multiple copies of genes across the assemblies (Suppl Table 14a), but authors only discuss the duplications (Fig
+2g). There is large variance in the number of genes with additional copies relative to GRCH38, ranging from 24 to 206 genes. The number of additional copies itself ranges between 1 and 46. Some samples have unrealistically high number of impacted genes such as APR_023_hap2 which has 206 genes with additional multiple copies relative to GRCH38, which is 4 times higher than the average, while hap 1 of the same sample has only 43 genes! Ultimately by looking at the trio, it is odd to see that the child has much lower number of genes with multiple copies in comparison to the father and the mother.
+
+Response: We appreciate your detailed comment regarding gene duplications and the observed variance. The APR cohort shows a gene duplication range of 24–206, with an outlier (a haplotype of APR_023 having 206 duplications) and only 4 haplotypes exceeding 100 duplications. The rest fall within the HPRC range (16–79). A similar pattern is observed in the HPRC dataset, where HG01358 has 79 duplicated genes on hap 2 and only 18 on hap 1. Such outliers are expected in these analyses due to natural biological variation, as outlier haplotypes may reflect individual-specific duplications. This pattern is not unique to our dataset and is part of the natural variation reported in genomic studies. For example, 60-80% of healthy individuals have a copy number variation (CNV) of at least 100 kb, while 5-10% have a CNV of at least 500 kb(Girirajan, Campbell, & Eichler, 2011).
+
+In the HPRC dataset, additional gene copies range from 1 to 37 (Liao et al., 2023)( Supplementary Table 9), highlighting comparable observations across datasets. These findings underscore the inherent variability in gene duplication patterns across individuals and cohorts.
+
+For the trio, the observed discrepancy in duplications may be due to various biological factors, including the variable inheritance of duplicated genes. Additionally, the high rate of loss of heterozygosity (LOH) in consanguineous populations, such as Arabs, may further contribute to these observed patterns. Such phenomena are indicative of the unique genetic dynamics within populations with high consanguinity, emphasizing the importance of considering population-specific genomic contexts in these analyses.
+
+• In view of the above, I wonder how significant are the results of the comparison with HPRC.
+
+Response: While the sample size for all three pangenomes remains relatively small, which may limit the ability to fully characterize the true distribution of duplicated genes per individual, our findings provide valuable insights. The number of unique duplicated genes observed in the APR (883 genes) and HPRC (801 genes) datasets is comparable, suggesting consistent trends in gene duplication between these populations. This consistency supports the validity of our comparison and highlights the utility of the APR dataset as a complementary resource to existing pangenome references. Further studies with larger cohorts will be necessary to refine these observations and fully capture the spectrum of gene duplication across diverse populations.
+
+Variant analysis
+
+• No mention of what reference the variant calls were made against and if any filtering was applied including the low-quality regions of the assemblies.
+Response: Thank you for your observation. Variants from pangenome were called using T2T-CHM13 as the reference, which was also utilized as the backbone for graph construction in our study. Although this was previously mentioned in our manuscript, we explicitly clarified this in the methods and results section of the revised version.
+
+We employed UniAligner, to assess mapping within highly complex genomic regions, including centromeric regions. We observed a mapping rate of 26.97% for APR samples, surpassing the 20.42% observed in the HPRC dataset. This highlights the better representation of complex genomic regions in our data, as detailed in Supplementary Fig. 11. These steps ensured robust variant calling, including low-quality regions. We have also included statistics on the coverage of subtelomeric and pericentric regions, which are enriched with rDNA (Fig. 1d).
+
+• The authors report a high proportion of novel variants, including for SNVs and Indels where it is 15-19% per genome. This is odd because for this type of variation there is genomic completeness in the references used, and samples from Middle East have already been extensively sampled in the literature. The authors do not attempt to check in GNOMAD and other Middle Eastern cohorts such as Mineral et al. 2021 and Scott et al. 2016.
+
+Response: Thank you for your comment. We have applied Inspector polishing algorithm to eliminate systematic errors, ensuring the reliability of our data. The reported number of small variants, including SNVs and indels, is consistent with findings from HPRC and CPC. For APR specific variants, we have conducted extensive filtering using dbSNP, gnomAD, the 1000 Genomes Project, and Greater Middle East (GME) - a comprehensive database that curates variants from numerous genome sequencing studies focused on Middle Eastern samples. The Arab representation is very limited in large genomic resources, including HPRC, 1000 Genomes Project and gnomAD (mostly exome). Our study will serve as a significant resource for understanding the genomic landscape of the MENA region and the broader global genomic community. We believe the reviewer is referencing Mineta et al, 2021 but this a genotyping array-based study that would not be useful to calculate novelty of variants.
+
+• For the SV calls, there is no attempt to QC the calls which are more prone to mis-assembly errors
+
+Response: Thank you for your comment. The number of structural variant (SV) calls per sample in our study is comparable to those reported in the HPRC dataset, indicating consistency in SV detection. We followed established methodologies aligned with HPRC QC standards, ensuring accurate and reliable SV calls. Furthermore, we built assemblies with >100X coverage (including 2011 GB of data with minimum >100kb stretch DNA with 12.53X coverage per sample) using long-read sequencing data, which are known for their high sensitivity in detecting structural variants. As the field of pangenome continues to evolve, these approaches contribute to standardizing SV detection processes, minimizing errors and maximizing data accuracy.
+
+Novel euchromatic sequences
+
+• Authors report their assemblies contain 112 Mbs of novel sequence relative to CHM13, other
+pan genomes, and 1.4 kb of Mt DNA. This amounts to ~ 4% of the human genome which seems excessive and may not as striking. How much does that compare to the inverse situation i.e the amount of novel sequences in HPRC samples relative to the samples in this study?
+
+Response: Our novel sequence estimation is based on SV insertions absent from GRCh38, T2T-CHM13, HPRC, CPC, and DGV. While HPRC and CPC reported novel sequences relative to the linear references GRCh38 and T2T-CHM13, our analysis incorporates all references, including both linear and graph pangenomes, to exclude previously reported sequences. Additionally, we filtered out insertions for variants present in the DGV database. The APR pangenome growth curve (Fig. 3h) and the per-sample contribution to novel sequences in APR (2.6 MB) is comparable to those observed in HPRC (1.9 MB) and CPC (2.8 MB).
+
+Complex structural variation
+
+• Strange the criteria used here for naming this category of SVs. Authors should just refer to SVs larger than certain length and frequent above a certain threshold.
+
+Response: The term "Complex Structural Variation Site and Region" was used to describe regions in the graph genome containing multiple overlapping structural variations (SVs) from multiple haplotypes, including rearrangements, deletions, and insertions. This is illustrated in Figure 4, which highlights loci with multiallelic SVs in both HPRC and APR datasets. This aligns with terminology used in prior studies (e.g., "complex multiallelic SV loci"), as employed by HPRC, to denote such regions with high allelic diversity.
+
+• Again novelty claimed without assessing relevant datasets beyond HPRC+CPC.
+
+Response: The structural variants we reported underwent rigorous filtering, including comparisons against SVs in the reference genomes, HPRC, and CPC. Additionally, we excluded SVs present in the Database of Genomic Variants (DGV), which hosts one of the largest global datasets of structural variants.
+
+Mitochondrial pangeome
+
+• No mention which reference variants were called against
+
+Response: Thanks for your comment. We used T2T-CHM13 mitochondrial genome sequence for mapping and variant calling. This information has now been explicitly stated in the revised manuscript.
+
+• The Mt assemblies are based on PacBio only. No mention of what particular QC was performed for these genomes nor what mis-assemblies were found
+
+Response: For Mt pangenome construction, we used PCR free high fidelity PacBio long reads with an average Q score of 33, a minimum length of 15kb, and at least 90% sequence identity. This stringent quality control allowed us to detect a set of high quality heteroplasmy variants from Arab populations.
+As the Mt pangenome was directly built from high-fidelity heteroplasmy reads and not from assembly, concerns about potential misassemblies are not applicable in this context. Given that Minigraph-Cactus cannot handle construction of de novo cyclic genomes, we utilized PGGB algorithm to construct the Mt pangenome graph.
+
+Performance gain from pangenome-aided
+
+• Higher calling rate using Pan genomes has been shown before so that is not specific to the APR graph.
+
+Response: Thank you for your comment. We agree that the higher calling rate using pangenome has been demonstrated previously, including the HPRC study. However, it is important to note that the HPRC dataset does not include Arab samples. In our analysis, we observed a higher genotype and structural variation calling rate in APR compared to HPRC (Fig. 5f and g). While this represents an overall improvement, it holds greater significance because some of these variants are located within APR-specific sequences that may not be adequately represented or mapped in HPRC. We illustrate this using trio data we generated: the APR pangenome variant calls exhibited lower errors in Mendelian concordance (with error rates of 0.0071 for SNPs and 0.0172 for indels), significantly lower than those observed by HPRC (with error rates of 0.0504 for SNPs and 0.0749 for indels).
+
+• The mapping rate is between the APR and HPRC is almost identical. Unclear what is the added value here
+
+Response: The mapping rate of APR is significantly higher compared to T2T-CHM13, confirming its advantage over the current linear references. While the mapping rate is comparable to HPRC, the sequence and haplotype structure in APR is specific to the Arab population, offering a distinct advantage in detecting structural variations and population-specific variants. See the Mendelian concordance analysis we mentioned above. Our analysis using short read mapping demonstrates that APR achieves a superior structural variation detection rate compared to HPRC.
+
+References:
+
+Almarri, M. A., Haber, M., Lootah, R. A., Hallast, P., Al Turki, S., Martin, H. C., . . . Tyler-Smith, C. (2021). The genomic history of the Middle East. Cell, 184(18), 4612-4625 e4614. doi:10.1016/j.cell.2021.07.013
+Arauna, L. R., Mendoza-Revilla, J., Mas-Sandoval, A., Izabel, H., Bekada, A., Benhamamouch, S., . . . Comas, D. (2017). Recent Historical Migrations Have Shaped the Gene Pool of Arabs and Berbers in North Africa. Mol Biol Evol, 34(2), 318-329. doi:10.1093/molbev/msw218
+Bergstrom, A., McCarthy, S. A., Hui, R., Almarri, M. A., Ayub, Q., Danecek, P., . . . Tyler-Smith, C. (2020). Insights into human genetic variation and population history from 929 diverse genomes. Science, 367(6484). doi:10.1126/science.aay5012
+
+Chawla, V., Kumar, R., & Shankar, R. (2016). Identifying wrong assemblies in de novo short read primary sequence assembly contigs. J Biosci, 41(3), 455-474. doi:10.1007/s12038-016-9630-0
+
+Cornet, L., & Baurain, D. (2022). Contamination detection in genomic data: more is not enough. Genome Biol, 23(1), 60. doi:10.1186/s13059-022-02619-9
+
+Girirajan, S., Campbell, C. D., & Eichler, E. E. (2011). Human copy number variation and complex genetic disease. Annu Rev Genet, 45, 203-226. doi:10.1146/annurev-genet-102209-163544
+
+Jakobsson, M., Scholz, S. W., Scheet, P., Gibbs, J. R., VanLiere, J. M., Fung, H. C., . . . Singleton, A. B. (2008). Genotype, haplotype and copy-number variation in worldwide human populations. Nature, 451(7181), 998-1003. doi:10.1038/nature06742
+
+Kolmogorov, M., Billingsley, K. J., Mastoras, M., Meredith, M., Monlong, J., Lorig-Roach, R., . . . Paten, B. (2023). Scalable Nanopore sequencing of human genomes provides a comprehensive view of haplotype-resolved variation and methylation. bioRxiv. doi:10.1101/2023.01.12.523790
+
+Li, J. Z., Absher, D. M., Tang, H., Southwick, A. M., Casto, A. M., Ramachandran, S., . . . Myers, R. M. (2008). Worldwide human relationships inferred from genome-wide patterns of variation. Science, 319(5866), 1100-1104. doi:10.1126/science.1153717
+
+Li, Q., Yan, B., Lam, T. W., & Luo, R. (2022). Assembly-free discovery of human novel sequences using long reads. DNA Res, 29(6). doi:10.1093/dnarese/dsac039
+
+Liao, W. W., Asri, M., Ebler, J., Doerr, D., Haukness, M., Hickey, G., . . . Paten, B. (2023). A draft human pangenome reference. Nature, 617(7960), 312-324. doi:10.1038/s41586-023-05896-x
+
+Mikheenko, A., Prjibelski, A., Saveliev, V., Antipov, D., & Gurevich, A. (2018). Versatile genome assembly evaluation with QUAST-LG. Bioinformatics, 34(13), i142-i150. doi:10.1093/bioinformatics/bty266
+
+Mineta, K., Goto, K., Gojobori, T., & Alkuraya, F. S. (2021). Population structure of indigenous inhabitants of Arabia. PLoS Genetics, 17(1), e1009210. https://doi.org/10.1371/journal.pgen.1009210
+
+Muñoz-Barrera, A., Rubio-Rodriguez, L. A., Jáspez, D., Corrales, A., Marcelino-Rodriguez, I., Lorenzo-Salazar, J. M., . . . Flores, C. (2024). Benchmarking of bioinformatics tools for the hybrid &lt;em&gt;de novo&lt;/em&gt; assembly of human whole-genome sequencing data. bioRxiv, 2024.2005.2028.595812. doi:10.1101/2024.05.28.595812
+
+Sherman, R. M., Forman, J., Antonescu, V., Puiu, D., Daya, M., Rafaels, N., . . . Salzberg, S. L. (2019). Assembly of a pan-genome from deep sequencing of 910 humans of African descent. Nat Genet, 51(1), 30-35. doi:10.1038/s41588-018-0273-y
+
+Zhang, W., Jia, B., & Wei, C. (2019). PaSS: a sequencing simulator for PacBio sequencing. BMC Bioinformatics, 20(1), 352. doi:10.1186/s12859-019-2901-7

09b079d80b3efdeef69e111bf52a9dde06da16423667f03702d57d11bbbeec9b/preprint/preprint.md

@@ -0,0 +1,525 @@
+Meta Selective C–H Borylation Directed by Secondary Silicon Oxygen Interaction
+
+Buddhadeb Chattopadhyay (✉ buddhadeb.c@cbmr.res.in )
+Center of Biomedical Research (CBMR) https://orcid.org/0000-0001-8473-2695
+Saikat Guria
+Center of Biomedical Research (CBMR)
+Mirja Md Hassan
+Centre of Biomedical Research (CBMR)
+Sayan Dey
+Center of Biomedical Research (CBMR)
+
+Article
+
+Keywords:
+
+Posted Date: August 17th, 2022
+
+DOI: https://doi.org/10.21203/rs.3.rs-1837437/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: There is NO Competing Interest. We declare that the authors have no competing interests except that We have filled an Indian Patent (Patent Application No: 202211036590) based on this work (including the ligand and catalyst).
+Meta Selective C–H Borylation Directed by Secondary Silicon Oxygen Interaction
+
+Saikat Guria†, Mirja Md Mahamudul Hassan†, Sayan Dey†, Buddhadeb Chattopadhyay†*
+
+†Department of Biological & Synthetic Chemistry, Centre of Biomedical Research, SGPGIMS Campus,
+Raebareli Road, Lucknow 226014, Uttar Pradesh, India
+*Correspondence to: buddhadeb.c@cbmr.res.in
+
+Summary Paragraph: Remote meta selective C–H functionalization\( ^{1,2,3} \) of aromatic compounds remains a challenging problem in chemical synthesis. Here, we report an iridium catalyst bearing a bidentate pyridine-pyridone (PY-PYRI) ligand framework that efficiently catalyzes this meta selective borylation reaction. We demonstrate that the developed concept can be employed to introduce a boron functionality at the remote meta position of phenols, phenol containing bioactive and drug molecules, which was an extraordinary challenge. Moreover, we have demonstrated that the method can also be applied for the remote C6 borylation of indole derivatives including tryptophan that was the key synthetic precursor for the total synthesis of Verruculogen and Fumitremorgin A alkaloids. The origin of the remote meta selectivity was described as a secondary silicon oxygen interaction\( ^{4} \) that was never used in C–H functionalization chemistry.
+Transition metal-catalyzed C–H bond activation and functionalization\(^{5,6,7,8,9,10,11}\) of aromatic compounds has been branded as one of the most significant chemical transformations. This has a profound impact in modern synthetic organic chemistry, ranging from laboratory methods to industrial deployment.\(^{12,13}\) However, the key underlying principles for the success of the metal catalysis lies on the two important factors, such as: (i) design and synthesis of new generation ligand framework that can produce highly reactive catalyst system\(^{14,15}\) and (ii) substrates’ structure modifications\(^{16}\) by which site selectivity could be controlled by the steric crowding\(^{17,18,19,20,21}\) or various weak interactions\(^{22,23,24,25}\) of the aromatic compounds among several similar type of C–H bonds via the ligand–substrate pre-organization\(^{26,27}\). In recent times, many elegant approaches\(^{28}\) have been developed for the functionalization of proximal\(^{15,29,30,31}\) and remote C–H bonds\(^{1,3,32,33,34,35,36,37,38}\) of arenes by the design of either new ligand frameworks with an extended architectures featuring a weak coordinating functional groups\(^{39}\) or templates\(^{40}\) as well as transient mediators\(^{41}\) or transient directing groups\(^{42}\) attached with the substrates. While ligand having an extended architecture or template approaches are extremely important to functionalize the remotely located C–H bonds of arenes, but requirement of multi-step preparation of the linkers of the ligands and templates of the aromatic substrates significantly limit the wide application of the methods.\(^{43}\)
+
+Among numerous aromatic substrates, phenols are the most widespread aromatic compounds that acquired household products including several bioactive to important drug molecules.\(^{44}\) Moreover, it is well-documented that 10% of the top 200 selling pharmaceuticals contain a phenol and several others employ phenols as synthetic intermediates.\(^{45}\) Furthermore, phenols are also key components of the biopolymers melanin, lignin, resins and polyphenylene oxides.\(^{44,45,46}\) In industry, phenol is routinely used as a raw material to make numerous important components by means of its diversification via the synthetic manipulation.\(^{44,45}\) Thus, direct functionalization of phenols would be a significant development for the rapid access of numerous important products.\(^{46}\) In this context, traditional electrophilic substitution is an alternative methods that affords variously substituted phenols (Fig. 1A, a).\(^{47}\) Employing this method, one can easily access ortho and para substituted phenol derivatives, although often remain a chance to have mixture of isomers. However, functionalization of the remote meta C–H bonds of phenols is extremely difficult because of the extreme inertness of the meta C–H bonds. Several pioneering approaches have been developed by Yu and others either using template method\(^{48,49,50}\) or transient directing group by Larrosa\(^{2}\)(Fig 1A, b, c). But, achieving the meta functionalized products using these methods, it is essential to have specialized substrates that limits the application of the methods.
+
+Having tremendous importance of catalytic C-H borylation\(^{51,52,53,54,55,56}\) in organic synthesis, we report here a concept for the meta selective C–H borylation of phenols via an unprecedented Si–O interaction that has never been utilized in C-H functionalization chemistry. Literature reports revealed\(^{4}\) that the most common structural motif for this O–Si interaction can be found in the amide skeleton, where a filled p-orbital of oxygen atom can interact with the vacant d-orbital of the tetracoordinated silicon atom consisting of at least one electronegative atom. The role of this electronegative atom is to make silicon atom more electropositive by developing a partial positive charge on the silicon atom (Fig. 1B, a). Notably, while various reports of O–Si weak interactions have been shown in a number of intermediates (Fig 1B, b-e),\(^{4,57,58}\) there was no report of utilization of these interactions in the catalysis research. Inspired from this background reports,\(^{4,57,58}\) we have proposed a hypothesis where phenol is protected with an easily removable electropositive silane group and silane protected phenol meet all the necessary criteria (having electropositive silane with attached electronegative oxygen atom) for the O–Si weak interaction with 2-pyridone moiety having amide skeleton (4). (Fig. 1C). The reaction design (ligand and catalyst design) is shown in Fig. 1D, a, b. The designed ligand (PY-PYRI) consists of two parts, one part is the simple pyridine unit (PY) and another part is a 2-pyridone (PYRI) unit\(^{59}\) which was redesigned by the skeletal modification of bipyridine core structure. The origin of the remote meta selectivity is presented in Fig. 1D, b. We hypothesized that the designed ligand (PY-PYRI) would control the remote meta selectivity owing to the following two considerations. Firstly, in presence of [Ir(cod)(OMe)]\(_2\), the ligand (PY-PYRI) will generate a complex (Int-A) without tautomerization of the 2-pyridone unit. Secondly, the p-orbital of the oxygen atom of the 2-pyridone unit will interact with the vacant d-orbital of the tetracoordinated electropositive silicon atom of the substrate (Fig. 1D, b).
+A. Meta functionalization of phenol:
+
+meta functionalization not well explored
+
+ortho & para functionalization
+Highly accessible and well explored
+
+Electrophilic approach
+
+phenol is ortho and para directing
+overcomes
+
+Difficulties:
+• Pre-installation of template
+• Deprotection of the template
+• Multi-step method
+
+b Template approach
+covalent binding
+
+c Transient directing group approach
+harsh conditions
+transient DG
+Required harsh conditions for installation of DG
+
+B. Conceptual background
+Dative O–Si interaction in amide skeleton:
+amide skeleton
+tetracoordinate silicon atom
+electronegative atom
+Common structural motif for O–Si interaction
+X = halogen atom
+
+Known Literature reports
+Organometallics, 2011
+J. Mol. Struct., 2021
+Organometallics, 2013
+
+C. Hypothesis:
+tautomerism in 2-pyrindone
+
+Electronegativity difference:
+Si–O bond: 1.7
+weak interaction
+
+Important considerations:
+a. Si have vacant d-orbital
+b. pyridone O have filled p-orbital
+
+polarised O–Si bond generates partial opposite charges on O and Si atom
+
+Structural modification of bpy based ligand
+
+Int-A
+partial (+) charge at O atom of pyridone
+
+PY-PYRI
+catalyst design
+
+D. Reaction design
+Proposed approach for meta borylation
+easily removable silane group
+Ir/Bpin2
+Pyridone ligand
+Weak interaction mediated meta borylation
+
+Fig. 1. Conceptual outline and proposed concept for the meta borylation of phenol. A. Meta functionalization of phenol, a, electrophilic approach. b, template approach. c, transient directing group approach. B. Conceptual Background. a, common structural skeleton for O–Si interaction. b-e, known literature reports of O–Si interactions. C. Hypothesis, D. Reaction design. a, ligand design by structural modification, b, proposed approach for meta borylation.
+
+The designed ligand (PY-PYRI) was prepared by the known synthetic methods (SI, for details), which was employed for the reaction optimization of steering silane group attached with phenol (Fig. 2A). We started our initial studies with the substrate (1a-I) featuring SiMe3 as the steering group under iridium-catalyzed conditions using the designed ligand (L1: PY-PYRI) at 40 °C temperature, which afforded good meta/para selectivity (m/p = 87/13) with 68% conversion. Evaluating other silane based steering groups under the same reaction conditions, we observed that Si(Pr)3 produced best meta selectivity (entry 2a: m/p = 94/6) with excellent conversion (95%). With this optimized Si(Pr)3 as steering group, different other ligands have also been tested to observe the effect on the selectivity (Fig. 2B). It was found that replacing the tert-butyl group with methyl group, ligand (L2) gave 91% meta selectivity with less conversions (68%). Similar selectivity was obtained when the reaction was performed with the ligand (L3) without any substituents on the pyridine unit of the ligand. Notably, the meta selectivity and conversion was found to be less using the bipyridine ligand (L4) compared to the ligands (L1-L3), which indicated the important role of the
+2-pyridone unit of the designed (PY-PYRI) ligand. Employment of other ligands (L5–L8) resulted in no reaction except the ligands (L9 & L10), which gave moderate meta selectivity.
+
+A. Optimization of steering groupa
+
+<table>
+  <tr>
+    <th>Reaction Development</th>
+    <th>B. Deleterious result with other ligandsa</th>
+  </tr>
+  <tr>
+    <td>
+      <img src="page_146_130_670_312.png" alt="Optimization of steering group">
+    </td>
+    <td>
+      <img src="page_820_130_670_312.png" alt="Deleterious result with other ligands">
+    </td>
+  </tr>
+</table>
+
+<table>
+  <tr>
+    <th>Ligand design</th>
+    <th>structural modification</th>
+  </tr>
+  <tr>
+    <td>L1</td>
+    <td>dtbpy</td>
+  </tr>
+</table>
+
+C. Origin of selectivity:
+
+![Diagram showing electron-rich pyridone carbonyl oxygen center and polar bond](page_146_563_670_246.png)
+
+electronically riched pyridone carbonyl oxygen center (filled p-orbital)
+polar bond
+electronegativity difference C–Si : 0.7, O–Si : 1.7
+vacant d-orbital
+
+D. Proof of concept:a
+
+<table>
+  <tr>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+  </tr>
+  <tr>
+    <td>CH2Si(Pr)3</td>
+    <td>2a-VII, 68% m/p = 77/23</td>
+    <td>XSi(R)3</td>
+    <td>2a, 95% m/p = 94/6</td>
+  </tr>
+  <tr>
+    <td>CH2Si(Me)3</td>
+    <td>2a-VIII, 59% m/p = 77/23</td>
+    <td>X = O, CH2</td>
+    <td>2a-I, 68% m/p = 87/13</td>
+  </tr>
+</table>
+
+Fig. 2. Ligand design and optimization of reaction conditions. A. Optimization of steering group, B. Deleterious results with other ligands. C. Origin of meta selectivity, D. Proof of concept. Reactions are on 0.2 mmol scales. aConversion was reported. In parenthesis, isolated yields are reported. See SI for details.
+
+At the outset, we proposed the tentative hypothesis that an O–Si secondary interaction between the oxygen atom of the 2-pyridone unit of the ligand and the silicon atom of the substrate’s steering group would interact each other via the filled p-orbital and empty d-orbital (Fig. 2C).4 Moreover, due to the high electronegativity difference between oxygen and silicon, the O–Si bond will be highly polarized, thereby may interact via a weak O–Si interaction. To prove this hypothesis for the origin of the meta selectivity, we performed a reaction with substrates featuring C–Si bond (–CH2SiPr3: 1a–VII and –CH2SiMe3: 1a–VIII, Fig. 2D), in which lacking of electronegative atom attached with silicon atom causes non-polarised bond, resulted in very low meta selective borylation due to loss of interaction between 2-pyridone of catalyst and substrates silicon atom. This experiment indicated an “O–Si” secondary interaction between the ligand and substrate that guides the selectivity of the borylation.
+
+Using L1 (PY-PYRI) as ligand and Si(iPr)3 as steering group, we next performed the iridium-catalyzed meta borylation of a variety of phenols that afforded excellent meta selectivity and yields of the isolated borylated products (Fig. 3). For example, we first tested 2-chlorophenol for the borylation reaction, while our designed ligand (L1) gave high meta selectivity (m/p = 92/8), traditional dtbpy ligand provided poor meta selectivity (m/p = 63/37), which clearly demonstrated the utility of the designed (L1: PY-PYRI) ligand. Other 2-substituted phenols, such as 2-bromo (1c) and 2-iodo (1d) afforded high meta selectivity that have great synthetic values owing to the two different types of handles on the phenols. Likewise, phenols bearing various alkyl chain ranging from methyl to pentyl (1e–1g) at the
+2-positions along with trifluoromethyl (1h), isopropyl (1i) and trifluoromethoxy (1l) smoothly underwent meta borylation irrespective of the nature of the substituent. Amino phenol (1k), substrate of momentous importance for the chemical and pharmaceutical industries60 is borylated with high meta selectivity (m/p = 97/3) without borylation next to the amino group, which is known to give ortho borylation under iridium-catalyzed borylation conditions via in situ generation of NHBpin group.61 Thioether (1l) that usually directs borylation at the ortho position62 also underwent borylation with good meta selectivity. We observed that phenols containing functional groups such as cyano (1m), Bpin (1n), cyclic amine (1o), cyclohexyl (1p), ketomethyl (1q) and homologous ester (1r) afforded high level of meta selectivity and tolerated well under the employed reaction conditions.
+
+![Substrates scope for substituted arenes. Reactions are in 0.5 mmol scale. Conversion was reported. 1.5 equiv. B2pin2 was used. 2.0 equiv. B2pin2 was used. See SI for details.](page_484_624_1012_624.png)
+
+Fig. 3. Substrates scope for substituted arenes. Reactions are in 0.5 mmol scale. aConversion was reported. b1.5 equiv. B2pin2 was used. c2.0 equiv. B2pin2 was used. See SI for details.
+
+Amide functionalities (1s & 1t) that are known to undergo numerous synthetic transformations63 exhibited excellent meta selective borylation. Borylation of phenols having CF3 (1h) and CN (1m) substituents at the ortho position afforded exclusively meta borylation, the same substituents at the meta position of phenols (1u & 1v) also gave meta selective borylation, which indicated the generality of the developed method. Moreover, fluoro-substituted
+arene, which typically gives borylation next to the fluorine atom under standard iridium-catalyzed conditions, in this case, 3-fluorophenol (1w) gave meta borylation as the major product. Several disubstituted phenols (1x-1ag) were also examined under the developed conditions that reacted smoothly to afford variously substituted meta borylated products in high yields. 2,2'-Biphenol, compound of paramount importance in medicinal chemistry as well as in chemical industry64 can selectively be mono- and diborylation (2ah & 2ai) by tuning the amount of boron reagent. A bulky substituent at the ortho positions (1aj) did not hamper the reaction that gave 96% meta borylated product with 90% isolated yield.
+
+![Extremely difficult & rare due to sterically crowding at the para positions](page_158_186_670_140.png)
+
+<table>
+  <tr>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+  </tr>
+  <tr>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+    <td>OSi(Pr)<sub>3</sub></td>
+  </tr>
+  <tr>
+    <td>Me</td>
+    <td>Et</td>
+    <td>pent</td>
+    <td>hex</td>
+    <td>iPr</td>
+    <td>Ome</td>
+    <td>CN</td>
+    <td>CF<sub>3</sub></td>
+    <td>m</td>
+    <td>Bpin</td>
+    <td>Bpin</td>
+    <td>Bpin</td>
+  </tr>
+  <tr>
+    <td>4a, 74%<br>(m/others = 91/9)</td>
+    <td>4b, 77%<br>(m = 100)</td>
+    <td>4c, 83%<br>(m = 100)</td>
+    <td>4d, 84%<br>(m = 100)</td>
+    <td>4e, 63%<br>(m = 100)</td>
+    <td>4f, 86%<br>(m = 100)</td>
+    <td>4k, 89%<br>(m = 100)</td>
+    <td>4l, 86%<br>(m = 100)</td>
+    <td>4m, 85%<br>(m = 100)</td>
+    <td>4n, 72%<br>(m = 100)</td>
+    <td>4o, 82%<sup>a</sup><br>(m = 100)</td>
+    <td>4p, 81%<br>(m = 100)</td>
+  </tr>
+  <tr>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+  <tr>
+    <td></td>
+    <td>OCF<sub>3</sub></td>
+    <td>SCF<sub>3</sub></td>
+    <td>SCH<sub>3</sub></td>
+    <td>OEt</td>
+    <td>OEt</td>
+    <td>OEt</td>
+    <td>OEt</td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+  <tr>
+    <td>4g, 90%<br>(m = 100)</td>
+    <td>4h, 92%<br>(m = 100)</td>
+    <td>4i, 99%<sup>a</sup><br>(m = 100)</td>
+    <td>4j, 87%<br>(m = 100)</td>
+    <td>4k, 89%<br>(m = 100)</td>
+    <td>4l, 86%<br>(m = 100)</td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+  <tr>
+    <td></td>
+    <td>Cl</td>
+    <td>Br</td>
+    <td>OMe</td>
+    <td>OMe</td>
+    <td>OMe</td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+  <tr>
+    <td>4m, 85%<br>(m = 100)</td>
+    <td>4n, 72%<br>(m = 100)</td>
+    <td>4o, 82%<sup>a</sup><br>(m = 100)</td>
+    <td>4p, 81%<br>(m = 100)</td>
+    <td>4q, 72%<br>(m = 100)</td>
+    <td>4r, 43%<br>(m = 100)</td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+</table>
+
+Borylation of tryptophan
+
+<table>
+  <tr>
+    <td colspan="2">5.0 mol% [Ir(cod)OMe]<sub>2</sub></td>
+    <td colspan="2">10.0 mol% 1,10-phen</td>
+    <td colspan="2">0.25 equiv B<pin>Bpin<sub>2</sub></p></td>
+    <td colspan="2">4.0 equiv B<pin>Bpin<sub>2</sub></p></td>
+    <td colspan="2">heexane, 80 °C, 24 h</td>
+    <td colspan="2"></td>
+  </tr>
+  <tr>
+    <td colspan="2"><img src="page_1092_186_134_140.png" alt="Borylation of tryptophan"></td>
+    <td colspan="2"><img src="page_1092_326_134_140.png" alt="Previous Work"></td>
+    <td colspan="2"><img src="page_1092_466_134_140.png" alt="Bioactive alkaloids"></td>
+    <td colspan="2"><img src="page_1092_606_134_140.png" alt="Keystep: Ligand cotrolled C-H borylation"></td>
+    <td colspan="2"></td>
+    <td colspan="2"></td>
+  </tr>
+  <tr>
+    <td colspan="12">77% (C6:C5 = 89:11)<br>previous selectivity</td>
+  </tr>
+</table>
+
+Borylation of tryptophan
+
+<table>
+  <tr>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+    <th></th>
+  </tr>
+  <tr>
+    <td>pinB<sup>6</sup></td>
+    <td>pinB<sup>6</sup></td>
+    <td>pinB<sup>6</sup></td>
+    <td>pinB<sup>6</sup></td>
+    <td>pinB<sup>6</sup></td>
+    <td>pinB<sup>6</sup></td>
+    <td>pinB<sup>6</sup></td>
+    <td>pinB<sup>6</sup></td>
+  </tr>
+  <tr>
+    <td>pHs, 94%<sup>a</sup><br>(C6:C5 = 91:9)</td>
+    <td></td>
+    <td>4t, 99%<sup>a,b</sup><br>(C6:C5 = 93:7)</td>
+    <td>4u, 84%<sup>a,b</sup><br>(C6:C5 = 93:7)</td>
+    <td>4v, 99%<sup>a,b</sup><br>(C2,C7,C3,C7 = 84/16)</td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+  <tr>
+    <td></td>
+    <td>3.0 mol% [Ir(cod)OMe]<sub>2</sub></td>
+    <td>3.0 mol% L1</td>
+    <td>3.0 equiv B<pin>Bpin<sub>2</sub></p></td>
+    <td>developed conditions</td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+  <tr>
+    <td></td>
+    <td>hexane, 80 °C, 24 h</td>
+    <td>PY-PYRI: L1</td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+    <td></td>
+  </tr>
+</table>
+
+Fig. 4. Substrates scope for the 4-substituted arenes and C6 borylation of indoles. Reactions are in 0.5 mmol scale.
+<sup>a</sup>Conversions were reported. <sup>b</sup>2.0 equiv. B<pin>Bpin<sub>2</sub></p>. See SI for details.
+
+Next, we focused on the meta borylation of those phenols bearing a substituent at the para position (Fig. 4). Because, borylation at the remote meta position in presence of a para substituent remains an extraordinary challenge due to the steric reason. Moreover, we selected those substituents at the para position that already provided exclusive meta borylation of phenols when they were located at either ortho or meta positions. The reason for this selection is mainly to observe the overall effects of the borylation by the same substituents. For the
+testification, we begun with the 4-methyl phenol (3a) that afforded 91% meta selective borylation. Increasing the chain length from small methyl group to the relatively bulkier alkyl groups such as, ethyl (3b), pentyl (3c), hexyl (3d) and isopropyl (3e), the borylation underwent smoothly with further enhancement of the meta selectivity from 91% to 100%. Para-substituted ethers and thioethers bearing electronically different substituents (3f-3j) reacted with 100% meta selectivity, which revealed that the scope of the meta borylation is very general regardless of the nature of the substituents. While 2-CN, 3-CN as well as 2-CF3 and 3-CF3 bearing phenols resulted in excellent meta borylation, the same substituents at the para position reacted to yield 100% meta borylation. Likewise, we also observed that chloro (3m) and bromo (3n) containing phenols reacted to give the meta borylation products solely irrespective of their position in the phenol. Moreover, it has been found that the phenols featuring bulky substituents at the para position (3o-3r) also gave exclusively meta borylation, although conversion was moderate in case of the cyclohexyl group.
+
+In 2015, Baran et al. reported65 the first total synthesis of Verruculogen and Fumitremorgin A enabled by ligand-controlled C–H borylation as the key step of TIPS protected tryptophan. We were curious if our designed ligand system could provide the remote C6 borylation of TIPS protected indoles and TIPS protected tryptophan. For that reason we performed borylation of TIPS-protected tryptophan (3s) (synthetic key precursor of bioactive alkaloids Verruculogen and Fumitremorgin A) which provided C6 borylation with 91% selectivity with excellent conversions. We also found that TIPS-protected other indole derivatives (3t & 3u) and TIPS-protected carbazole (3v) smoothly underwent remote borylation affording excellent selectivity and conversion. This developed method provided a simple way to borylate the 3-substituted indoles derivatives that might be beneficial for the total synthesis or the late-stage functionalization of several bioactive molecules.
+
+Late-stage functionalization66 of complex bioactive and medicinally important molecules by the site selective C–H activation is a powerful method for the development of new drug candidates.67 In this context, introducing a boron functionality into the bioactive and medicinally important molecules would further enhance the identification of new lead molecules not only for the enormous importance of the boron-bearing small molecules68 but also for the uniqueness of the boron group towards the diverse derivatization towards numerous other functional groups. Thus, we tested our developed method for several commercially available bioactive and drug molecules (Fig. 5A). For example, cannabinoid core (5a: used as a psychoactive drug), methyl salicylate derivatives (5b: an anti-inflammatory and analgesic agent), tyrosol derivatives (5c: an antioxidant), eugenol derivatives (5d: a flavouring agent), sesamol derivatives (5e: an antioxidant), naproxen derivatives (5f: a nonsteroidal anti-inflammatory drug, NSAID), deoxyarbutin derivatives (5g: used for treatment of hyperpigmentation disorders) and homosalate (5h: used as a sunscreen) were meta borylated with high yield and selectivity. The steering silane group from the borylated phenols has been removed under a very mild reaction conditions at room temperature (ethylene glycol, KF, 1h) that afforded the meta borylated phenols in high yields (Fig. 5B). Notably, the meta borylated phenols can further be transformed to a number of substituted phenols/resorcinols that are difficult to prepare by otherwise.
+
+Next, we aimed to prepare the active catalyst (10) that was proposed to form in situ between the reaction of the designed ligand (L1: PY-PYRI) and [Ir(cod)(OMe)]2 during the meta selective borylation conditions. Accordingly, we performed the reaction and isolated the catalyst [10: Ir(cod)(PY-PYRI)] in 90% yield (Fig. 5C). The catalyst structure was confirmed by X-ray crystallography and other spectroscopic data. The catalytic efficiency of this catalyst [10: Ir(cod)(PY-PYRI)] was further tested in the meta borylation reactions, which exhibited same level of meta selectivity with better product conversion (compared to the in situ generation) (Fig. 5D). Moreover, we checked the stability of the catalyst [10: Ir(cod)(PY-PYRI)] and found highly stable that can even be stored in open air. Furthermore, to verify the broad utility of this air stable catalyst [10: Ir(cod)(PY-PYRI)], we performed several test experiments using this catalyst [10: Ir(cod)(PY-PYRI)] that was stored in open air and found no loss of catalytic activity even after 30 days (Fig. 5D, SI for details).
+A. Late-Stage meta C-H borylation:
+
+Bioactive developed conditions Bioactive Bpin
+
+6a, from cannabidiol 64% (m/others = 93/7)
+6b, from methyl salicylate 95% (m/p = 98/4)
+6c, from tyrosol, 76% (m = 100)
+6d, from Eugenol 70% (m = 100)
+6e, from sesamol, 95% (m = 100)
+6f, from naproxen, 58% (m/others = 96/4)
+6g, from deoxybutin, 91% (m = 100)
+6h, from homosalate 99%, m/p = 94/6
+
+B. Removal of Silane:
+tetraethylene glycol (0.1 M) KF (1.1 equiv.), rt, 1 h
+R = H, 7, yield: 87%
+R = 2-CF3, 8, yield: 85%
+R = 4-OCF3, 9, yield: 95%
+
+C. Catalyst synthesis:
+L1
+0.5 equiv. [Ir(cod)OMe]2 THF, rt, 1 h
+Air stable
+Catalyst 10: 90%. Confirmed by X-Ray, NMR, HRMS
+X-ray Structure of 10 CCDC No: 2180880
+
+D. Test of reactivity of Catalyst 10:
+3.0 mol% 10 1.0 equiv. Bpin2 CyH, 40-80 °C, 24 h
+R = H, 2a, 98% (m/p = 94/6)a
+R = 2-Cl, 2b, 100%, (m/p = 92/8)a
+R = 4-Cl, 4m, 100%, (m = 100)b
+
+Fig. 5. Applications, Catalyst Preparation and Testing. A. Late-stage meta-C-H borylation, B. Removal of silane group, C. Catalyst synthesis, D. Test of reactivity of catalyst 10. Reactions are in 0.5 mmol scale. aConversions were reported. See SI for details.
+
+In conclusion, we report a new class of ligand and catalyst that has demonstrated remarkable efficiency for the remote meta selective borylation of phenols featuring all types of substitutions at the arene ring. In addition, we have seen that our developed ligand system is beneficial for the remote C6-borylation of indoles derivatives including tryptophan which is a synthetic precursor of bioactive alkaloids (Verruculogen and Fumitremorgin A). Several late-stage meta borylations have been showcased with bioactive and drug molecules that might be useful for repurposing medicines and identification of new lead drug candidates. For the first time, an “O–Si” secondary interaction has been employed to tune the remote selectivity. We anticipate that the designed ligand and catalyst will also find wide application in the context of other C–H functionalization reactions.
+
+References and notes
+1. Zhang, Z., Tanaka, K. & Yu, J-Q. Remote site-selective C–H activation directed by a catalytic bifunctional template. Nature 543, 538–542 (2017).
+2. Luo, J., Preciado, S. & Larrosa, I. Overriding Ortho-Para Selectivity via a Traceless Directing Group Relay Strategy: The Meta-Selective Arylation of Phenols, J. Am. Chem. Soc. 136, 4109-4112 (2014).
+3. Sinha, S. K., Guin, S., Maiti, S., Biswas, J. P., Porey, S. & Maiti, D. Toolbox for Distal C–H Bond Functionalizations in Organic Molecules. Chem. Rev. 122, 5682-5841 (2022).
+4. Lazareva, N. F., Sterkhova, I. V. & Vashchenko, A. V. N-[difluoro(methyl)silyl]carboxamides: Synthesis, structural features and theoretical estimating of Si←O dative bond energy. Journal of Molecular Structure 1225, 129130 (2021).
+5. Lyons, T. W. & Sanford, M. S. Palladium-Catalyzed Ligand-Directed C–H Functionalization Reactions. Chem. Rev. 110, 1147-1169 (2010).
+6. Davies, H. M. L., Bois, J. D. & Yu, J-Q. C–H Functionalization in organic synthesis. Chem. Soc. Rev. 40, 1855-1856 (2011).
+7. Arockiam, P. B., Bruneau, C. & Dixneuf, P. H. Ruthenium (II)-Catalyzed C–H Bond Activation and Functionalization. Chem. Rev. **112**, 5879-5918 (2012).
+8. Sambiagio, C. et al. A comprehensive overview of directing groups applied in metal-catalysed C–H functionalisation chemistry. Chem. Soc. Rev. **47**, 6603-6743 (2018).
+9. Crabtree, R. H. & Lei, A. Introduction: CH Activation. Chem. Rev. **117**, 8481–8482 (2017).
+10. Shilov, A. E. & Shul’pin, G. B. Activation of C–H Bonds by Metal Complexes. Chem. Rev. **97**, 2879–2932(1997).
+11. Rogge, T. et al. C–H activation. Nat Rev Methods Primers **1**, (2021). doi.org/10.1038/s43586-021-00041-2.
+12. Yamaguchi, J., Yamaguchi, A. D. & K. Itami, C–H Bond Functionalization: Emerging Synthetic Tools for Natural Products and Pharmaceuticals. Angew. Chem. Int. Ed. **51**, 8960 – 9009 (2012).
+13. Dalton, T., Faber, T. & Glorius, F. C–H Activation: Toward Sustainability and Applications. ACS Cent. Sci. **7**, 245–261 (2021).
+14. Liao, K. et al. Design of catalysts for site-selective and enantioselective functionalization of non-activated primary C–H bonds. Nature Chem **10**, 1048–1055 (2018).
+15. Hoque, M. E., Hassan, M. M. M. & Chattopadhyay, B. Remarkably Efficient Iridium Catalysts for Directed C(sp2)–H and C(sp3)–H Borylation of Diverse Classes of Substrates. J. Am. Chem. Soc. **143**, 5022-5037 (2021).
+16. Hoveyda, A. H., Evans, D. A. & Fu, G. C. Substrate-directable chemical reactions. Chem. Rev. **93**, 1307–1370 (1993).
+17. Cheng, C. & Hartwig, J. F. Rhodium-Catalyzed Intermolecular C–H Silylation of Arenes with High Steric Regiocontrol. Science, **343**, 853-857(2014).
+18. Ramadoss, B., Jin, Y., Asako, S. & Ilies, L. Remote steric control for undirected meta-selective C–H activation of arenes, Science, **375**, 658-663(2022).
+19. Cho, J. Y., Tse, M. K., Holmes, D., Maleczka, R. E. & Smith, M. R. Remarkably selective iridium catalysts for the elaboration of aromatic C-H bonds. Science **295**, 305-308 (2002).
+20. Saito, Y., Segawa Y. & Itami, K. para-C–H Borylation of Benzene Derivatives by a Bulky Iridium Catalyst. J. Am. Chem. Soc., **137**, 5193-5198 (2015).
+21. Mondal, A., Chen, H., Fläming, L., Wedi, P. & Gemmeren, M. V. Sterically Controlled Late-Stage C–H Alkynylation of Arenes. J. Am. Chem. Soc. **141**, 18662–18667 (2019).
+22. Kuninobu, Y., Ida, H., Nishi, M. & Kanai, M. A meta-selective C-H borylation directed by a secondary interaction between ligand and substrate. Nat. Chem. **7**, 712-717 (2015).
+23. Fanourakis, A., Docherty, P. J., Chuentragool, P. & Phipps, R. J. Recent Developments in Enantioselective Transition Metal Catalysis Featuring Attractive Noncovalent Interactions between Ligand and Substrate, ACS Catalysis, **10**, 10672-10714 (2020).
+24. Zhang, T., Luan, Y., Lam, N. Y. S., Li, J., Li, Y., Ye, M. Yu, & J-Q. A directive Ni catalyst overrides conventional site selectivity in pyridine C–H alkynylation. Nat. Chem. **13**, 1207–1213 (2021).
+25. Hoque, M. E., Bisht, R., Haldar, C. & Chattopadhyay, B. Noncovalent Interactions in Ir-Catalyzed C–H Activation: L-Shaped Ligand for Para-Selective Borylation of Aromatic Esters. J. Am. Chem. Soc. **139**, 7745-7748 (2017).
+26. Dydio, P. & Reek, J. N. H. Supramolecular control of selectivity in transitionmetal catalysis through substrate preorganization. Chem. Sci., **5**, 2135–2145 (2014).
+27. Lou, Y., Wei, J., Li, M. & Zhu, Y. Distal Ionic Substrate–Catalyst Interactions Enable Long-Range Stereocontrol: Access to Remote Quaternary Stereocenters through a Desymmetrizing Suzuki–Miyaura Reaction. J. Am. Chem. Soc. **144**, 123–129 (2022).
+28. Bisht, R., Haldar, C., Hassan, M. M. M., Hoque, M. E., Chaturvedi, J. & Chattopadhyay, B. Metal-catalysed C–H bond activation and borylation. Chem. Soc. Rev., **51**, 5042–5100 (2022).
+29. Ros, A., Fernandez, R. & Lassaletta, J. M. Functional group directed C–H borylation. Chem. Soc. Rev. **43**, 3229-3243 (2014).
+30. Kawamorita, S., Ohmiya, H., Hara, K., Fukuoka, A. & Sawamura, M. Directed Ortho Borylation of Functionalized Arenes Catalyzed by a Silica-Supported Compact Phosphine–Iridium System. J. Am. Chem. Soc. , **131**, 5058–5059 (2019).
+31. Boebel, T. A. & Hartwig, J. F. Silyl-Directed, Iridium-Catalyzed ortho-Borylation of Arenes. A One-Pot ortho-Borylation of Phenols, Arylamines, and Alkylarenes. J. Am. Chem. Soc., **130**, 7534-7535 (2008).
+32. Bisht R. & Chattopadhyay, B. Formal Ir-Catalyzed Ligand-Enabled Ortho and Meta Borylation of Aromatic Aldehydes via In Situ-Generated Imines. J. Am. Chem. Soc., **138**, 84-87 (2016).
+33. Yang, L., Uemura, N. & Nakao, Y. meta-Selective C–H Borylation of Benzamides and Pyridines by an Iridium–Lewis Acid Bifunctional Catalyst. J. Am. Chem. Soc. **141**, 7972-7979 (2019).
+34. Davis, H. J., Madalina, M. T. & Phipps, R. J. Ion Pair-Directed Regiocontrol in Transition-Metal Catalysis: A Meta-Selective C–H Borylation of Aromatic Quaternary Ammonium Salts. J. Am. Chem. Soc. 138, 12759–12762 (2016).
+35. Chaturvedi, J., Haldar, C., Bisht, R., Pandey, G. & Chattopadhyay, B. Meta Selective C–H Borylation of Sterically Biased and Unbiased Substrates Directed by Electrostatic Interaction. J. Am. Chem. Soc. 143, 7604–7611 (2021).
+36. Mihai, M. T., Williams, B. D. & Phipps, R. J. Para-Selective C–H Borylation of Common Arene Building Blocks Enabled by Ion-Pairing with a Bulky Countercation. J. Am. Chem. Soc. 141, 15477–15482 (2019).
+37. Bastidas, J. R. M., Oleskey, T. J., Miller, S. L., Smith, M. R. & Maleczka, R. E. Para-Selective, Iridium-Catalyzed C–H Borylations of Sulfated Phenols, Benzyl Alcohols, and Anilines Directed by Ion-Pair Electrostatic Interactions. J. Am. Chem. Soc. 141, 15483–15487 (2019).
+38. Chang, W. et al. Computationally designed ligands enable tunable borylation of remote C–H bonds in arenes. Chem, 8, 1775–1788 (2022).
+39. Engle, K. M., Mei, T., Wasa, M. & Yu, J-Q. Weak Coordination as a Powerful Means for Developing Broadly Useful C–H Functionalization Reactions. Acc. Chem. Res. 45, 788–802 (2012).
+40. Leow, D., Li, G., Mei, T.-S. & Yu, J-Q. Activation of remote meta-C–H bonds assisted by an end-on template. Nature 486, 518–522 (2012).
+41. Shi, H., Herron, A. N., Shao, Y., Shao, Q. & Yu, J-Q. Enantioselective remote meta-C–H arylation and alkylation via a chiral transient mediator, Nature 558, 581–585 (2018).
+42. Gandeeapan, P. & Ackermann, L. Transient Directing Groups for Transformative C–H Activation by Synergistic Metal Catalysis. Chem 4, 199–222 (2018).
+43. Meng, G. et al. Achieving Site-Selectivity for C–H Activation Processes Based on Distance and Geometry: A Carpenter’s Approach. J. Am. Chem. Soc. 142, 10571–10591 (2020).
+44. Scott, K. A., Cox, P. B. & Njardarson, J. T. Phenols in Pharmaceuticals: Analysis of a Recurring Motif. J. Med. Chem. 65, 7044–7072 (2022).
+45. Bartolomei, B., Gentile, G., Rosso, C., Filippini, G. & Prato, M. Turning the Light on Phenols: New Opportunities in Organic Synthesis. Chem. Eur. J. 27, 16062–16070 (2021).
+46. Quideau, S., Deffieux, D., Douat-Cassasus, C. & Pouységú, L. Angew. Chem.Int. Ed. 50, 586–621 (2011).
+47. Huang, Z. & Lumb, J-P. Phenol-Directed C–H Functionalization. ACS Catal. 9, 521–555 (2019).
+48. Dai, H-X., Li, G., Zhang, X-G., Stepan, A. F. & Yu, J-Q. Pd(II)-Catalyzed ortho- or meta-C–H Olefination of Phenol Derivatives. J. Am. Chem. Soc. 135, 7567–7571(2013).
+49. Wan, L., Dastbaravaradeh, N., Li, G. & Yu, J-Q. Cross-Coupling of Remote meta-C–H Bonds Directed by a U-Shaped Template. J. Am. Chem. Soc. 135, 18056–18059 (2013).
+50. Xu, J. et al. Sequential Functionalization of meta-C–H and ipso-C–O Bonds of Phenols, J. Am. Chem. Soc. 141, 1903–1907 (2019).
+51. Iverson, C. N. & Smith, M. R. III., Stoichiometric and Catalytic B–C Bond Formation from Unactivated Hydrocarbons and Boranes. J. Am. Chem. Soc. 121, 7696 –7697 (1999).
+52. Ishiyama, T. et al. Mild iridium-catalyzed borylation of arenes. High turnover numbers, room temperature reactions, and isolation of a potential intermediate. J. Am. Chem. Soc. 124, 390 –391 (2002).
+53. Mkhalid, I. A. I., Barnard, J. H., Marder, T. B., Murphy, J. M. & Hartwig, J. F. C–H Activation for the Construction of C–B Bonds. Chem. Rev. 110, 890 –931 (2010).
+54. Hartwig, J. F. Regioselectivity of the borylation of alkanes and arenes. Chem. Soc. Rev. 40, 1992 –2002 (2011).
+55. Boller, T. M., Murphy, J. M., Hapke, M., Ishiyama, T., Miyaura, N. & Hartwig, J. F. Mechanism of the Mild Functionalization of Arenes by Diboron Reagents Catalyzed by Iridium Complexes. Intermediacy and Chemistry of Bipyridine-Ligated Iridium Trisboryl Complexes. J. Am. Chem. Soc. 127, 14263 –14278 (2005).
+56. Haldar, C., Hoque, M. E., Chaturvedi, J., Hassan, M. M. M. & Chattopadhyay, B. Ir-catalyzed proximal and distal C–H borylation of arenes. Chem. Commun., 57, 13059-13074 (2021).
+57. Muhammad, S., Bassindale, A. R., Taylor, P. G., Male, L., Coles, S. J. & Hursthouse, M. B. Study of Binuclear Silicon Complexes of Diketopiperazine at SN2 Reaction Profile. Organometallics 30, 564–571 (2011).
+58. Sohail, M. et al. Synthesis and Hydrolysis–Condensation Study of Water-Soluble Self-Assembled Pentacoordinate Polysilylamides. Organometalics 32, 1721–1731 (2013).
+59. Li, Z. et. al. A tautomeric ligand enables directed C–H hydroxylation with molecular oxygen. Science 372, 1452–1457 (2021).
+60. Lajiness, J. P. et. al. Design, Synthesis, and Evaluation of Duocarmycin O-Amino Phenol Prodrugs Subject to Tunable Reductive Activation. J. Med. Chem. 53, 7731–7738 (2010)
+61. Preshlock S. M. et al. A Traceless Directing Group for C-H Borylation. Angew. Chem., Int. Ed. **52**, 12915–12919 (2013).
+62. Li, H. L., Kuninobu, Y. & Kanai, M. Lewis Acid-Base Interaction- Controlled ortho-Selective C-H Borylation of Aryl Sulfides. Angew. Chem., Int. Ed. **56**, 1495–1499 (2017).
+63. Sun, W. et al. Chemodivergent transformations of amides using gem-diborylalkanes as pro-nucleophiles. Nat Commun **11**, 3113 (2020).
+64. Hua, Z., Vassar, V. C., Choi, H. & Ojima, I. New biphenol-based, fine-tunable monodentate phosphoramidite ligands for catalytic asymmetric transformations. Proc Natl Acad Sci, **101**, 5411-5416 (2004).
+65. Feng, Y., Holte, D., Zoller, J., Umemiya, S., Simke, L. R., Baran, P. S. Total Synthesis of Verruculogen and Fumitremorgin A Enabled by Ligand-Controlled C–H Borylation. J. Am. Chem. Soc. **137**, 10160–10163 (2015).
+66. Zhang, L. & Ritter, T. A Perspective on Late-Stage Aromatic C–H Bond Functionalization. J. Am. Chem. Soc. **144**, 2399–2414 (2022).
+67. Guillemard, L., Kaplaneris, N., Ackermann, L. & Johansson, M. J. Late-stage C–H functionalization offers new opportunities in drug discovery. Nat. Rev. Chem. **5**, 522–545 (2021).
+68. Thareja, S., Zhu, M., Ji, X., Wang, B. Boron-based small molecules in disease detection and treatment (2013 -2016). Heterocycl. Commun. **23**, 137-153 (2017).
+
+Acknowledgements
+We thank Centre of Biomedical Research (CBMR) for providing research facility. We also thank IIT Kanpur for the X-ray crystallography data collection.
+
+Funding: This work was supported by SERB-SUPRA grant (SPR/2019/000158). SD and SG thank CSIR for their JRF, MMMH thanks UGC for an SRF.
+
+Author contributions: BC conceived the concept. SG developed the ligand. SG, MMMH and SD performed the experiments. BC supervised the project. All authors contributed to writing and proofreading of manuscript and SI.
+
+Competing interests: We have filled an Indian Patent (Patent Application No: 202211036590) based on this work (including the ligand and catalyst).
+
+Data and materials availability: X-ray dataset for catalyst **10** is freely available at the Cambridge Crystallographic Data Centre under deposition number 2180880.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SupportingInformation.pdf

0ab7b0bb9760a2d34a3e8c55df78035e6e4b32c3b6a2028b02979e3615445422/preprint/preprint.md

@@ -0,0 +1,223 @@
+Decreased cloud cover partially offsets the cooling effects of surface albedo change due to deforestation
+
+Hao Luo
+luoh93@mail2.sysu.edu.cn
+
+Sun Yat-sen University    https://orcid.org/0000-0002-6648-4234
+Johannes Quaas
+Universitaet Leipzig    https://orcid.org/0000-0001-7057-194X
+Yong Han
+Sun Yat-sen University    https://orcid.org/0000-0002-3297-2782
+
+Article
+
+Keywords:
+
+Posted Date: July 23rd, 2024
+
+DOI: https://doi.org/10.21203/rs.3.rs-4019501/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on August 26th, 2024. See the published version at https://doi.org/10.1038/s41467-024-51783-y.
+Decreased cloud cover partially offsets the cooling effects of surface albedo change due to deforestation
+
+Hao Luo1,2*, Johannes Quaas2,3, Yong Han1,4*
+
+1Advanced Science & Technology of Space and Atmospheric Physics Group (ASAG), School of Atmospheric Sciences, Sun Yat-sen University, 519082 Zhuhai, China
+2Leipzig Institute for Meteorology, Leipzig University, 04103 Leipzig, Germany
+3German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig, 04103 Leipzig, Germany
+4Key Laboratory of Tropical Atmosphere-Ocean System (Sun Yat-sen University), Ministry of Education, 519082 Zhuhai, China
+
+*Corresponding author(s). Email(s): luoh93@mail2.sysu.edu.cn (Hao Luo); hany66@mail.sysu.edu.cn (Yong Han)
+
+Abstract
+
+Biophysical processes of forests affect climate through the regulation of surface water and heat fluxes, which leads to further effects through the adjustment of clouds and water cycles. These indirect biophysical effects of forests on clouds and their radiative forcing are poorly understood but highly relevant in the context of large-scale deforestation or afforestation, respectively. Here, we provide evidence for local decreases in global low-level clouds and tropical high-level clouds from deforestation through both idealized deforestation simulations with climate models and from observations-driven reanalysis using space-for-time substitution. The decreased cloud cover can be explained by alterations in surface turbulent heat flux, which diminishes uplift and moisture to varying extents. Deforestation-induced reduction in cloud cover warms the climate, partially counteracting the cooling effects of increased surface albedo. The findings from idealized deforestation experiments and space-for-time substitution exhibit disparities, with global average offsets of, respectively, approximately 44% and 26%, suggesting the necessity for further constraints.
+Introduction
+
+Forests have the capacity to buffer global warming by storing large amounts of carbon from the atmosphere via photosynthesis \(^{1-3}\). Alongside the biochemical effects, forests can influence the local and regional climate through biophysical processes, including alterations in land surface water and energy balance \(^{4-7}\). On the local scale, the higher albedo and lower evapotranspiration (ET) following deforestation cause either surface cooling or warming, depending on which process holds dominance \(^{8-10}\). These cooling or warming impacts have the potential to offset or intensify, respectively, the warming effects connected to the released carbon caused by deforestation \(^{11-16}\). Extensive studies on the direct biophysical effects of deforestation on surface temperature have unveiled a latitudinal shift from tropical warming to boreal cooling \(^{8,9,17-19}\). Nevertheless, globally, alterations in surface albedo are more prevalent in the direct biophysical temperature response than ET because of its wider-scale impact \(^{17}\). This suggests that the global warming attributed to the biochemical effects of deforestation could potentially be mitigated by the cooling effects resulting from increased surface albedo and consequently altered radiative balance \(^{12,14,16}\). Yet, the impact of forest indirect biophysical processes on clouds and their associated radiative balance has not been well addressed, and the assessment of how changes in cloud radiative effects interact with the surface albedo effects remains unquantified. Understanding the response of clouds and their radiative effects to deforestation, however, is crucial due to the overwhelming effect clouds play for the Earth energy budget. It stands as a major challenge in evaluating land-use-change-driven climate change \(^{20-24}\).
+
+Observational studies allow for the conclusion that deforestation may predominantly reduce global cloud cover \(^{22,23,25}\), but with contrasting impacts across various regions \(^{21}\). These studies mostly compare clouds above forests and open land in adjacent geographical units (i.e. space-for-time substitution) and find larger cloudiness over forests. This commonly adopted method assumes that forests and
+neighboring land units share the same climate background, thereby deducing local effects through distinctions in land surface conditions. Apart from observations-based studies, general circulation models (GCMs) have been widely employed to quantify the impacts of deforestation \(^{26-28}\). GCMs show a global average enhancement in cloud cover with deforestation \(^{29}\). Unlike the observational studies that concentrate solely on local effects, GCMs probably possess the ability to encompass both local and non-local effects of deforestation. Hence, separating local and non-local effects could facilitate comparisons between these two distinct methods and enhance comprehension of the biophysical mechanisms of deforestation on clouds \(^{24}\).
+
+Given the essential roles of cloud vertical structures in influencing radiative processes \(^{30,31}\), a sole concentration on overall cloud cover may be insufficient for a comprehensive analysis of the changes in cloud radiative effects from deforestation. Typically, low, highly reflective clouds have a cooling effect as they reflect solar radiation. In contrast, high, semi-transparent clouds contribute to warming by allowing shortwave radiation to pass through while impeding longwave radiation \(^{32,33}\). The alterations in cloud vertical profiles following deforestation have not received adequate attention, and addressing this gap is essential for gaining a deeper understanding of the consequent changes in cloud radiative effects.
+
+In this study, we approach the evaluations of cloud profiles and associated radiative response to deforestation from two distinct viewpoints: the space-for-time substitution method and the idealized deforestation experiments available from GCM simulations. Given that the outcomes from GCMs contain both local and non-local signals, we then isolate the local signals using a chessboard-like method \(^{24,34}\), enabling a comparative analysis between the two distinct ways. Using both methods, this work consistently indicates a global reduction in low-level clouds and a decline in high-level clouds over tropical regions in response to deforestation. In addition, we explore the potential physical mechanisms through which deforestation induces alterations in cloud profiles, suggesting that changes in turbulent heat flux could be a crucial factor.
+Finally, we quantify the impact of deforestation on cloud radiative forcing within the Earth-atmosphere system, with findings indicating that the warming effects of clouds to substantial extent counterbalance the cooling effects of surface albedo at a global scale.
+
+Results
+
+Cloud profile changes
+
+Two distinct approaches (see Methods) are employed in this study to assess the potential impact of deforestation on cloud fraction profiles. The first method draws upon five available GCMs (Table S1) participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) \(^{35}\). It entails analyzing the idealized global deforestation simulations (deforest-glob) conducted in the Land Use Model Intercomparison Project (LUMIP) \(^{36}\), and comparing them against the pre-industrial control simulations (piControl). The second method uses the space-for-time substitution to contrast the multi-year average cloud fraction profiles between the neighboring unaltered forested and unaltered open land grids. In this approach, the potential effects of deforestation on cloud profiles are measured by land cover data from the Moderate Resolution Imaging Spectroradiometer (MODIS) and cloud profiles from the European Centre for Medium-Range Weather Forecasts (ECMWF) fifth reanalysis (ERA5). One significant drawback of the cloud profile data from active satellites is that they have relatively small footprint and sample sizes. As a result, data from numerous satellite passes must be averaged or combined to create a product with sufficient coverage. Given the finer spatial resolution of ERA5 cloud profiles, and their much larger coverage, in comparison to the available gridded data derived from active satellite observations, along with the strong correlation exhibited between ERA5 and the observations (Fig. S1), we employ long-term ERA5 data instead. As GCMs contain both local and no-local effects, we extract the local effects from the total signals (see Methods). Isolating local effects can aid in understanding the biophysical mechanisms of deforestation on clouds. Despite the differing
+principles behind the two methods, it is noted that the space-for-time substitution also solely considers local effects, allowing for a comparison between these two approaches.
+
+While Boysen, et al. \(^{28}\) outlined diverse spatial patterns in how cloud cover responds to deforestation across GCMs in LUMIP, once the local effects are isolated, they reveal consistent spatial patterns (Fig. 1a). This implies that the inconsistencies across models documented by Boysen, et al. \(^{28}\) primarily arise from discrepancies in non-local effects. Even with distinct principles, both methods demonstrate consistent spatial signals regarding cloud vertical profile responses to deforestation (Fig. 1). The results are consistent in terms of sign, albeit with different magnitudes that can be explained by the differences between the two methods. Globally, cloud cover below 700 hPa decreases in response to deforestation, showing consistency with satellite observations \(^{21-23}\). The decrease in tropical cloud cover is restricted to relatively low altitudes according to the ERA5 space-for-time substitution method. The response to deforestation is most pronounced in tropical low-level clouds, with additional reductions found for tropical high-level clouds (>500 hPa).
+Fig. 1. Changes in cloud profile due to deforestation. (a) Zonal mean of the cloud fraction profile difference between the deforest-glob and piControl simulations (deforest-glob minus piControl). The data is the ensemble mean of the local effect extracted from CMIP6 model simulations (see Methods). The stippling represents four or more of the five models showing the same sign. (b) Zonal mean ERA5 cloud fraction profile variations that deforestation would imply using the space-for-time substitution (open land minus forest; see Methods). Only latitudes possessing more than 10 data are considered to ensure representativeness.
+
+Discussion of physical mechanisms of forest-cloud impacts
+
+Various biophysical processes are engaged in the interactions between forests and clouds, yet identifying the factors that dictate where cloud enhancement or reduction occurs across global deforested areas has remained unclear \(^{21,22,29}\). In terms
+of the thermodynamics and moisture factors involved in cloud formation, cloud cover in certain areas might be restricted by the heating needed for uplift \(^{37}\). In others, it might be restricted by the availability of moisture \(^{38}\). In the following, we explore these two fundamental factors.
+
+In comparison to forests, open land typically exhibits higher surface albedo (Fig. S2) and lower ET (Fig. S3). Increased surface albedo from deforestation causes cooling by reflecting more shortwave radiation. This cooling effect is counterbalanced by lower ET \(^{8}\). Both the cooling caused by the surface albedo difference and the warming due to ET difference vary across latitudes, indicating that the magnitude and even the sign of local land surface temperature (LST) changes resulting from alterations in forests differ across climate regions. When examining LST changes in deforested areas, shifting from forests to open land induces surface warming in tropical regions (Fig. S4). This is primarily due to the prevailing impact of ET on the temperature signal, although alterations in surface albedo partially counteract this surface warming. In contrast, the overall biophysical effect of deforestation leads to cooling in the boreal zone (Fig. S4). Notably, the impact of surface albedo becomes more pronounced as latitude increases, while the influence of evapotranspiration tends to diminish with higher latitudes. Hence, in boreal regions, increased surface albedo emerges as the predominant factor of surface cooling. Moreover, the reduction in incoming solar radiation and the drop in LST caused by the higher surface albedo results in a substantial decrease in sensible heat flux (SH) within the boreal zone; however, in the tropics, the decline in the surface turbulent heat flux primarily stems from the reduction in latent heat flux (LH) due to the dominant role of ET (Fig. S5 and Fig. 2). Thus, when combining the alterations in LH and SH, the decrease in surface turbulent heat flux depicted in Fig. 2 is evident globally. In conclusion, the response of cloud cover to the reduction in turbulent heat flux is illustrated through the decrease in water vapor supply due to decreased LH in the tropics and the weakening in uplifting process caused by decreased SH in the boreal regions.
+Fig. 2. Changes in surface turbulent heat flux due to deforestation. (a) Global pattern of the surface turbulent heat flux (latent heat (LH) + sensible heat (SH)) difference between the deforest-glob and piControl simulations (deforest-glob minus piControl). The diagonal grids indicate four or more of the five models showing the same symbol. (b) Box plots of the CMIP6 surface turbulent heat flux (LH+SH, LH and SH) differences between the deforest-glob and piControl simulations over both tropical and boreal areas. (c) ERA5 surface turbulent heat flux (LH+SH) variations due to deforestation using the space-for-time substitution (see Methods). (d) Box plots of the ERA5 surface turbulent heat flux (LH+SH, LH and SH) variations due to deforestation. The data in (a-b) is the ensemble mean of the local effect extracted from CMIP6 model simulations (see Methods). Boxes in (b and d) show the 25th to 75th percentiles of the data, whiskers display the 5th to 95th percentiles, horizontal yellow lines in the boxes represent the median values, and red dots are the mean values.
+
+Implications for radiation and climate
+
+Previous studies have concentrated on alterations in surface albedo following deforestation, yet there is a lack of quantitative analysis on changes in cloud albedo
+subsequent to deforestation \(^{22}\). Clouds on average exert a cooling effect on climate \(^{39}\).
+The decrease in cloud cover with deforestation therefore implies a warming effect on climate. The increase in surface albedo resulting from deforestation, in turn, contributes to a cooler climate \(^{17}\). Hence, clarifying the competitive relationship between these two elements is essential to the area of forest biophysical effects.
+
+For a complete analysis, we also examine the disturbance of the outgoing radiation at the top of the atmosphere (TOA). The perturbation of outgoing radiation under all-sky conditions reflects the combined impacts of alterations in both surface and cloud properties from deforestation. Under clear-sky conditions, the radiation perturbations solely arise from alterations in surface properties. Thus, the alterations in TOA outgoing radiation due to cloud cover changes can be obtained through the difference between all-sky and clear-sky conditions (all-sky minus clear-sky, also known as cloud radiative effect). As denoted in Fig. 3, a universal pattern prevails worldwide: alterations in surface properties largely govern the overall outgoing radiation changes, with changes in cloud cover acting as a buffer. When comparing the shortwave and longwave components (Figs. S6 and S7), however, it becomes evident that the perturbations to the climate come mainly from shortwave, further indicating that changes in surface and cloud albedo are the most main causes. From a global average standpoint, the quantitative competition between clouds and surface albedo becomes apparent. On average, from the CMIP6 idealized deforestation experiments, reduced cloud cover offsets approximately 44% of the surface albedo cooling effect; while from the space-for-time substitution method based on ERA5, the relative offset is about 26% (Fig. 3g). The disparities in numerical outcomes primarily result from methodological differences. Nonetheless, both methods lead to consistent conclusions. Given the saturation of CMIP6 latitudinal data, we proceed to examine the zonal disparities (Fig. 3h). The discernible result reveals that the compensatory impact of cloud cover compared to the surface albedo change is stable across latitudes, at roughly 50%. Considering that alterations in cloud cover following deforestation
+approximately counterbalance half of the cooling effect caused by changes in surface albedo, neglecting the shifts in cloud-climate interactions introduces a large bias when investigating the biophysical effects of forests in the future.
+
+![Global patterns of TOA outgoing radiation (shortwave) for CMIP6 MME All-sky, ERA5 All-sky, Clear-sky, All-sky minus clear-sky, and bar and line plots showing global mean and zonal mean differences](page_374_282_1097_1012.png)
+
+Fig. 3. Changes in outgoing radiation at the top of atmosphere (TOA) due to deforestation. (a, c, and e) Global pattern of the TOA outgoing radiation (shortwave
++ longwave) difference between the deforest-glob and piControl simulations (deforest-glob minus piControl), respectively, under all-sky, clear-sky, and all-sky minus clear-sky circumstances. The diagonal grids indicate four or more of the five models showing the same symbol. (b, d, and f) ERA5 TOA outgoing radiation (shortwave + longwave) variations due to deforestation using the space-for-time substitution (see Methods). Global mean values and standard errors for (a-f) are shown in (g). The offset ratio is the proportion of all-sky minus clear-sky to the all-sky value. (h) Zonal mean of the TOA outgoing radiation (shortwave + longwave) difference between the deforest-glob and piControl simulations under both clear-sky and all-sky minus clear-sky circumstances. The black line indicates the zonal mean offset ratio and the dashed yellow line is the ratio equal to −0.5. The CMIP6 data is the ensemble mean of the local effect extracted from multi-model simulations (see Methods).
+
+Methods
+
+CMIP6 simulations
+
+Cloud fraction profile, tree cover fraction, surface latent heat flux (LH), sensible heat flux (SH), air temperature, radiation fluxes and evapotranspiration (ET), as well as radiation fluxes at the top of atmosphere (TOA) from five available general climate models (GCMs) (Table S1) participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) are adopted in this study \(^{38}\). The idealized global deforestation simulations (deforest-glob) from the Land Use Model Intercomparison Project (LUMIP) \(^{39}\) are analyzed in comparison to the pre-industrial control simulations (piControl). The deforest-glob setup assumes that a total forest area of 20 million km\(^2\) is linearly removed from the top 30% forested area with a fixed rate of 400 000 km\(^2\) yr\(^{-1}\) over a period of 50 years across the globe. This is then followed by at least a 30-year simulation with a constant land cover to achieve stable conditions. The last 30 years of the deforest-glob and piControl simulations are compared (deforest-glob minus piControl) to derive the mean response to deforestation \(^{28}\). Due
+to differences in resolution among GCMs, the ensemble mean statistics are calculated by bilinear remapping of diagnostics from individual GCMs to a \(2^\circ \times 2^\circ\) grid, and vertically to 27 pressure levels from 1000 to 100 hPa.
+
+Reanalysis datasets
+
+From the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5\(^{40}\), we utilize the ERA5 cloud fraction profiles data alongside elevation, surface LH, SH, air temperature, radiation fluxes, ET, and TOA radiation fluxes to examine the impacts of deforestation. Datasets spanning from 2001 to 2021, featuring a spatial resolution of \(0.25^\circ \times 0.25^\circ\) and encompassing 28 vertical pressure levels from 1000 to 100 hPa, are employed for the analysis.
+
+Observed land cover
+
+For delineating forested and open land areas, we use land cover data from the Moderate resolution imaging spectroradiometer (MODIS) dataset (MCD12C1, version 6.1)\(^{41}\), relying on the International Geosphere-Biosphere Program (IGBP) classification layer to define the land cover types. Annual data for the years 2001–2021 with a spatial resolution of \(0.05^\circ \times 0.05^\circ\) are adopted. Here, five forest types (evergreen needleleaf forest, evergreen broadleaf forest, deciduous needleleaf forest, deciduous broadleaf forest and mixed forest) are merged into a single forest classification. The forest fraction is bilinearly gridded spatially into \(0.25^\circ \times 0.25^\circ\) to align with the ERA5 data.
+
+Observed cloud profile
+
+In assessing the accuracy of ERA5 cloud profiles, we analyse active satellite-observed cloud profiles. The cloud profile retrievals from Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) and CloudSat between 2007 and 2010, are aggregated to a spatial resolution of \(2^\circ \times 2^\circ\) and a vertical resolution of 480 meters\(^{42}\). The fusion of data from both sensors facilitates an extensive depiction of the vertical cloud structure. This comprehensive view is achieved by leveraging the
+distinct wavelengths each sensor employs (CloudSat: approximately 2 mm, CALIPSO: 532 nm and 1064 nm), catering to various cloud and precipitation particles in both liquid and solid phases.
+
+Climate zones
+
+In this study, climate zones are defined according to the global maps of the Köppen-Geiger climate classification (Version 1) \(^{43}\). The Köppen-Geiger historical map contains 30 climate zones at a resolution of 1 km. Tropical and boreal regions are each merged from corresponding subdivided climate zones.
+
+Extracting local effect from GCMs
+
+Deforestation exerts a local impact on the climate within deforested areas (local effect) by modifying land surface characteristics such as albedo, roughness, and ET. Additionally, it affects both deforested and open land grids by altering the advection of heat and moisture, as well as influencing atmospheric circulation (non-local effect) \(^{44}\). Distinguishing between local and non-local effects within GCMs is crucial as coupled models encompass the complete climate response to deforestation, incorporating both local and non-local impacts. Moreover, it allows to develop a more profound insight into the mechanisms influencing the local effects in comparison to those governing the non-local effects.
+
+Here, we use a chessboard method as outlined by Winckler, et al. \(^{34}\) to assess the local effect. This method assumes that the unaltered and adjacent deforested grids share the same non-local effect \(^{21,44}\). To generate a global map of the non-local effect, we spatially interpolate the non-local signal to the adjacent deforested regions, maintaining the original values over the unaltered grids unchanged. The local effect over the deforested grids thus can be derived by subtracting the interpolated non-local effect from the total effect. Notably, employing a chessboard-like method introduces horizontal interpolation errors, given that the local effect relies solely on interpolation from neighboring, unaltered grids. However, our study is centered on idealized deforestation scenarios and prior a study \(^{24}\) has demonstrated the possibility of
+isolating local effects using similar methodologies and datasets. Winckler, et al. \(^{34}\) conducted comparisons between simulations involving both sparse and extensive idealized deforestation, finding small differences in derived local effects from spatial interpolation.
+
+**Space-for-time substitution**
+
+In addition to idealized deforestation simulations, this study employs a space-for-time substitution method to assess the impacts of deforestation combining MODIS land cover and ERA5 reanalysis datasets. Such an approach has previously been applied in various studies to evaluate the effect of alterations in land cover on temperature \(^{8,26}\), the surface energy budget \(^{5,45}\), or cloud cover \(^{21,22}\). The fundamental premise of this method is that neighboring land patches share the same climatic background and variations in their characteristics can act as a proxy for temporal changes. This method exclusively includes the local effects, making it well-suited for assessing the isolated local effects derived from GCMs.
+
+Areas designated as unaltered forested (or unaltered open land) are identified as pixels where the initial (in 2001) tree cover fraction exceeds 60% (or is below 40%) and with a net change in forest cover <10% from 2001 to 2021. Pixels with water coverage >10% are excluded. We use a moving window approach to search for comparison samples between unaltered forested and unaltered open land pixels. We choose for each moving window a size of \(7 \times 7\) pixels (\(1.75^\circ \times 1.75^\circ\)). To reduce the influence of topography, we calculate the standard deviation (s.d.) of elevation within specific moving windows and omit samples where this s.d. exceeds 100 m following Xu, et al. \(^{21}\). Finally, the potential effect of deforestation on a specific variable (\(\Delta\)Var) is quantified as:
+
+\[
+\Delta \mathrm{Var} = \mathrm{Var}_{\text{open land}} - \overline{\mathrm{Var}}_{\text{surrounding forests}}
+\]
+(1)
+
+or
+
+\[
+\Delta \mathrm{Var} = \mathrm{Var}_{\text{surrounding open lands}} - \mathrm{Var}_{\text{forest}}
+\]
+(2)
+where equations (1) and (2) are applicable to the case where the central pixel of the moving window is unaltered open land and unaltered forest, respectively. \( \mathrm{Var}_{\text{open land}} \) and \( \mathrm{Var}_{\text{forest}} \) are multi-year mean variables over unaltered open land and unaltered forest pixels, respectively. \( \mathrm{Var}_{\text{surrounding forests}} \) and \( \mathrm{Var}_{\text{surrounding open lands}} \) are the average values of the surrounding \( \mathrm{Var}_{\text{forest}} \) and \( \mathrm{Var}_{\text{open land}} \) within a moving window when the central pixel is unaltered open land and unaltered forest, respectively.
+
+**Data availability**
+
+The data that support the findings of this study are publicly available. The CMIP6 data are taken from https://esgf-data.dkrz.de/search/cmip6-dkrz/. The ERA5 cloud fraction profile data are obtained from https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-pressure-levels-monthly-means?tab=overview. Other ERA5 datasets are available from https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels-monthly-means?tab=overview. MODIS land cover data are obtained from https://lpdaac.usgs.gov/products/mcd12c1v061/. CALIPSO-CloudSat cloud profile data are taken from https://www.cen.uni-hamburg.de/en/icdc/data/atmosphere/calipso-cloudsat-cloudcover.html. The Köppen-Geiger historical map is available from https://figshare.com/articles/dataset/Present_and_future_K_ppen-Geiger_climate_classification_maps_at_1-km_resolution/6396959/2.
+
+**Code availability**
+
+The code used in the work can be obtained upon request from the corresponding author.
+
+**Acknowledgments**
+
+This research has been supported by the National Natural Science Foundation of China (grant nos. 42027804, 41775026, 41075012).
+
+**Author contributions**
+J.Q., H.L. and Y.H. designed the research. H.L. performed the research and drafted the paper. H.L., J.Q. and Y.H. contributed to analysis and interpretation of the results, as well as revising the paper.
+
+Competing interests
+
+The authors declare no competing interests.
+
+References
+
+1 Bonan, G. B. Forests and Climate Change: Forcings, Feedbacks, and the Climate Benefits of Forests. Science **320**, 1444-1449, doi:10.1126/science.1155121 (2008).
+2 Nabuurs, G.-J. *et al.* First signs of carbon sink saturation in European forest biomass. *Nature Climate Change* **3**, 792-796, doi:10.1038/nclimate1853 (2013).
+3 Baccini, A. *et al.* Estimated carbon dioxide emissions from tropical deforestation improved by carbon-density maps. *Nature Climate Change* **2**, 182-185, doi:10.1038/nclimate1354 (2012).
+4 Runyan, C. W., D’Odorico, P. & Lawrence, D. Physical and biological feedbacks of deforestation. *Reviews of Geophysics* **50**, doi:10.1029/2012RG000394 (2012).
+5 Duveiller, G., Hooker, J. & Cescatti, A. The mark of vegetation change on Earth’s surface energy balance. *Nature Communications* **9**, 679, doi:10.1038/s41467-017-02810-8 (2018).
+6 Perugini, L. *et al.* Biophysical effects on temperature and precipitation due to land cover change. *Environmental Research Letters* **12**, 053002, doi:10.1088/1748-9326/aa6b3f (2017).
+7 Bright, R. M. *et al.* Local temperature response to land cover and management change driven by non-radiative processes. *Nature Climate Change* **7**, 296-302, doi:10.1038/nclimate3250 (2017).
+8 Li, Y. *et al.* Local cooling and warming effects of forests based on satellite observations. *Nature Communications* **6**, 6603, doi:10.1038/ncomms7603 (2015).
+9 Lee, X. *et al.* Observed increase in local cooling effect of deforestation at higher latitudes. *Nature* **479**, 384-387, doi:10.1038/nature10588 (2011).
+10 Williams, C. A., Gu, H. & Jiao, T. Climate impacts of U.S. forest loss span net warming to net cooling. *Science Advances* **7**, eaax8859, doi:10.1126/sciadv.aax8859 (2021).
+11 Bala, G. *et al.* Combined climate and carbon-cycle effects of large-scale deforestation. *Proceedings of the National Academy of Sciences* **104**, 6550-6555, doi:10.1073/pnas.0608998104 (2007).
+Betts, R. A. Offset of the potential carbon sink from boreal forestation by decreases in surface albedo. Nature **408**, 187-190, doi:10.1038/35041545 (2000).
+
+Claussen, M., Brovkin, V. & Ganopolski, A. Biogeophysical versus biogeochemical feedbacks of large-scale land cover change. Geophysical Research Letters **28**, 1011-1014, doi:10.1029/2000GL012471 (2001).
+
+Arora, V. K. & Montenegro, A. Small temperature benefits provided by realistic afforestation efforts. Nature Geoscience **4**, 514-518, doi:10.1038/ngeo1182 (2011).
+
+Li, Y. *et al.* Deforestation-induced climate change reduces carbon storage in remaining tropical forests. Nature Communications **13**, 1964, doi:10.1038/s41467-022-29601-0 (2022).
+
+Windisch, M. G., Davin, E. L. & Seneviratne, S. I. Prioritizing forestation based on biogeochemical and local biogeophysical impacts. Nature Climate Change **11**, 867-871, doi:10.1038/s41558-021-01161-z (2021).
+
+Davin, E. L. & de Noblet-Ducoudrè, N. Climatic Impact of Global-Scale Deforestation: Radiative versus Nonradiative Processes. Journal of Climate **23**, 97-112, doi:10.1175/2009JCLI3102.1 (2010).
+
+Snyder, P. K., Delire, C. & Foley, J. A. Evaluating the influence of different vegetation biomes on the global climate. Climate Dynamics **23**, 279-302, doi:10.1007/s00382-004-0430-0 (2004).
+
+Zhang, M. *et al.* Response of surface air temperature to small-scale land clearing across latitudes. Environmental Research Letters **9**, 034002, doi:10.1088/1748-9326/9/3/034002 (2014).
+
+Norris, J. R. *et al.* Evidence for climate change in the satellite cloud record. Nature **536**, 72-75, doi:10.1038/nature18273 (2016).
+
+Xu, R. *et al.* Contrasting impacts of forests on cloud cover based on satellite observations. Nature Communications **13**, 670, doi:10.1038/s41467-022-28161-7 (2022).
+
+Duveiller, G. *et al.* Revealing the widespread potential of forests to increase low level cloud cover. Nature Communications **12**, 4337, doi:10.1038/s41467-021-24551-5 (2021).
+
+Teuling, A. J. *et al.* Observational evidence for cloud cover enhancement over western European forests. Nature Communications **8**, 14065, doi:10.1038/ncomms14065 (2017).
+
+Hua, W., Zhou, L., Dai, A., Chen, H. & Liu, Y. Important non-local effects of deforestation on cloud cover changes in CMIP6 models. Environmental Research Letters **18**, 094047, doi:10.1088/1748-9326/acf232 (2023).
+
+Cerasoli, S., Yin, J. & Porporato, A. Cloud cooling effects of afforestation and reforestation at midlatitudes. Proceedings of the National Academy of Sciences **118**, e2026241118, doi:10.1073/pnas.2026241118 (2021).
+
+Chen, L. & Dirmeyer, P. A. Reconciling the disagreement between observed
+and simulated temperature responses to deforestation. Nature Communications **11**, 202, doi:10.1038/s41467-019-14017-0 (2020).
+27 Ge, J. *et al.* Local surface cooling from afforestation amplified by lower aerosol pollution. Nature Geoscience **16**, 781-788, doi:10.1038/s41561-023-01251-x (2023).
+28 Boysen, L. R. *et al.* Global climate response to idealized deforestation in CMIP6 models. Biogeosciences **17**, 5615-5638, doi:10.5194/bg-17-5615-2020 (2020).
+29 Portmann, R. *et al.* Global forestation and deforestation affect remote climate via adjusted atmosphere and ocean circulation. Nature Communications **13**, 5569, doi:10.1038/s41467-022-33279-9 (2022).
+30 Luo, H., Quaas, J. & Han, Y. Examining cloud vertical structure and radiative effects from satellite retrievals and evaluation of CMIP6 scenarios. Atmos. Chem. Phys. **23**, 8169-8186, doi:10.5194/acp-23-8169-2023 (2023).
+31 Chen, T., Rossow, W. B. & Zhang, Y. Radiative Effects of Cloud-Type Variations. Journal of Climate **13**, 264-286, doi:10.1175/1520-0442(2000)013<0264:REOCTV>2.0.CO;2 (2000).
+32 Slingo, A. Sensitivity of the Earth's radiation budget to changes in low clouds. Nature **343**, 49-51, doi:10.1038/343049a0 (1990).
+33 Lohmann, U. & Roeckner, E. Influence of cirrus cloud radiative forcing on climate and climate sensitivity in a general circulation model. Journal of Geophysical Research: Atmospheres **100**, 16305-16323, doi:10.1029/95JD01383 (1995).
+34 Winckler, J., Reick, C. H. & Pongratz, J. Robust Identification of Local Biogeophysical Effects of Land-Cover Change in a Global Climate Model. Journal of Climate **30**, 1159-1176, doi:10.1175/JCLI-D-16-0067.1 (2017).
+35 Eyring, V. *et al.* Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev. **9**, 1937-1958, doi:10.5194/gmd-9-1937-2016 (2016).
+36 Lawrence, D. M. *et al.* The Land Use Model Intercomparison Project (LUMIP) contribution to CMIP6: rationale and experimental design. Geosci. Model Dev. **9**, 2973-2998, doi:10.5194/gmd-9-2973-2016 (2016).
+37 Bony, S. *et al.* Clouds, circulation and climate sensitivity. Nature Geoscience **8**, 261-268, doi:10.1038/ngeo2398 (2015).
+38 Sherwood, S. C., Roca, R., Weckwerth, T. M. & Andronova, N. G. Tropospheric water vapor, convection, and climate. Reviews of Geophysics **48**, doi:10.1029/2009RG000301 (2010).
+39 Zelinka, M. D., Randall, D. A., Webb, M. J. & Klein, S. A. Clearing clouds of uncertainty. Nature Climate Change **7**, 674-678, doi:10.1038/nclimate3402 (2017).
+40 Hersbach, H. *et al.* The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society **146**, 1999-2049, doi:10.1002/qj.3803 (2020).
+Sulla-Menashe, D., Gray, J. M., Abercrombie, S. P. & Friedl, M. A. Hierarchical mapping of annual global land cover 2001 to present: The MODIS Collection 6 Land Cover product. Remote Sensing of Environment 222, 183-194, doi:10.1016/j.rse.2018.12.013 (2019).
+
+Kay, J. E. & Gettelman, A. Cloud influence on and response to seasonal Arctic sea ice loss. Journal of Geophysical Research: Atmospheres 114, doi:10.1029/2009JD011773 (2009).
+
+Beck, H. E. et al. Present and future Köppen-Geiger climate classification maps at 1-km resolution. Scientific Data 5, 180214, doi:10.1038/sdata.2018.214 (2018).
+
+Pongratz, J. et al. Land Use Effects on Climate: Current State, Recent Progress, and Emerging Topics. Current Climate Change Reports 7, 99-120, doi:10.1007/s40641-021-00178-y (2021).
+
+Liu, Z., Ballantyne, A. P. & Cooper, L. A. Biophysical feedback of global forest fires on surface temperature. Nature Communications 10, 214, doi:10.1038/s41467-018-08237-z (2019).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• forestcloudSupplementary.pdf

0acd819abe48c47a84cbc21e7054100bd0d1363bbc114db43e3926604a5e536e/peer_review/peer_review.md

@@ -0,0 +1,191 @@
+Peer Review File
+Electrically Driven Spin Resonance of 4f Electrons in a Single Atom on a Surface
+
+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/.
+Reviewers' Comments:
+
+Reviewer #2:
+Remarks to the Author:
+In the manuscript titled “Electrically Driven Spin Resonance of 4f Electrons in a Single Atom on a Surface” submitted by Reale et al., the authors perform ESR-STM experiments on hydrogenated Ti to drive spin resonance in a nearby 4f atom, namely Erbium. Experiments were performed at different inter-atomic distances of this dimer and it was tried to learn about the T1 and T2 times as well as the driving force of the 4f electrons via this remote ESR-STM technique.
+
+Overall the manuscript is very well and clearly written and the figures are clear. The performed experiments are sound and explained well. Comparing the work to other ESR-STM related publications in this journal, I would adjudicate the presented manuscript of similar quality, importance and novelty. Hence, I believe Nature Communications is a good fit for this manuscript.
+
+However, I do have some points that have to be addressed by the authors prior considering the manuscript for publication. Please read below:
+
+1. Line 40: Reference 18 is a wrong link. But should be updated to be pointing towards the recent Science publication of the group (https://www.science.org/stoken/author-tokens/ST-1487/full)
+
+2. While the introduction focuses a lot on the strategy to improve coherence times and driving mechanisms of spin states and use 4f states for this purpose, the later results in the manuscript cannot hold up to this anymore. In the end, the coherence time could not directly be measured (probably due to the weak coupling) and the driving force was only estimated, as the authors state. Therefore, I suggest to soften the claims given in the introduction, especially towards the end, a bit. Or to shift it more towards the successful experiments that are presented. Another, maybe additional, option could also be to bring parts of the supplement into the main text and show, e.g. Fig S8 in the main paper. This way more of the discussion about this topic enters the main text and is not omitted so quickly towards the end. I also would appreciate more clear statements about what can be said or not from the presented experiments in the conclusions. If more experiments need to be done in the future this is exciting and something to be written.
+
+3. I am wondering whether in the presented manuscript the same Ti species as previously and also in the recent Science publication (ref. 18) has been used. In the latter, the authors mention that the Ti is a hydrogenated atom, whereas this is not mentioned anymore in the present manuscript. Is there new insights that the Ti is not hydrogenated, or was it just not mentioned? This should be clarified.
+
+4. For example in Forrester et al., a 4f atom (Ho) was directly addressed via SP-STM methods. I assume the authors did, but don’t mention it: Did the authors try different elements, e.g. also Ho, to drive the ESR directly and not remotely via a Ti spin? If so, was it not successful? Besides the remote driving, I believe this would be very exciting, despite the probable reduction in T1/T2 times. A brief statement about this in the manuscript would be very helpful for the whole community.
+
+5. The authors write about a 5x longer T1 time compared to 3d atoms measured on the same surface. However, while I agree with the general statement, I find it a bit too much to directly put numbers to these measured times. There are many parameters at play that can be different in these two scenarios that I would not say this can be a 1:1 comparison. Magnetic moments, couplings, the actual applied magnetic field, the state mixing, influences of the tip to name a few. I would ask for a more general statement of this finding.
+
+6. Maybe I am misunderstanding, but in Figure 1c in the case of AFM coupling, should the first peak not only be smaller in the regime where J is smaller than B? Here we have a coupling of 400
+MHz but a magnetic field of 0.3T, which at 1\( \mu_B \) already corresponds to about 4.6GHz of Zeeman splitting. Could the authors please clarify my confusion here?
+
+7. In Figure 3c: How do I have to picture an upward relaxation mechanism (Gamma_2)? Or are the levels drawn there no energy levels?
+
+8. Line 167: What are the significant digits and error bars of both of these g-factors? How were they determined? And as for the latter one, wasn’t it reported to be anisotropic?
+
+9. When reading lines 149/150 and 200, it is not fully clear to me how the peaks/dips behave for the ESR signal of the Er. Are the Ti and Er peaks always opposite (peak vs dip)? Or is the Er signal tip dependent and can be both, either a peak or a dip?
+
+10. Line 219: Should the f4Ti not be f4Er?
+
+Reviewer #3:
+Remarks to the Author:
+The authors present in this work an experimental approach to drive electron-spin resonance (ESR) on Er atoms (lanthanide atoms with 4f electrons) and find relatively long spin relaxation times, as compared with atoms of transition elements. The approach is based on the coupling of Er atoms with Ti atoms, which act as detectors.
+
+The main conclusion of this work is of great importance for the ESR community and widens the possibilities to use single atoms to do electrically driven quantum coherent control of single spins on surfaces. The work is scientifically sound, the manuscript is well written, and the main conclusions are clearly stated. So, in this sense, I am certainly inclined to recommend the publication of this manuscript provided several issues are clarified or better described in a revised version. Below, I proceed to list those issues.
+
+1) For didactic reasons, and as a reference, it would have been good to add in the main manuscript ESR spectra of isolated Ti atoms. The authors could at least mention in a revised version that those spectra are reported in Fig. S3 in the SI. In general, I think that the authors could do a better job in the manuscript specifying where to find the information in the SI.
+
+2) As the authors explained in the manuscript, no direct ESR signal was possible to acquire when they placed the SP-STM tip on top of the Er atoms. They also speculate about the reason for that. Maybe they can elaborate a bit more using the information provided in Fig. S1, namely it would be interesting to know if the differential conductance spectra in the absence of microwaves already hint at the absence of an ESR signal when the microwaves are applied.
+
+3) From the text in the manuscript, it is a little bit unclear what the authors mean by a term originating from the spin-polarized current in the context of the rate equation model summarized in Fig. 3c. I know that they elaborate on this in section 7 of the SI, but readers would benefit from a more detailed/didactic explanation of the meaning of this spin pumping.
+
+4) Related to the previous issue, in the discussion of the rate equation model (section 7 of the SI), it is a bit unclear to me how the solution of the rate equation is finally converted into the ESR signal (\Delta I in Fig. S7). Could the authors elaborate more in the SI about this issue?
+
+5) The authors measure the relaxation time in the Er atoms with a sophisticated protocol that I must admit I have not quite understood. Could the authors make this discussion at the beginning of section “Relaxation Time Measurement ...” somehow more accessible to non-specialists?
+6) Related to the previous issue, the authors report the Er relaxation time as a single number. Should I understand that this is the intrinsic relaxation time of Er on this surface? How can one avoid the influence of the Ti atom?
+Reply to Reviewers
+
+Reviewer #2 (Remarks to the Author):
+
+In the manuscript titled “Electrically Driven Spin Resonance of 4f Electrons in a Single Atom on a Surface” submitted by Reale et al., the authors perform ESR-STM experiments on hydrogenated Ti to drive spin resonance in a nearby 4f atom, namely Erbium. Experiments were performed at different inter-atomic distances of this dimer and it was tried to learn about the T1 and T2 times as well as the driving force of the 4f electrons via this remote ESR-STM technique.
+
+Overall, the manuscript is very well and clearly written, and the figures are clear. The performed experiments are sound and explained well. Comparing the work to other ESR-STM related publications in this journal, I would adjudicate the presented manuscript of similar quality, importance, and novelty. Hence, I believe Nature Communications is a good fit for this manuscript.
+
+However, I do have some points that have to be addressed by the authors prior considering the manuscript for publication. Please read below:
+
+1. Line 40: Reference 18 is a wrong link. But should be updated to be pointing towards the recent Science publication of the group (https://www.science.org/stoken/author-tokens/ST-1487/full)
+
+Reply: We appreciate the reviewer’s careful reading and valuable comments with her/his support for our paper. We corrected the reference and updated the link.
+
+2. While the introduction focuses a lot on the strategy to improve coherence times and driving mechanisms of spin states and use 4f states for this purpose, the later results in the manuscript cannot hold up to this anymore. In the end, the coherence time could not directly be measured (probably due to the weak coupling) and the driving force was only estimated, as the authors state. Therefore, I suggest to soften the claims given in the introduction, especially towards the end, a bit. Or to shift it more towards the successful experiments that are presented. Another, maybe additional, option could also be to bring parts of the supplement into the main text and show, e.g. Fig S8 in the main paper. This way more of the discussion about this topic enters the main text and is not omitted so quickly towards the end. I also would appreciate more clear statements about what can be said or not from the presented experiments in the conclusions. If more experiments need to be done in the future this is exciting and something to be written.
+
+Reply: We agree with the reviewer. We corrected the manuscript by softening the claims. The corresponding part in abstract reads "The erbium spin states exhibit extended spin relaxation time and a higher driving efficiency compared to the 3d atoms with spin \( \frac{1}{2} \) in similarly coupled structures." In the introduction, we changed "coherent manipulation" to "resonant driving" (page 2).
+
+We appreciate the reviewer’s suggestion about moving contents (e.g. Fig. S8) from supplementary materials. However, the model descriptions in the SI might be too specific for a broad readership. In addition, while Ti shows clear Rabi oscillations in Fig. S8, we could not observe such oscillations for Er. Since it is unclear whether this absence of Rabi oscillations is due to the initialization of Er spins, to insufficient driving of Er, or to something else, we decided to keep this figure as a part of the supplementary information.
+
+In addition, we added a short perspective to the conclusions at page 13: “This allows one to develop more advanced pulse sequences for quantum coherent manipulation on atomic-scale spin platforms. We expect that, by employing a similar approach in different atomic structures, we can reduce the influence of the
+spin fluctuations of the atom used for the detection and amplify the ESR driving on the 4f electrons, enabling the use of lanthanide atoms as surface spin qubits with superior properties compared to the routinely adopted 3d elements.".
+
+3. I am wondering whether in the presented manuscript the same Ti species as previously and also in the recent Science publication (ref. 18) has been used. In the latter, the authors mention that the Ti is a hydrogenated atom, whereas this is not mentioned anymore in the present manuscript. Is there new insights that the Ti is not hydrogenated, or was it just not mentioned? This should be clarified.
+
+Reply: The reviewer is correct. The Ti species is the same as in previous ESR-STM works. While the hydrogenation of Ti has been supported by previous works to explain the spin-1/2 state of Ti on MgO, recent studies suggest other possibilities, such as charge transfer. Since this point is still under debate, we originally preferred not to specify the hydrogenation of Ti atoms. However, we agree with the reviewer on the need of being consistent with other recent works. Hence, we added one sentence in Methods at page 14, which reads: "As described in previous works the Ti atoms on 2 ML MgO/Ag(100) shows spin \( \frac{1}{2} \) behavior which is presumably originating from hydrogenation."
+
+4. For example in Forrester et al., a 4f atom (Ho) was directly addressed via SP-STM methods. I assume the authors did, but don’t mention it: Did the authors try different elements, e.g. also Ho, to drive the ESR directly and not remotely via a Ti spin? If so, was it not successful? Besides the remote driving, I believe this would be very exciting, despite the probable reduction in T1/T2 times. A brief statement about this in the manuscript would be very helpful for the whole community.
+
+Reply: We appreciate the reviewer's suggestion. While we focus on the potential use of 4f atoms on surfaces as a qubit, Ho on MgO is known as a single atom magnet. We haven't tried to do ESR driving on Ho, but we expect the resonant driving of the Ho to be quite challenging or even not feasible since the ground state of Ho has a large magnetic quantum number and the two lowest-lying states are separated by a large anisotropy barrier. We added a brief statement about this issue in the revised manuscript (page 2), which reads: "This magnetic level scheme differs from the ones of lanthanide single atom magnets studied so far on MgO/Ag(100). For instance, dysprosium and holmium present a ground state characterized by a large \( J_n \). The level scheme presents two lowest-lying states well separated by a significant anisotropy barrier and greatly suppresses the reversal of angular momentum, thereby stabilizing the magnetic states. Additionally, it impedes the first-order ESR transition induced by the exchange of a single quantum of angular momentum." In addition, we added the reference mentioned by the reviewer as the new reference 27.
+
+5. The authors write about a 5x longer T1 time compared to 3d atoms measured on the same surface. However, while I agree with the general statement, I find it a bit too much to directly put numbers to these measured times. There are many parameters at play that can be different in these two scenarios that I would not say this can be a 1:1 comparison. Magnetic moments, couplings, the actual applied magnetic field, the state mixing, influences of the tip to name a few. I would ask for a more general statement of this finding.
+
+Reply: The reviewer is totally correct. We revised the statement in the abstract, which now reads: "The erbium spin states exhibit an extended spin relaxation time and a higher driving efficiency compared to 3d
+atoms with spin \( \frac{1}{2} \) in similarly coupled structures." In addition, in the introduction (page 2), we specified the case of comparison as "a remotely-driven spin-\( \frac{1}{2} \) system" in the sentence starting "We observed an Er T_1 of close to 1 \( \mu \)s, ..."
+
+6. Maybe I am misunderstanding, but in Figure 1c in the case of AFM coupling, should the first peak not only be smaller in the regime where J is smaller than B? Here we have a coupling of 400 MHz but a magnetic field of 0.3T, which at 1\( \mu \)B already corresponds to about 4.6GHz of Zeeman splitting. Could the authors please clarify my confusion here?
+
+Reply: To better explain the effect of the Er-Ti interaction on the Ti ESR peak intensity, we added a new section (now Supplementary Section 4) in the SI, entitled “Effect of the Er-Ti coupling on the ESR spectra of Ti”. In this section, we introduce the total interaction tensor \( \vec{J}_{int} \) (Eq. S2) as the sum of the dipolar tensor (Eq. S1) and exchange interaction energy (Eq. 2). For a fixed Er-Ti separation at a constant magnetic field angle, this interaction tensor \( \vec{J}_{int} \) can be reduced to a scalar \( J_{int} \), providing a direct visualization of its influence on the energy levels and the resulting ESR spectra. Figure S4 represents three possible cases for \( J_{int} \) being zero, positive, and negative with its magnitude smaller than the Zeeman energy, as experimentally realized in this work.
+
+In Fig. 1c of the main text, the stronger signal appears at a higher frequency, which indicates the FM coupling in the given experimental condition, now schematized in Fig. S4b,e. As summarized in Fig. 1d, the coupling strength and polarity changes depending on the magnetic field direction due to the dominant contribution of the dipole interaction in the Ti-Er dimer with 0.928 nm separation. In the rest of figures in the main text, the Ti-Er dimer with 0.72 nm separation shows exceedingly higher contribution of antiferromagnetic exchange interactions than the dipole coupling, which shows the higher peaks at lower frequencies corresponding to the schematic in Fig. S4c,f.
+
+![Four-level scheme of Er-Ti dimers with different interaction strengths and resulting ESR spectra](page_352_1012_1127_377.png)
+
+Figure S4 | Influence of the Er-Ti interaction on the energy levels and ESR spectra. a,b,c, Four-level scheme of Er-Ti dimers with \( J_{int} = 0 \), \( J_{int} < 0 \) (ferromagnetic) and \( J_{int} > 0 \) (antiferromagnetic) respectively. d,e,f, Schematics of the resulting ESR spectra on Ti in its resonance frequency range. When no interaction is present (a) only one peak is detectable (d). A ferromagnetic interaction (\( J_{int} < 0 \)) shifts the antiparallel levels to higher energies (b), resulting in two distinguishable peaks with the higher intensity one (\( f_1^{Ti} \)) at a higher frequency (e). An antiferromagnetic interaction (\( J_{int} > 0 \)) shifts the antiparallel levels to lower energies (c), resulting in two peaks with the higher intensity one (\( f_1^{Ti} \)) at a lower frequency (f).
+
+To further clarify this point, we added a sentence at page 3, which reads “When isolated, a nuclear spin-free Ti atom presents a single ESR signal under an external magnetic field (see Fig. S3a). The ESR peak of
+Ti splits when coupled to an Er atom (Supplementary Section 4)." We also revised the caption of Fig. 1c, which now reads "For the latter, the relative peak intensity indicates a ferromagnetic interaction." In addition, all over the text and SI we aligned the sign of \( J_{int} \) to match the previous literature, with positive/negative sign indicating antiferromagnetic/ferromagnetic coupling, respectively.
+
+7. In Figure 3c: How do I have to picture an upward relaxation mechanism (Gamma_2)? Or are the levels drawn there no energy levels?
+
+Reply: To clarify the reviewer’s point and explain this upward relaxation, we included a new section in the Supplementary Section 9 (rate equation model). Here we introduce a new schematic (Fig. S9) representing the four-level scheme and the respective populations for each level in 3 different cases: at thermal equilibrium (Fig. S9a), driving the system into resonance with \( f_3^{Er} \) excluding (Fig. S9b) and including relaxations (Fig. S9c). By driving \( f_3^{Er} \) at saturation, we can equalize the populations of the \( |↓↓⟩ \) and \( |↓↑⟩ \) state, which drives the system far from the thermal equilibrium state (Fig. S9b). Given that the spin relaxation of Er is much slower than the one of Ti, the system tends to relax through the latter (dotted purple arrows in Fig. S9b). Consequently the relaxation path starts from transferring the excess of population from \( |↑↓⟩ \) to \( |↓↓⟩ \), which is then transferred to \( |↓↑⟩ \) by the excitation \( f_3^{Er} \), and finally towards \( |↓↑⟩ \) by the Ti upward relaxation mechanism, with the required energy transferred from the environment.
+
+![Four-level schemes showing populations and relaxation paths for Er-Ti dimer under different conditions](page_355_682_1187_312.png)
+
+Figure S9 | Populations of the Er-Ti dimer calculated with the rate equation model at different conditions. a, Four-level scheme and respective populations for each level at thermal equilibrium. b, Driving the system into resonance at \( f_3^{Er} \) (shown as solid red arrow) and prior to activating the relaxation terms described in the model. The ESR driving leads to an equalization of \( n_{00} \) and \( n_{01} \), which makes the Ti populations far from the Boltzmann distribution. Dashed arrows represent the relaxation path realized by the Ti relaxation mechanism, where part of the population of \( n_{10} \) is transferred to \( n_{00} \) through downward relaxation mechanism. The rf excitation re-equilibrate the excess of population towards \( n_{01} \), which is further transferred to \( n_{11} \) via upward Ti relaxation mechanisms. c, Resulting populations after the inclusion of the relaxation terms, with the Er relaxation shown as dashed yellow arrows and the Ti relaxation as pink solid arrows.
+
+We additionally revised the manuscript at page 9, which now reads "These relaxations happen due to an exchange of energy with the environment which tends to relax the populations towards the thermal equilibrium. An exchange of energy to (from) the environment leads to a transition to a lower (higher) energy level."
+
+8. Line 167: What are the significant digits and error bars of both of these g-factors? How were they determined? And as for the latter one, wasn’t it reported to be anisotropic?
+Reply: The reviewer is correct. The g-factor for Ti is known to be anisotropic. Using the anisotropic g-tensor given in ref. [29], we calculated the g-factor projected along the applied magnetic field. Since the g-tensor in ref. [29] contains experimental errors, we gave the error values considering the error propagation. In contrast, the \( g_{Er} = 1.2 \) is the Er Landé g-factor calculated from its atomic quantum numbers, which gives no error bar. To clarify this, we revised the manuscript by mentioning the projection direction of g-value, error bar for Ti, and how we obtained the Er g-factor (page 4 and 7).
+
+9. When reading lines 149/150 and 200, it is not fully clear to me how the peaks/dips behave for the ESR signal of the Er. Are the Ti and Er peaks always opposite (peak vs dip)? Or is the Er signal tip dependent and can be both, either a peak or a dip?
+
+Reply: Both ESR signals are influenced by the STM tip, but in a different way. In our experiment, we observed only positive ESR signals for Ti. However, we observed that the polarity of Er ESR signals differs depending on the STM tip. For the Er, the ESR signal is detected through the Ti and, thus, we have to consider the spin relaxation processes of both Ti and Er and the spin pumping of Ti due to the spin-polarized electrons. In general, the ESR signal is obtained by modulating the RF signals and detecting the change in conductance. While there is no change in the junction conductance at the off-resonance frequency, the change occurs once the system is on-resonance. This can be expressed as given in Eq. S5:
+
+\[
+\Delta I = C \cdot \{[(n_{10} + n_{11}) - (n_{00} + n_{01})] - [(n_{10} + n_{11}) - (n_{00} + n_{01})]_{undriven}\}
+\]
+
+(S5)
+
+In our experiment, the STM tip is located over the Ti atom in the Ti-Er pair. For the Ti ESR signal, the polarity is determined by the relative spin polarization between Ti and the magnetic tip. The latter depends on the tip while the former stays almost the same at given experiment conditions. At tunnel currents with several tens of pA, this spin pumping does not play the major role for Ti ESR, since this process is much weaker and slower than the ESR-driving. However, for the indirect sensing, this spin pumping occurs in a similar time scale to the spin relaxation process and thus plays an important role in determining the ESR signal intensity and polarity.
+
+To clarify this point, we specified in the main text that "...while the Ti transitions always yield positive peaks \( f_{1,2}^{Ti} \), Er ESR signals differ depending on specific tip conditions, i.e., different tips show positive or negative sign for \( f_{3,4}^{Er} \)." In addition, we modified Supplementary Section 9 to include Eq. S5 and the related discussion.
+
+10. Line 219: Should the f4Ti not be f4Er?
+
+Reply: The reviewer is correct. We appreciate the reviewer's careful reading. We corrected it accordingly.
+
+Reviewer #3 (Remarks to the Author):
+
+The authors present in this work an experimental approach to drive electron-spin resonance (ESR) on Er atoms (lanthanide atoms with 4f electrons) and find relatively long spin relaxation times, as compared with atoms of transition elements. The approach is based on the coupling of Er atoms with Ti atoms, which act as detectors.
+
+The main conclusion of this work is of great importance for the ESR community and widens the possibilities to use single atoms to do electrically driven quantum coherent
+control of single spins on surfaces. The work is scientifically sound, the manuscript is well written, and the main conclusions are clearly stated. So, in this sense, I am certainly inclined to recommend the publication of this manuscript provided several issues are clarified or better described in a revised version. Below, I proceed to list those issues.
+
+1) For didactic reasons, and as a reference, it would have been good to add in the main manuscript ESR spectra of isolated Ti atoms. The authors could at least mention in a revised version that those spectra are reported in Fig. S3 in the SI. In general, I think that the authors could do a better job in the manuscript specifying where to find the information in the SI.
+
+Reply: We are grateful to the Reviewer for supporting the publication of our paper. We agree that the manuscript can be improved by better referencing the SI content in the main text. Since the spectrum given in Fig. 1c is quite comparable to the single Ti ESR spectrum, we preferred not to add it to the main text. Instead, in the revised manuscript at page 6 we provide a more precise referencing to Fig. S3 as follows: “However, despite using a tip showing ESR signal on an isolated Ti atom (Fig. S3a), we observed no ESR when positioning the tip over an Er atom...” In addition, we further refer Fig. S5 in the main text (page 6) with the following sentence: “These transitions are not observed for all Er atoms, possibly due to the presence of isotopes with large nuclear spins for which the intensity of the ESR signal is spread over multiple peaks and is below the sensitivity of our measurements (Fig. S5).”
+
+2) As the authors explained in the manuscript, no direct ESR signal was possible to acquire when they placed the SP-STM tip on top of the Er atoms. They also speculate about the reason for that. Maybe they can elaborate a bit more using the information provided in Fig. S1, namely it would be interesting to know if the differential conductance spectra in the absence of microwaves already hint at the absence of an ESR signal when the microwaves are applied.
+
+Reply: The reviewer is correct. However, the absence of spectral features is not directly related to the absence of the ESR signal from Er. For example, the Ho on MgO shows no features in dl/dV spectra, but the spin switching of 4f electrons of Ho can be detected in the tunnel current. In fact, the absence of spectral features indicates no unpaired electrons in 6s/5d orbitals and therefore a weak coupling between the tunneling electrons and the 4f orbitals. To link the featureless spectrum of Er and follow the hint suggested by the reviewer, we revised the manuscript (page 6): "The weak polarization of the outer shell is reflected in the absence of spin excitations in the dl/dV spectra, as reported in Fig. S1d,e,h. These factors ..."
+
+3) From the text in the manuscript, it is a little bit unclear what the authors mean by a term originating from the spin-polarized current in the context of the rate equation model summarized in Fig. 3c. I know that they elaborate on this in section 7 of the SI, but readers would benefit from a more detailed/didactic explanation of the meaning of this spin pumping.
+
+Reply: As the reviewer suggested, we included a more detailed description about the spin pumping at page 9 of the revised manuscript, which reads: "In addition, to account for the tip-dependent sign and intensity of Er ESR signals, we included a spin pumping term originating from the spin-polarized tunnel current (Fig. 3c for a negatively polarized tip). In inelastic scattering events, the exchange of angular momenta occurs while retaining the total angular momentum of the system. That is, the spin-polarized tunneling electrons lead to scattering events with preferential polarization, as depicted in the inset of Fig. 3a and
+Fig. 3c. Thus, the tunneling electrons can shift the Ti spin occupation altering the population balance with respect to the thermal equilibrium (see Supplementary Section 9).”
+
+4) Related to the previous issue, in the discussion of the rate equation model (section 7 of the SI), it is a bit unclear to me how the solution of the rate equation is finally converted into the ESR signal (\Delta I in Fig. S7). Could the authors elaborate more in the SI about this issue?
+
+Reply: As the reviewer suggested, we added an additional section in the Supplementary Section 9 explaining how we calculated the ESR intensity from the rate equation model. The newly added equations (Eq. S5 and S6) show how the populations of each level are converted to the ESR signals.
+
+5) The authors measure the relaxation time in the Er atoms with a sophisticated protocol that I must admit I have not quite understood. Could the authors make this discussion at the beginning of section “Relaxation Time Measurement ...” somehow more accessible to non-specialists?
+
+Reply: Following the reviewer's suggestion, we added detailed descriptions at the beginning of the section at page 11 which reads: "As previously discussed about the Er-Ti dimer with 0.928 nm separation, the relative peak intensity of the Ti peaks \( f_1^{Ti} \) and \( f_2^{Ti} \) reflects the time averaged population of the Er states (Fig. 4a), i.e. changes in the time-averaged spin states of Er will induce the change of the relative peak intensity of \( f_1^{Ti} \) and \( f_2^{Ti} \). Thus, by monitoring the Ti ESR signals, we can observe the evolution of the Er spin state. In this way, we characterize the characteristic relaxation time \( T_1^{Er} \) by exciting the Er into an out-of-equilibrium state and monitoring its relaxation towards the thermal state." In addition, in the following paragraph, we added one sentence, which reads: "With this scheme, it is possible to drive Er transitions and sense the change in population through the Ti ones."
+
+6) Related to the previous issue, the authors report the Er relaxation time as a single number. Should I understand that this is the intrinsic relaxation time of Er on this surface? How can one avoid the influence of the Ti atom?
+
+Reply: In general, it is difficult to access the intrinsic relaxation time of a certain material system since there are always interactions with the environment: a higher isolation of the system leads to a longer relaxation time. In our experiment, we did isolate the Er from the direct impact of the tunneling electrons, which is one of the main relaxation sources in STM experiments. However, the Er is strongly coupled to the Ti atom (about 2.7 GHz), which exposes the Er states to the Ti spin noise, as generated by the rapidly fluctuating spin under the influence of tunneling electrons. Thus, as the reviewer pointed out, the measurement result is not the intrinsic relaxation time of Er on this surface. To avoid this, one should probably build different atomic structures to reduce the spin noise. To include the reviewer's point, we revised the manuscript at page 11, which now reads: "The \( T_1^{Er} \) observed through Ti likely differs from the intrinsic relaxation time of Er on this surface due to its strong interaction with the Ti atom. Nevertheless, the large \( T_1^{Er} \)..."
+
+In addition, in the conclusion at page 13 we added a perspective sentence as follows: “We expect that, by employing a similar approach to different atomic structures, we can reduce the influence of the spin fluctuations of the atom used for the detection...”
+Reviewers' Comments:
+
+Reviewer #2:
+Remarks to the Author:
+I thank the authors for addressing all my previous concerns I had with the manuscript. The answers have improved the manuscript and help clarifying questions I had, and a potential reader might have as well. Especially the additional explanation of models, what ESR peaks to expect and the softening of some claims is appreciated.
+I only have one small remark left, that I would ask the authors to adjust.
+
+3. I am not fully in line with the added sentence in the methods about the Ti atoms. Even though the authors are using the bridge site Ti species, I find the sentence now too unprecise. I much prefer what has been written in line 71, namely that the spin magnitude is h/2 and that there is a (weak) g-factor anisotropy. The added sentence in the methods now suggests the Ti being a pure spin \( \frac{1}{2} \) atom, to which I strongly disagree. Especially since the added text does not specify the adsorption site. I’d suggest rephrasing this to be similarly as has been written in line 71 of the manuscript.
+
+Reviewer #3:
+Remarks to the Author:
+The authors have certainly made an effort to address all the issues that I raised in my first report and, in particular, I am satisfied with the answers and the changes introduced accordingly both in the manuscript and in the supplementary information. So, in this sense, I reiterate here my support for the publication of the this manuscript, which is going to be welcome by the ESR-STM community.
+Reply to Reviewers
+
+Reviewer #2 (Remarks to the Author):
+
+I thank the authors for addressing all my previous concerns I had with the manuscript. The answers have improved the manuscript and help clarifying questions I had, and a potential reader might have as well. Especially the additional explanation of models, what ESR peaks to expect and the softening of some claims is appreciated.
+I only have one small remark left, that I would ask the authors to adjust.
+
+3. I am not fully in line with the added sentence in the methods about the Ti atoms. Even though the authors are using the bridge site Ti species, I find the sentence now too unprecise. I much prefer what has been written in line 71, namely that the spin magnitude is h/2 and that there is a (weak) g-factor anisotropy. The added sentence in the methods now suggests the Ti being a pure spin 1/2 atom, to which I strongly disagree. Especially since the added text does not specify the adsorption site. I’d suggest rephrasing this to be similarly as has been written in line 71 of the manuscript.
+
+Reply: We thank the reviewer for his/her contribution in improving the manuscript. We modified the sentence in the methods to make it more precise. The sentence now reads: “As described in previous works, the Ti atoms on 2 ML MgO/Ag(100) show a spin magnitude of h/2 with g-factor anisotropy29, a behavior previously attributed to hydrogenation31.”.
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have certainly made an effort to address all the issues that I raised in my first report and, in particular, I am satisfied with the answers and the changes introduced accordingly both in the manuscript and in the supplementary information. So, in this sense, I reiterate here my support for the publication of the this manuscript, which is going to be welcome by the ESR-STM community.
+Reply: We thank the reviewer for her/his valuable comments and for supporting the publication.

0ae031331c2d266697df1ca15bce36f1cb2333132590b2a9b295f659d1dffe9c/peer_review/peer_review.md

@@ -0,0 +1,562 @@
+Peer Review File
+
+Long-term durability of metastable β-Fe2O3 photoanodes in highly corrosive seawater
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The article “Ultradurability of metastable β-Fe2O3 photoanodes in highly corrosive seawater” reports an important effect of tin when used to dop β-Fe2O3 photoanodes, rendering them more stable and energy efficient – higher current density.
+The manuscript is adequate for being published at Nature Communications after addressing the comments indicated below.
+
+Some comments:
+Line 42 – “a theoretical solar-to-hydrogen efficiency of 20.9%” – what theory do the authors refer to? This expression is often used but it is also often meaningless, since it is not the thermodynamic efficiency. Please review and correct it.
+
+Line 43 – please review references 6 and 7 since they seem to be not appropriate;
+
+Lines 68-70 – “The saturated photocurrent density of the 2% Sn/β-Fe2O3 photoanodes reaches 2.21 mA cm–2 at 1.6 VRHE, which is 8.5 times that of β-Fe2O3 photoanodes (Supplementary Fig. 2a).” – normally the current density is obtained either at the on-set of the dark current (which is not indicated) or at the thermodynamic electrolysis potential – 1.48 V @ 25 °C.
+
+Lines 71-73 – “…while the PEC performance improvement is partly due to the increased carrier concentration (Supplementary Fig. 3)” – please clarify this statement.
+
+Figure S2 – where is the on-set of the dark current? Why the current density increases from 4 -> 1 -> 3 -> 2 and the impedance goes from 2 -> 1 -> 3 -> 4?
+The onset of the photocurrent is quite shifted to the higher potentials, at ca. 1.15 VRHE. This means that the generated photopower is quite small. Please comment on that. Compared with for example http://dx.doi.org/10.1016/j.nanoen.2017.05.051, where the onset is at ca. 0.5 VRHE.
+
+Lines 181-182 – “The advantages of high valence cation-doped Sn dispersed in the bulk phase in increasing the carrier concentration and conductivity can also be fully demonstrated.” – the oxidation state of the tin can be +4 or -4; which carrier you mean? electrons or holes. Please elaborate more the discussion.
+
+Reviewer #2 (Remarks to the Author):
+
+Changhao Liu et al's manuscript presents a photoanode that is Sn doped β-Fe2O3 and is able to perform photoelectrocatalytic seawater splitting for 1440 hours without significant instability issues. The stability of PEC catalysts in seawater is a crucial problem that presents a significant challenge to realizing their practical application. Even though the activity is not exceptional, the authors' work highlights the potential of low-cost materials like Fe2O3 in achieving long-term stable PEC seawater splitting, which is a promising development. Additionally, the manuscript presents strong evidence of the stable OER of Sn/β-Fe2O3 through the use of X-ray absorption near-edge structure and time-of-flight secondary ion mass spectrometry. However, before it can be published in Nature Communications, a few questions must be addressed.
+
+1. The authors are encouraged to provide an explanation for the over 20% increase in current observed in Fig. 2d during the first 700 hours.
+2. It is necessary for the authors to clarify the experimental conditions under which the ICP samples were collected, which can be found in Supplementary Table 2.
+
+3. The Raman spectrum presented in Fig. 3d may be perplexing to readers, and it would be helpful if the authors explained the reason for annealing the samples and why they preferred to use Raman instead of XRD to characterize the materials, which may produce clearer results. Additionally, it would be interesting to know if the samples were adequately washed before characterization, as even a small amount of residual seawater could contain a significant amount of NaCl.
+
+4. In line 116, Fig. 3g should be corrected to Fig. 3c, while Fig. 3c in line 120 should be corrected to Fig. 3d.
+
+5. It would be beneficial if the authors could measure the conductivities of β-Fe2O3 and Sn/β-Fe2O3, as they claim that the addition of Sn can enhance the conductivity of β-Fe2O3.
+
+6. Deeper XPS analysis for Sn should be provided, as the authors suggest that uniformly dispersed Sn in the bulk phase can help stabilize β-Fe2O3. However, since the thickness of the Sn/β-Fe2O3 layer is over 200 nm, the 10 nm measurement is likely surface-based.
+
+Reviewer #3 (Remarks to the Author):
+
+Photoelectrochemical (PEC) water splitting using Earth-abundant seawater and sunlight is a promising way to produce green hydrogen on a large scale. However, this is still challenging due to the high corrosiveness of seawater, particularly the presence of high-concentration Cl- in seawater. In this paper, Liu and co-workers report a Sn-doped β-Fe2O3 photoanode with exceptional stability for PEC seawater oxidation. Sn dopant was found to enhance the metal-oxygen bonding energy in β-Fe2O3 and hinder the transfer of protons to the lattice oxygen, thereby inhibiting excessive surface hydration and Cl- coordination. As a result, a record durability of 1440 h was achieved with the Sn/β-Fe2O3 photoanode without any surface modification.
+
+Overall, the results presented here are intriguing and the manuscript is concisely and clearly written. I have only some minor questions regarding this paper:
+
+1) As pointed out by the authors, the Sn dopants also slowly leached out of the β-Fe2O3 photoanode due to lattice reconstruction after long-term operation (1000 h). Could the author comment on if the self-healing or dynamic stability concept (as proposed in Refs. 20-21 for Fe-based OER catalysts) can be brought to this system to stabilize the Sn dopant? Can this lead to the ultimate stability of the β-Fe2O3 photoanode?
+2) Although the size (1 × 1 cm) of the photoanode is reported in the Methods, it would be more straightforward to report photocurrent density instead of photocurrent in Figure 2a and 2d.
+3) If possible, the current-potential curves of the photoanodes after the stability tests are suggested to be provided, as they provide meaningful information on the change of the onset potentials after the stability test.
+4) Line 43, Refs 6 and 7 are not about β-Fe2O3, please double-check the accuracy of the reference list.
+Point-by-point responses for Nature Communications manuscript
+
+(ID: NCOMMS-22-49861-T)
+
+Manuscript Type: Article
+
+Title: Ultradurability of metastable \( \beta \)-Fe\(_2\)O\(_3\) photoanodes in highly corrosive seawater.
+
+Author(s): Changhao Liu, Ningsi Zhang, Yang Li, Rongli Fan, Wenjing Wang, Jianyong Feng, Chen Liu, Jiaou Wang, Weichang Hao, Zhaosheng Li, Zhigang Zou
+
+General response:
+We sincerely thank the editor, editorial staff and all reviewers for their critical comments that we have based on to improve the quality of our manuscript. The manuscript has been modified point-by-point after addressing all the suggestions as listed below.
+(Our response is given in blue and the corrections in the revised manuscript are shown in red)
+
+Point-by-point responses to Reviewer(s)
+
+Reviewer #1:
+
+The article “Ultra-durability of metastable \( \beta \)-Fe\(_2\)O\(_3\) photoanodes in highly corrosive seawater” reports an important effect of tin when used to dop \( \beta \)-Fe\(_2\)O\(_3\) photoanodes, rendering them more stable and energy efficient – higher current density.
+The manuscript is adequate for being published at *Nature Communications* after addressing the comments indicated below.
+
+Response:
+We are very grateful to the reviewer. The review comments are of great significance to further improve the manuscript. We have addressed the comments point-by-point and made the corresponding changes accordingly in the revised manuscript.
+
+Some comments:
+1. Line 42 – ‘a theoretical solar-to-hydrogen efficiency of 20.9%’ – what theory do the authors refer to? This expression is often used but it is also often meaningless, since it is not the thermodynamic efficiency. Please review and correct it.
+
+Response:
+We thank the reviewer for the insightful comments. The statement here is not proper. The theoretical
+solar-to-hydrogen efficiency of 20.9% described here is simply calculated based on the optical absorption range corresponding to the band gap of 1.9 eV of \( \beta \)-Fe\(_2\)O\(_3\), combined with the AM 1.5 G solar spectrum\(^{r1}\). However, a series of factors, such as specific band structure, transition model of electrons, carrier relaxation and recombination, optical absorption loss, and resistance loss, are not considered here. These processes will inevitably occur and have a great impact. Our original intention here is only to emphasize that \( \beta \)-Fe\(_2\)O\(_3\) has a narrower band gap and a wider optical absorption range than \( \alpha \)-Fe\(_2\)O\(_3\).
+
+[r1] Li, Z. S. et al. Photoelectrochemical cells for solar hydrogen production: current state of promising photoelectrodes, methods to improve their properties, and outlook. Energy Environ. Sci. **6**, 347–370, (2013).
+
+**Therefore, the corresponding description in the manuscript is modified as follows.**
+Lines 41–45 (Original manuscript lines 41–43):
+Recently, \( \beta \)-Fe\(_2\)O\(_3\) entered our research field as a metastable phase of iron oxide. Because of its narrower band gap (1.9 eV) compared with \( \alpha \)-Fe\(_2\)O\(_3\) (2.1 eV), the theoretical optical absorption band edge can be extended to approximately 650 nm. Thus, it has a higher theoretical solar-to-hydrogen efficiency than \( \alpha \)-Fe\(_2\)O\(_3\). At the same time, \( \beta \)-Fe\(_2\)O\(_3\) also shows good stability in photoelectrochemical alkaline water splitting\(^{20,21}\).
+
+2. Line 43 – please review references 6 and 7 since they seem to be not appropriate;
+
+**Response:**
+Thanks for the reviewer’s comment. Here is a citation error. We have added the correct and corresponding references here, and other corresponding reference numbers have also been changed.
+
+**References**
+Lines 325–329 (Original manuscript lines 298):
+[20] Zhang, N. S. et al. Paving the road toward the use of \( \beta \)-Fe\(_2\)O\(_3\) in solar water splitting: Raman identification, phase transformation and strategies for phase stabilization. *Natl. Sci. Rev.* **7**, 1059–1067, (2020).
+[21] Li, Y. et al. Metastable-phase \( \beta \)-Fe\(_2\)O\(_3\) photoanodes for solar water splitting with durability exceeding 100 h. *Chinese J. Catal.* **42**, 1992–1998, (2021).
+
+3. Lines 68-70 – “The saturated photocurrent density of the 2% Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes reaches 2.21 mA cm\(^{-2}\) at 1.6 V\(_{RHE}\), which is 8.5 times that of \( \beta \)-Fe\(_2\)O\(_3\) photoanodes (Supplementary Fig. 2a).” – normally the current density is obtained either at the on-set of the dark current (which is not indicated) or at the thermodynamic electrolysis potential – 1.48 V @ 25 °C.
+
+**Response:**
+We are greatly grateful to the reviewer for the nice question, which can help us revise the discussion
+in the manuscript more accurately. The photoelectric current at 1.6 \( V_{\text{RHE}} \) is selected for comparison, considering the following two factors. i. The photocurrent of the common \( \alpha \)-Fe\(_2\)O\(_3\) photoanode has reached saturation at 1.6 \( V_{\text{RHE}} \), so we choose this potential to compare. Since photogenerated carriers have reached saturation, the photocurrent will not increase with the increase of bias voltage. ii. In the supplementary dark current test, it is found that the on-set of the dark current of \( \beta \)-Fe\(_2\)O\(_3\) photoanodes is usually at 1.7–1.8 \( V_{\text{RHE}} \).
+
+![Dark current density of 0% to 4% Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes.](page_355_393_484_324.png)
+
+Fig. R1. Dark current density of 0% to 4% Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes.
+
+During the long-term stability test, we found that the on-set of the dark current would shift negatively due to the formation of FeOOH on the surface in situ as electrocatalysts. However, the on-set of the dark current is always after 1.6 \( V_{\text{RHE}} \), so it can be determined that the current at 1.6 \( V_{\text{RHE}} \) is contributed by the photocurrent. Considering all \( \beta \)-Fe\(_2\)O\(_3\) with different Sn concentrations and the photoanodes before and after the stability tests, we choose the photocurrent density at 1.6 \( V_{\text{RHE}} \) for comparison. And the subsequent stability test is also conducted under this bias voltage.
+
+We modified Supplementary Fig. 2 and added the dark current of 0% to 4% Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes as follows.
+
+![Photoelectrochemical tests of \( \beta \)-Fe\(_2\)O\(_3\) with different Sn concentrations.](page_355_1042_484_324.png)
+
+Supplementary Fig. 2. Photoelectrochemical tests of \( \beta \)-Fe\(_2\)O\(_3\) with different Sn concentrations.
+a, Photocurrent density of 0% to 4% Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes at 0.6–1.7 \( V_{\text{RHE}} \) in 1 M KOH + 0.5 M NaCl simulated seawater under one sun illumination. b, Dark current density of 0% to 4% Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes.
+
+The corresponding description in the manuscript is modified as follows.
+Lines 70–72 (Original manuscript lines 68–70):
+The photocurrent density of the 2% Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes reaches 2.21 mA cm\(^{-2}\) at 1.6 \( V_{\text{RHE}} \).
+which is 8.5 times that of \( \beta\text{-Fe}_2\text{O}_3 \) photoanodes (Supplementary Fig. 2).
+
+4. Lines 71-73 – “…while the PEC performance improvement is partly due to the increased carrier concentration (Supplementary Fig. 3)” – please clarify this statement.
+
+Response:
+Thank the reviewer for the constructive suggestions. In the original manuscript, we do not have a detailed description and analysis of the increased carrier concentration in Sn/\( \beta\text{-Fe}_2\text{O}_3 \). The carrier concentrations in \( \beta\text{-Fe}_2\text{O}_3 \) and Sn/\( \beta\text{-Fe}_2\text{O}_3 \) can be calculated by the slope of the tangent of the M-S curve. With the increase of carrier concentration, the conductivity of Sn/\( \beta\text{-Fe}_2\text{O}_3 \) is also improved, which is consistent with the AC electrochemical impedance spectra. We have supplemented the detailed discussion in Supplementary Fig. 4 and revised our manuscript.
+We have added relevant discussion in Supplementary Fig. 4.
+
+![Mott-Schottky plots and AC impedance spectra of β-Fe2O3 and Sn/β-Fe2O3 photoanodes](page_362_670_1047_393.png)
+
+Supplementary Fig. 4. Carrier concentration and AC impedance of \( \beta\text{-Fe}_2\text{O}_3 \) and Sn/\( \beta\text{-Fe}_2\text{O}_3 \).
+a, Mott-Schottky plots of \( \beta\text{-Fe}_2\text{O}_3 \) and Sn/\( \beta\text{-Fe}_2\text{O}_3 \) photoanodes measured at 1.6 V_{RHE}. b, High-frequency part of AC electrochemical impedance spectra of \( \beta\text{-Fe}_2\text{O}_3 \) and Sn/\( \beta\text{-Fe}_2\text{O}_3 \) photoanodes.
+We used Mott–Schottky relationship to determine the donor concentration (\( N_D \)):
+
+\[
+\frac{1}{C_{SC}^2} = \frac{2}{q \varepsilon \varepsilon_0 N_D} \left( V - V_{fb} - \frac{kT}{q} \right)
+\]
+
+where \( C_{SC} \) is the space charge capacitance, \( q \) is the elementary charge, \( \varepsilon_0 \) is the permittivity of free space, and \( \varepsilon \) is the dielectric constant of \( \beta\text{-Fe}_2\text{O}_3 \)^{37}. The slope of the tangent in Supplementary Fig. 4a is inversely proportional to the carrier concentration:
+
+\[
+Slope = \frac{2}{\varepsilon \varepsilon_0 N_D}
+\]
+
+where \( e \) is the electron charge. Thus, it can be estimated that the carrier (electrons) concentration of Sn/\( \beta\text{-Fe}_2\text{O}_3 \) is 8.4 times that of \( \beta\text{-Fe}_2\text{O}_3 \).
+In Supplementary Fig. 4b, the half-circle fitted by the AC electrochemical impedance spectra in the high frequency region is related to R3/CPE2 in the equivalent circuit model, which is assigned to the electron transport inside the electrode. The R3 values of \( \beta\text{-Fe}_2\text{O}_3 \) and Sn/\( \beta\text{-Fe}_2\text{O}_3 \) calculated
+from the fitting results are 1065.0 \( \Omega \) and 299.8 \( \Omega \) respectively, which means that the conductivity of Sn/\( \beta \)-Fe\(_2\)O\(_3\) is much higher than that of \( \beta \)-Fe\(_2\)O\(_3\).
+
+References
+Lines 369–371 (Original manuscript lines 343):
+[37] Franking, R. et al. Facile post-growth doping of nanostructured hematite photoanodes for enhanced photoelectrochemical water oxidation. Energy Environ. Sci. 6, 500–512, (2013).
+
+The corresponding description in the manuscript is modified as follows.
+Lines 72–80 (Original manuscript lines 70–74):
+The Sn dopants do not affect the light absorption of the \( \beta \)-Fe\(_2\)O\(_3\) photoanodes, while the PEC performance improvement is partly due to the increased electron concentration caused by the doping of high-valence cation Sn\(^{4+}\). The carrier concentration of Sn/\( \beta \)-Fe\(_2\)O\(_3\) is approximately 8.4 times that of\( \beta \)-Fe\(_2\)O\(_3\), and correspondingly, its bulk conductivity is also improved, according to Mott-Schottky plots and AC electrochemical impedance spectra in the low-frequency region. Sn can simultaneously adjust the chemical field at the semiconductor/electrolyte interface, which significantly reduces the AC impedance of the interface transfer kinetics (Supplementary Fig. 3 and 4).
+
+5. Figure S2 – where is the on-set of the dark current? Why the current density increases from 4 -> 1 -> 3 -> 2 and the impedance goes from 2 -> 1 -> 3 -> 4?
+The onset of the photocurrent is quite shifted to the higher potentials, at ca. 1.15 V\(_{RHE}\). This means that the generated photo-power is quite small. Please comment on that. Compared with for example http://dx.doi.org/10.1016/j.nanoen.2017.05.051, where the onset is at ca. 0.5 V\(_{RHE}\).
+
+Response:
+Thanks for the reviewer’s valuable comments.
+
+![Dark current density vs Potential graph showing curves for β-Fe2O3 and Sn-doped β-Fe2O3 photoanodes](page_1012_1042_388_246.png)
+
+Fig. R2. Dark current density of 0% to 4% Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes.
+The on-set of the dark current is between 1.7~1.8 V\(_{RHE}\). It is related to the electrocatalytic properties of the surface of photoanodes. The AC electrochemical impedance spectra in the low frequency region in Supplementary Fig. 3b can reflect the transfer kinetics at the interface, which has a strong correlation with the change of doping concentration and the on-set of the dark current.
+We modified Supplementary Fig. 2 and added the dark current of 0% to 4% Sn/β-Fe2O3 photoanodes as follows.
+
+![Photoelectrochemical tests of β-Fe2O3 with different Sn concentrations.](page_186_153_1077_312.png)
+
+Supplementary Fig. 2. Photoelectrochemical tests of β-Fe2O3 with different Sn concentrations.
+a, Photocurrent density of 0% to 4% Sn/β-Fe2O3 photoanodes at 0.6–1.7 V_{RHE} in 1 M KOH + 0.5 M NaCl simulated seawater under one sun illumination. b, Dark current density of 0% to 4% Sn/β-Fe2O3 photoanodes.
+
+In addition to interface transfer kinetics, the photocurrent density is also related to the separation and transmission of carriers in the bulk. The low-frequency impedance shows a trend of 2 ->1 ->3 ->4. This is because the addition of a small amount of Sn improves the electronic environment of the surface and is conducive to charge transfer and matter exchange. When a large amount of Sn accumulates on the surface, Fe as the reactive active center was blocked, so the impedance increased. Another important factor affecting the photocurrent density is the conductivity of the semiconductor. The Fermi level of the semiconductor will be closer to the conduction band with the increase of doping concentration, and the carrier concentration and conductivity increased. The interface reaction rate and the bulk transfer of carriers jointly determine that the photocurrent density goes from 4 -> 1 -> 3 -> 2.
+
+The photocurrent on-set potential of β-Fe2O3 is quite high here, which is due to the absence of any electrocatalyst or passivation layer on the photoanodes surface. A large number of surface trapped states will have serious effects. But in this work, we only discuss the activity and stability of bare β-Fe2O3 photoanodes. Besides, the adding of Sn will bring a negative effect, that is, it will introduce more defect levels in the band gap. Photogenerated electrons and holes are trapped in the defect levels, and the charge recombination will also be more serious. More surface trapped states will also bind the carriers, resulting in the reduction of the photovoltage. For this reason, the on-set potential will shift to the higher potential with the increase of Sn concentration. Besides, the steady-state photovoltage test is carried out (Fig. R3). The photovoltage of Sn-doped photoanode reduced, which means that the generated photo-power became smaller.
+Fig. R3. Steady-state photovoltage of \( \beta\text{-Fe}_2\text{O}_3 \) and Sn/\( \beta\text{-Fe}_2\text{O}_3 \) photoanodes.
+
+6. Lines 181-182 – “The advantages of high valence cation-doped Sn dispersed in the bulk phase in increasing the carrier concentration and conductivity can also be fully demonstrated.” – the oxidation state of the tin can be +4 or -4; which carrier you mean? electrons or holes. Please elaborate more the discussion.
+
+Response:
+Thank you for raising this valuable question. The Sn mentioned here is a +4-valence cation, and the high valence cation doping can provide more electrons as a donor. For n-type semiconductors, more cation doping means that the Fermi level is closer to the conduction band, and the conductivity will be improved.
+
+The corresponding description in the manuscript is modified as follows, and the role of high-valent cations is discussed more.
+Lines 72–74 (Original manuscript lines 70–72):
+The Sn dopants do not affect the light absorption of the \( \beta\text{-Fe}_2\text{O}_3 \) photoanodes, while the PEC performance improvement is partly due to the increased electron concentration caused by the doping of high-valence cation Sn^{4+}.
+Lines 189–190 (Original manuscript lines 180–182):
+The advantages of donor Sn^{4+} dispersed in bulk phase in improving electron concentration and conductivity can also be fully demonstrated.
+Reviewer #2:
+
+Changhao Liu et al's manuscript presents a photoanode that is Sn doped \( \beta \)-Fe\(_2\)O\(_3\) and is able to perform photoelectrocatalytic seawater splitting for 1440 hours without significant instability issues. The stability of PEC catalysts in seawater is a crucial problem that presents a significant challenge to realizing their practical application. Even though the activity is not exceptional, the authors' work highlights the potential of low-cost materials like Fe\(_2\)O\(_3\) in achieving long-term stable PEC seawater splitting, which is a promising development. Additionally, the manuscript presents strong evidence of the stable OER of Sn/\( \beta \)-Fe\(_2\)O\(_3\) through the use of X-ray absorption near-edge structure and time-of-flight secondary ion mass spectrometry. However, before it can be published in *Nature Communications*, a few questions must be addressed.
+
+Response:
+We sincerely thank the reviewer’s comments. The proposed suggestions are valuable and helpful for improving our work We have carefully revised the manuscript and replied to the comments point-by-point shown below.
+
+1. The authors are encouraged to provide an explanation for the over 20% increase in current observed in Fig. 2d during the first 700 hours.
+
+Response:
+We thank the reviewer for the insightful comments. In the whole process of the stability test, the change of photocurrent density shows a trend of slowly increasing at first, then gradually decreasing, and finally stabilizing. This is because at the initial stage of the reaction, \( \beta \)-Fe\(_2\)O\(_3\) on the surface is converted into FeOOH, which XPS, Raman and other experiments can also prove (Fig. 3). FeOOH is an efficient electrocatalyst, so the photocurrent density increased at the early stage of the reaction. Because of the infiltration corrosion of Cl\(^-\) along with the excessive surface reconstruction and the loss of Sn, the photocurrent density had a downward trend after 700 h. However, due to the relatively uniform distribution of Sn in the whole film, the reconstruction and corrosion are controlled within a certain range of the Sn/\( \beta \)-Fe\(_2\)O\(_3\) surface. The evolution of the surface has reached a relatively stable state after a period of time, so the photocurrent density tends to be stable after 1000 h.
+In the past period, we have extended the stability test to 3000 h (Fig. 3d). In the later period of the stability test, the photocurrent is kept in a relatively stable range.
+
+The following is the replacement of the stability image in the manuscript, adding 3000 h stability data.
+Fig. 2. | PEC properties of the \( \beta \)-Fe\(_2\)O\(_3\) photoanode in simulated seawater.
+a, Stability test of \( \beta \)-Fe\(_2\)O\(_3\) (in 1 M KOH, 1 M KOH/0.5 M NaCl, 1 M KOH/saturated NaCl solution) and Sn/\( \beta \)-Fe\(_2\)O\(_3\) (in 1 M KOH/saturated NaCl solution) for 100 h. b, AC electrochemical impedance spectra of Sn/\( \beta \)-Fe\(_2\)O\(_3\) before and after the reaction. c, HAADF images of the Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanode after 100 h of reaction in 1 M KOH/0.5 M NaCl. d, Stability test of Sn/\( \beta \)-Fe\(_2\)O\(_3\) in 1 M KOH/0.5 M NaCl for 3000 h. e, Summary of the photoanode stability of PEC (simulated) seawater splitting over the years. Detailed information can be found in Supplementary Table 1.
+
+The following is the description added to the text.
+Lines 201–217 (Original manuscript lines 192):
+Such characteristics make the stability rise continuously at first, then decline slowly, and finally maintain a stable range of fluctuations. At the initial stage of the reaction, the \( \beta \)-Fe\(_2\)O\(_3\) on the surface is converted into FeOOH in situ (Fig. 3). FeOOH itself is an efficient electrocatalyst, so the photocurrent increased. Because of the infiltration corrosion of Cl\(^-\) along with the excessive surface reconstruction and the loss of Sn, the photocurrent density had a downwards trend after 700 h. It can be analyzed from Supplementary Fig. 10 that due to the electrocatalytic effect of surface FeOOH after 1000-h reaction, the on-set potential moved to the negative direction by approximately 0.1 \( V_{RHE} \). However, the photocurrent density at 1.6 \( V_{RHE} \) was reduced with the influence of surface reconstruction and corrosion. Owing to the relatively uniform distribution of Sn in the whole film, reconstruction and corrosion were controlled within a certain range of \( \beta \)-Fe\(_2\)O\(_3\) surface, instead of continuing to the deep layer. In addition, while the surface reconstruction was accompanied by the loss of Sn, there would also be a sedimentation equilibrium on the surface, and deposition would occur, which forms a dynamic balance of corrosion, metal loss, deposition and protection. A dynamic stable state of surface evolution was established after a period of time, so the photocurrent density tended to be stable after 1000 h.
+2. It is necessary for the authors to clarify the experimental conditions under which the ICP samples were collected, which can be found in Supplementary Table 2.
+
+Response: We appreciate the reviewer’s comments very much. We have supplemented the relevant information of the ICP test in the method and Supplementary Materials.
+
+The instrument information of ICP is supplemented in the Methods section.
+Lines 437–438 (Original manuscript lines 402):
+The concentrations of cations in condensates were examined by ICP-OES (PerkinElmer Instruments, PTIMA 5300 DV).
+
+We have added the test conditions of ICP experiment in Supplementary Table 2 as follows.
+
+<table>
+  <tr>
+    <th>Samples</th>
+    <th>Fe ion concentration in electrolyte (mg/L)</th>
+  </tr>
+  <tr>
+    <td>\( \beta\)-Fe<sub>2</sub>O<sub>3</sub></td>
+    <td>0.071</td>
+  </tr>
+  <tr>
+    <td>Sn/\( \beta\)-Fe<sub>2</sub>O<sub>3</sub></td>
+    <td>0.036</td>
+  </tr>
+</table>
+
+Supplementary Table 2. ICP measurement of Fe ions in the electrolyte after the reaction.
+Take the alkaline simulated seawater electrolyte after 100 h of reaction, and measure the content of Fe ion in the solution. Based on this, we can analyse the loss of Fe caused by surface reconstruction in the stability test. Dilution and pH adjustment of the same multiple was conducted before the test.
+
+3. The Raman spectrum presented in Fig. 3d may be perplexing to readers, and it would be helpful if the authors explained the reason for annealing the samples and why they preferred to use Raman instead of XRD to characterize the materials, which may produce clearer results. Additionally, it would be interesting to know if the samples were adequately washed before characterization, as even a small amount of residual seawater could contain a significant amount of NaCl.
+
+Response: We sincerely thank the reviewer’s comments. The annealing treatment further confirms the hydration and reconstruction of the surface of \( \beta\)-Fe<sub>2</sub>O<sub>3</sub> after a long-term reaction in the electrolyte. \( \beta\)-Fe<sub>2</sub>O<sub>3</sub> was transformed into FeOOH during the reconstruction, and FeOOH would be transformed into stable phase \( \alpha\)-Fe<sub>2</sub>O<sub>3</sub> after annealing. Through such treatment, the process of surface reconstruction and material transformation can be more intuitively displayed.
+In our tests, we found that the signal-to-noise ratio of the XRD of the \( \beta\)-Fe<sub>2</sub>O<sub>3</sub> layer grown on the surface of the FTO conductive glass was weak. Because the reconstruction during the reaction only occurred on the surface of the film, no more valuable signal can be detected in the XRD test. The confocal Raman can detect the signal of trace substances on the sample surface, so the surface reconstruction can be well detected here.
+
+The corresponding description in the manuscript is modified as follows.
+Lines 125–133 (original manuscript lines 119–124):
+Confocal Raman spectroscopy is used to observe the speciation evolution of trace substances on the surface of photoanode before and after the reaction. After 100 h and 1000 h of seawater splitting, there are two peaks of M–OOH at approximately 470 cm\(^{-1}\) and 550 cm\(^{-1}\)\(^{24,25}\), which echo the change
+in the O 1 s XPS peak. FeOOH, which is hydrated and reconstituted during the reaction, is also more prone to dehydration and sintering at high temperatures. Thus, the \( \alpha\text{-Fe}_2\text{O}_3 \) peak can be observed when the \( \beta\text{-Fe}_2\text{O}_3 \) photoanodes are further calcined at 600\(^\circ\)C.
+
+The specific steps of cleaning are as follows: remove the photoanodes after reaction for a period of time from the electrolyte, and immediately wash them with flowing deionized water for 20 s. Make sure there is no residual electrolyte. If Cl\(^-\) or NaCl is detected later, it is considered that they are strong adsorption ions formed on the surface of \( \beta\text{-Fe}_2\text{O}_3 \), or ions participating in the surface reconstruction and entering the lattice. In addition, no NaCl signal was detected for Sn/\( \beta\text{-Fe}_2\text{O}_3 \) photoanode after 100-h reaction which was treated with the same cleaning method. This is because Cl\(^-\) in the electrolyte has not yet formed strong adsorption or participated in surface reconstruction. This comparison also shows that the cleaning of the photoanode after the reaction is effective.
+
+In the Methods section, we added relevant descriptions.
+Lines 462–464 (original manuscript lines 425):
+All photoanodes after reaction were taken out of the electrolyte and washed with flowing deionized water for 20 s to remove the residual electrolyte on the surface. Then the cleaned photoanodes were further characterized and analyzed.
+
+4. In line 116, Fig. 3g should be corrected to Fig. 3c, while Fig. 3c in line 120 should be corrected to Fig. 3d.
+
+Response: We sincerely thank the reviewer’s correction.
+
+The corresponding description in the manuscript is modified as follows.
+Lines 121–123 (original manuscript lines 115–117):
+The XPS peak of the Sn 3d signal disappeared after 1000 h of reaction (Fig. 3c), indicating that Sn is also slowly lost during lattice reconstruction by surface hydration.
+
+Lines 125–129 (original manuscript lines 119–122):
+Confocal Raman spectroscopy is used to observe the speciation evolution of trace substances on the surface of photoanode before and after reaction (Fig. 3d). After 100 h and 1000 h of seawater splitting, there are two peaks of M–OOH at approximately 470 cm\(^{-1}\) and 550 cm\(^{-1}\) 26,27, which echo the change in the O 1 s XPS peak.
+
+5. It would be beneficial if the authors could measure the conductivities of \( \beta\text{-Fe}_2\text{O}_3 \) and Sn/\( \beta\text{-Fe}_2\text{O}_3 \), as they claim that the addition of Sn can enhance the conductivity of \( \beta\text{-Fe}_2\text{O}_3 \).
+
+Response: We are very grateful to the reviewers for the suggestions. We compare the conductivities of \( \beta\text{-Fe}_2\text{O}_3 \) and Sn/\( \beta\text{-Fe}_2\text{O}_3 \) by analysing the Nyquist impedance. The AC electrochemical impedance data are fitted into an equivalent circuit model including two RC (a sub-circuit containing a resistance and a capacitance in parallel) circuits, as shown in Fig R2.
+    (1) R1 is the solution resistance.
+    (2) R2/CPE1 in the low-frequency region with the larger resistance represents interface transfer
+kinetics between n-type semiconductor electrode and electrolyte (Supplementary Fig. 3b).
+
+(3) R3/CPE2 in the high-frequency region is assigned to the electron transport inside the electrode (Supplementary Fig. 4b).
+
+By fitting the data, we obtained the corresponding parameter values of \( \beta\)-Fe$_2$O$_3$ and Sn/\( \beta\)-Fe$_2$O$_3$ photoanodes in the equivalent circuit model, and the error was less than 2%.
+
+<table>
+  <tr>
+    <th>Samples</th>
+    <th>R1 (\( \Omega \))</th>
+    <th>R2 (\( \Omega \))</th>
+    <th>R3 (\( \Omega \))</th>
+    <th>CPE1-T (F)</th>
+    <th>CPE1-P /</th>
+    <th>CPE2-T (F)</th>
+    <th>CPE2-P /</th>
+  </tr>
+  <tr>
+    <td>\( \beta\)-Fe$_2$O$_3$</td>
+    <td>17.61</td>
+    <td>148,290</td>
+    <td>1,065</td>
+    <td>1.3719\times10^{-6}</td>
+    <td>0.90126</td>
+    <td>2.3622\times10^{-6}</td>
+    <td>0.97877</td>
+  </tr>
+  <tr>
+    <td>Sn/\( \beta\)-Fe$_2$O$_3$</td>
+    <td>14.93</td>
+    <td>19,137</td>
+    <td>299.8</td>
+    <td>3.5184\times10^{-6}</td>
+    <td>0.89427</td>
+    <td>1.6896\times10^{-5}</td>
+    <td>0.89833</td>
+  </tr>
+</table>
+
+Table R1. ICP measurement of Fe ions in the electrolyte after the reaction.
+
+The R3 values of \( \beta\)-Fe$_2$O$_3$ and Sn/\( \beta\)-Fe$_2$O$_3$ calculated from the fitting results are 1065.0 \( \Omega \) and 299.8 \( \Omega \) respectively, which means that the conductivity of Sn/\( \beta\)-Fe$_2$O$_3$ is much higher than that of \( \beta\)-Fe$_2$O$_3$. Improved conductivity is due to the increase of carrier concentration in Sn/\( \beta\)-Fe$_2$O$_3$, as discussed in Supplementary Fig. 4a.
+
+\( \beta\)-Fe$_2$O$_3$ grown on the surface of conductive glass substrate has a certain orientation along the lattice direction of FTO, and its conductivity is anisotropic. The resistance perpendicular to the plane direction of the film measured by AC electrochemical impedance is more practical. At present, we cannot grow high-quality single-crystal \( \beta\)-Fe$_2$O$_3$ on other substrates and accurately measure its intrinsic conductivity.
+
+We have added relevant discussion in Supplementary Fig. 4.
+
+![Mott-Schottky plots and AC impedance spectra of β-Fe2O3 and Sn/β-Fe2O3 photoanodes](page_101_682_1242_377.png)
+
+Supplementary Fig. 4. Carrier concentration and AC impedance of \( \beta\)-Fe$_2$O$_3$ and Sn/\( \beta\)-Fe$_2$O$_3$.
+a, Mott-Schottky plots of \( \beta\)-Fe$_2$O$_3$ and Sn/\( \beta\)-Fe$_2$O$_3$ photoanodes measured at 1.6 V$_{RHE}$. b, High-frequency part of AC electrochemical impedance spectra of \( \beta\)-Fe$_2$O$_3$ and Sn/\( \beta\)-Fe$_2$O$_3$ photoanodes.
+We used Mott–Schottky relationship to determine the donor concentration (N$_D$):
+
+\[
+\frac{1}{C_{SC}^2} = \frac{2}{q \varepsilon \varepsilon_0 N_D} \left( V - V_{fb} - \frac{kT}{q} \right)
+\]
+
+where $C_{SC}$ is the space charge capacitance, $q$ is the elementary charge, $\varepsilon_0$ is the permittivity of free space, and $\varepsilon$ is the dielectric constant of \( \beta\)-Fe$_2$O$_3$[37]. The slope of the tangent in Supplementary Fig. 4a is inversely proportional to the carrier concentration:
+Slope = \frac{2}{e \varepsilon \epsilon_0 N_D}
+
+where \( e \) is the electron charge. Thus, it can be estimated that the electron concentration of Sn/\( \beta \)-Fe\(_2\)O\(_3\) is 8.4 times that of \( \beta \)-Fe\(_2\)O\(_3\).
+
+In Supplementary Fig. 4b, the half-circle fitted by the AC electrochemical impedance spectra in high-frequency region is related to R3/CPE2 in the equivalent circuit model, which is assigned to the electron transport inside the electrode. The R3 values of \( \beta \)-Fe\(_2\)O\(_3\) and Sn/\( \beta \)-Fe\(_2\)O\(_3\) calculated from the fitting results are 1065 \( \Omega \) and 299.8 \( \Omega \) respectively, which means that the conductivity of Sn/\( \beta \)-Fe\(_2\)O\(_3\) is much higher than that of \( \beta \)-Fe\(_2\)O\(_3\).
+
+References
+Lines 369–371 (Original manuscript lines 343):
+[37] Franking, R. et al. Facile post-growth doping of nanostructured hematite photoanodes for enhanced photoelectrochemical water oxidation. Energy Environ. Sci. 6, 500–512, (2013).
+
+The corresponding description of conductivity in the manuscript is modified as follows.
+Lines 74–77 (original manuscript lines 70–72):
+The electron concentration of Sn/\( \beta \)-Fe\(_2\)O\(_3\) is approximately 8.4 times that of \( \beta \)-Fe\(_2\)O\(_3\), and correspondingly, its bulk conductivity is also improved, according to Mott-Schottky plots and AC electrochemical impedance spectra at low frequency region.
+
+6. Deeper XPS analysis for Sn should be provided, as the authors suggest that uniformly dispersed Sn in the bulk phase can help stabilize \( \beta \)-Fe\(_2\)O\(_3\). However, since the thickness of the Sn/\( \beta \)-Fe\(_2\)O\(_3\) layer is over 200 nm, the 10 nm measurement is likely surface-based.
+Response: We appreciate the reviewer’s valuable comments that help to improve the quality of our work. We conducted a deeper etching XPS experiment. To reflect the dispersion of bulk Sn, we increased the etching depth to 50 nm, 100 nm and 200 nm. Compared with the thickness of the film, this detection depth can better reflect the distribution of Sn in the bulk phase.
+We added a new etching XPS experiment and replaced Supplementary Fig. 9.
+Supplementary Fig. 9. Gradually varying Sn concentration in Sn/β-Fe2O3.
+Etching XPS of Sn 3d spectra of 2% Sn/β-Fe2O3. Take the positions at depths of 0 nm, 50 nm, 100 nm and 200 nm from the surface layer for analysis.
+
+![Etching XPS of Sn 3d spectra of 2% Sn/β-Fe2O3.](page_347_286_755_410.png)
+Reviewer #3:
+
+Photoelectrochemical (PEC) water splitting using Earth-abundant seawater and sunlight is a promising way to produce green hydrogen on a large scale. However, this is still challenging due to the high corrosiveness of seawater, particularly the presence of high-concentration Cl⁻ in seawater. In this paper, Liu and co-workers report a Sn-doped \( \beta \)-Fe₂O₃ photoanode with exceptional stability for PEC seawater oxidation. Sn dopant was found to enhance the metal-oxygen bonding energy in \( \beta \)-Fe₂O₃ and hinder the transfer of protons to the lattice oxygen, thereby inhibiting excessive surface hydration and Cl⁻ coordination. As a result, a record durability of 1440 h was achieved with the Sn/\( \beta \)-Fe₂O₃ photoanode without any surface modification.
+
+Overall, the results presented here are intriguing and the manuscript is concisely and clearly written. I have only some minor questions regarding this paper:
+
+Response: We thank the reviewer for the very positive assessment of our work. The comments lead to further improvements in the quality of our work. According to the comments, we have modified our manuscript discussion and corresponding responses.
+
+1) As pointed out by the authors, the Sn dopants also slowly leached out of the \( \beta \)-Fe₂O₃ photoanode due to lattice reconstruction after long-term operation (1000 h). Could the author comment on if the self-healing or dynamic stability concept (as proposed in Refs. 20-21 for Fe-based OER catalysts) can be brought to this system to stabilize the Sn dopant? Can this lead to the ultimate stability of the \( \beta \)-Fe₂O₃ photoanode?
+
+Response: We appreciate the reviewer’s valuable comments. Long-term stability test will lead to the slow leaching of Sn in the surface layer, resulting in a decline in photocurrent density after 700 h, as well as a certain degree of surface reconstruction and Cl⁻ erosion. In addition, we extended the stability test to 3000 h. We found that the subsequent photocurrent density did not decrease significantly, but remained relatively stable after the 1000-h reaction. Owing to the relatively uniform distribution of Sn in the whole film, reconstruction and corrosion were controlled within a certain range of \( \beta \)-Fe₂O₃ surface, instead of continuing to the deep layer. In addition, while the surface reconstruction was accompanied by the loss of Sn, there would also be a sedimentation equilibrium on the surface, and deposition would occur, which forms a dynamic balance of corrosion, metal loss, deposition and protection. A dynamic stable state of surface evolution was established, as the reviewer said.
+
+The following is the replacement of the stability image in the manuscript, adding 3000 h stability data.
+Fig. 2. | PEC properties of the \( \beta \)-Fe\(_2\)O\(_3\) photoanode in simulated seawater.
+a, Stability test of \( \beta \)-Fe\(_2\)O\(_3\) (in 1 M KOH, 1 M KOH/0.5 M NaCl, 1 M KOH/saturated NaCl solution) and Sn/\( \beta \)-Fe\(_2\)O\(_3\) (in 1 M KOH/saturated NaCl solution) for 100 h. b, AC electrochemical impedance spectra of Sn/\( \beta \)-Fe\(_2\)O\(_3\) before and after the reaction. c, HAADF images of the Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanode after 100 h of reaction in 1 M KOH/0.5 M NaCl. d, Stability test of Sn/\( \beta \)-Fe\(_2\)O\(_3\) in 1 M KOH/0.5 M NaCl for 3000 h. e, Summary of the photoanode stability of PEC (simulated) seawater splitting over the years. Detailed information can be found in Supplementary Table 1.
+
+The following is the description added to the text.
+Lines 201–217 (Original manuscript lines 192):
+Such characteristics make the stability rise continuously at first, then decline slowly, and finally maintain a stable range of fluctuations. At the initial stage of the reaction, the \( \beta \)-Fe\(_2\)O\(_3\) on the surface is converted into FeOOH in situ (Fig. 3). FeOOH itself is an efficient electrocatalyst, so the photocurrent increased. Because of the infiltration corrosion of Cl\(^-\) along with the excessive surface reconstruction and the loss of Sn, the photocurrent density had a downwards trend after 700 h. It can be analyzed from Supplementary Fig. 10 that due to the electrocatalytic effect of surface FeOOH after 1000-h reaction, the on-set potential moved to the negative direction by approximately 0.1 \( V_{RHE} \). However, the photocurrent density at 1.6 \( V_{RHE} \) was reduced with the influence of surface reconstruction and corrosion. Owing to the relatively uniform distribution of Sn in the whole film, reconstruction and corrosion were controlled within a certain range of \( \beta \)-Fe\(_2\)O\(_3\) surface, instead of continuing to the deep layer. In addition, while the surface reconstruction was accompanied by the loss of Sn, there would also be a sedimentation equilibrium on the surface, and deposition would occur, which forms a dynamic balance of corrosion, metal loss, deposition and protection. A dynamic stable state of surface evolution was established after a period of time, so the photocurrent density tended to be stable after 1000 h.
+2) Although the size (1 × 1 cm) of the photoanode is reported in the Methods, it would be more straightforward to report photocurrent density instead of photocurrent in Figure 2a and 2d.
+
+Response: We appreciate the reviewer’s comments. We unify the ordinates in the figure as the current density.
+
+We have made corresponding modifications to Figures 2a and 2d.
+
+![PEC properties of the β-Fe2O3 photoanode in simulated seawater. Panels a-d show current density vs. time plots and AC impedance spectra, while panel e shows stability over years.](page_184_370_1080_627.png)
+
+Fig. 2. | PEC properties of the β-Fe2O3 photoanode in simulated seawater.
+a, Stability test of β-Fe2O3 (in 1 M KOH, 1 M KOH/0.5 M NaCl, 1 M KOH/saturated NaCl solution) and Sn/β-Fe2O3 (in 1 M KOH/saturated NaCl solution) for 100 h. b, AC electrochemical impedance spectra of Sn/β-Fe2O3 before and after the reaction. c, HAADF images of the Sn/β-Fe2O3 photoanode after 100 h of reaction in 1 M KOH/0.5 M NaCl. d, Stability test of Sn/β-Fe2O3 in 1 M KOH/0.5 M NaCl for 3000 h. e, Summary of the photoanode stability of PEC (simulated) seawater splitting over the years. Detailed information can be found in Supplementary Table 1.
+
+3) If possible, the current-potential curves of the photoanodes after the stability tests are suggested to be provided, as they provide meaningful information on the change of the onset potentials after the stability test.
+
+Response: We are grateful for the reviewer’s suggestions. We have added the current-bias curve of Sn/β-Fe2O3 after 1000-h reaction. Due to the partial reconstruction of the surface, FeOOH was generated in situ. FeOOH could act as an electrocatalyst, and the on-set potential of the photoanode after the stability test moved towards the direction of low bias voltage. Due to the electrocatalytic effect of surface FeOOH after 1000-h reaction, the on-set potential moved to the negative direction by approximately 0.1 V_{RHE}. But the photocurrent density at 1.6 V_{RHE} was reduced with the influence of surface reconstruction and corrosion.
+We added the photocurrent curve before and after the reaction as Supplementary Fig. 10:
+
+![Photocurrent curve before and after 1000 h of reaction](page_384_256_768_496.png)
+
+Supplementary Fig. 10. j-V curves of Sn/β-Fe2O3 before and after 1000 h of reaction.
+
+The following is the description added to the text.
+Lines 202–212 (Original manuscript lines 192):
+At the initial stage of the reaction, the β-Fe2O3 on the surface is converted into FeOOH in situ (Fig. 3). FeOOH itself is an efficient electrocatalyst, so the photocurrent increased. Because of the infiltration corrosion of Cl− along with the excessive surface reconstruction and the loss of Sn, the photocurrent density had a downwards trend after 700 h. It can be analyzed from Supplementary Fig. 10 that due to the electrocatalytic effect of surface FeOOH after 1000-h reaction, the on-set potential moved to the negative direction by approximately 0.1 \( V_{RHE} \). However, the photocurrent density at 1.6 \( V_{RHE} \) was reduced with the influence of surface reconstruction and corrosion. Owing to the relatively uniform distribution of Sn in the whole film, reconstruction and corrosion were controlled within a certain range of β-Fe2O3 surface, instead of continuing to the deep layer.
+
+4) Line 43, Refs 6 and 7 are not about β-Fe2O3, please double-check the accuracy of the reference list.
+
+Response: Thanks for the reviewer’s comment. Here is a citation error. We have added the correct and corresponding references here, and other corresponding reference numbers have also been changed.
+
+References
+Lines 325–329 (Original manuscript lines 298):
+[20] Zhang, N. S. et al. Paving the road toward the use of β-Fe2O3 in solar water splitting: Raman identification, phase transformation and strategies for phase stabilization. Natl. Sci. Rev. 7, 1059–1067, (2020).
+[21] Li, Y. et al. Metastable-phase β-Fe2O3 photoanodes for solar water splitting with durability exceeding 100 h. Chinese J. Catal. 42, 1992–1998, (2021).
+REVIEWERS’ COMMENTS
+
+Reviewer #2 (Remarks to the Author):
+
+The author's response has satisfied me, and their extension of the stability experiment from 1000 hours to 3000 hours further demonstrates the stability of the material in PEC. Therefore, I would like to recommend publishing this article in its current version in Nature Communications.
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have adequately addressed my concerns and the manuscript can be published after fixing the following minor issues. No further review is required.
+
+1) Ref 20 in the main text differs from the one in the response letter.
+2) The stability data (1440 h) in Supplementary Table 1 needs to be updated since they provided longer stability data (3000 h).
+Point-by-point responses for Nature Communications manuscript
+
+(ID: NCOMMS-22-49861A)
+
+Manuscript Type: Article
+
+Title: Long-term durability of metastable β-Fe₂O₃ photoanodes in highly corrosive seawater.
+
+Author(s): Changhao Liu, Ningsi Zhang, Yang Li, Rongli Fan, Wenjing Wang, Jianyong Feng, Chen Liu, Jiaou Wang, Weichang Hao, Zhaosheng Li, Zhigang Zou
+
+General response:
+We sincerely thank the editor, editorial staff and all reviewers for their comments. The manuscript has been modified point-by-point after addressing all the suggestions as listed below.
+(Our response is given in blue and the corrections in the revised manuscript are shown in red)
+
+Point-by-point responses to Reviewer(s)
+
+Reviewer #2:
+
+The author's response has satisfied me, and their extension of the stability experiment from 1000 hours to 3000 hours further demonstrates the stability of the material in PEC. Therefore, I would like to recommend publishing this article in its current version in Nature Communications.
+
+Response:
+We thank the reviewer for the very positive assessment of our work, which have greatly helped us improve the quality of the manuscript.
+Reviewer #3:
+
+The authors have adequately addressed my concerns and the manuscript can be published after fixing the following minor issues. No further review is required.
+
+Response: We are very grateful to the reviewer for the suggestions and comments, which leads to further improvements in the quality of our work. According to the comments, we have modified our manuscript discussion and corresponding responses.
+
+1) Ref 20 in the main text differs from the one in the response letter.
+
+Response: Thanks for the reviewer’s comment. We apologize for the errors in references 5, 20, and 21 in the main text, and have made corrections accordingly.
+
+References
+Lines 350–351, and 391–395
+[5] Nishiyama, H. et al. Photocatalytic solar hydrogen production from water on a 100-m^2 scale. Nature **598**, 304–307 (2021).
+[20] Zhang, N. S. et al. Paving the road toward the use of \( \beta \)-Fe_2O_3 in solar water splitting: Raman identification, phase transformation and strategies for phase stabilization. Natl. Sci. Rev. **7**, 1059–1067 (2020).
+[21] Li, Y. et al. Metastable-phase \( \beta \)-Fe_2O_3 photoanodes for solar water splitting with durability exceeding 100 h. Chinese J. Catal. **42**, 1992–1998 (2021).
+
+2) The stability data (1440 h) in Supplementary Table 1 needs to be updated since they provided longer stability data (3000 h).
+
+Response: Thanks for the reviewer’s comment.
+
+The data in Supplementary Table 1 has been updated (Supplementary Materials lines 118).
+
+Supplementary Table 1. Comparison of photoelectrochemical OER (simulated) seawater splitting with different photoanodes.
+
+<table>
+  <tr>
+    <th>Photoanodes</th>
+    <th>Light intensity</th>
+    <th>Bias potential</th>
+    <th>Current density</th>
+    <th>Stability</th>
+    <th>Ref.</th>
+  </tr>
+  <tr>
+    <td>Sn/\( \beta \)-Fe_2O_3</td>
+    <td>AM 1.5G</td>
+    <td>1.6 V<sub>RHE</sub></td>
+    <td>2.21 mA cm<sup>-2</sup></td>
+    <td><b>3000 h</b></td>
+    <td>this work</td>
+  </tr>
+  <tr>
+    <td>RhO_2/Mo-BiVO_4</td>
+    <td>Full-arc xenon lamp (\( \lambda > 300 \) nm) with higher light intensity</td>
+    <td>1.0 V<sub>Ag/AgCl</sub></td>
+    <td>18 mA cm<sup>-2</sup></td>
+    <td>270 min</td>
+    <td>14</td>
+  </tr>
+  <tr>
+    <td>TiO_2@g-C_3N_4@CoPi</td>
+    <td>AM 1.5G</td>
+    <td>1.23 V<sub>RHE</sub></td>
+    <td>1.64 mA cm<sup>-2</sup></td>
+    <td>10 h</td>
+    <td>38</td>
+  </tr>
+</table>
+<table>
+  <tr>
+    <th>WO<sub>3</sub>/g-C<sub>3</sub>N<sub>4</sub></th>
+    <th>AM 1.5G</th>
+    <th>1.23 V<sub>RHE</sub></th>
+    <th>0.73 mA cm<sup>-2</sup></th>
+    <th>1 h</th>
+    <th>39</th>
+  </tr>
+  <tr>
+    <th>Fe<sub>2</sub>O<sub>3</sub>/WO<sub>3</sub></th>
+    <th>AM 1.5G</th>
+    <th>1.23 V<sub>RHE</sub></th>
+    <th>1 mA cm<sup>-2</sup></th>
+    <th>5 h</th>
+    <th>8</th>
+  </tr>
+  <tr>
+    <th>In<sub>2</sub>S<sub>3</sub>/ANP/RND</th>
+    <th>AM 1.5G</th>
+    <th>1.23 V<sub>RHE</sub></th>
+    <th>1.53 mA cm<sup>-2</sup></th>
+    <th>2 h</th>
+    <th>9</th>
+  </tr>
+  <tr>
+    <th>In<sub>2</sub>S<sub>3</sub>/In<sub>2</sub>O<sub>3</sub></th>
+    <th>AM 1.5G</th>
+    <th>0.981 V<sub>RHE</sub></th>
+    <th>~0.2 mA cm<sup>-2</sup></th>
+    <th>1000 s</th>
+    <th>40</th>
+  </tr>
+  <tr>
+    <th>Mg doped ZnO</th>
+    <th>AM 1.5G</th>
+    <th>0.5 V<sub>Ag/AgCl</sub></th>
+    <th>~1 μA cm<sup>-2</sup></th>
+    <th>5 h</th>
+    <th>41</th>
+  </tr>
+  <tr>
+    <th>MoB/BiVO<sub>4</sub></th>
+    <th>AM 1.5G</th>
+    <th>1.23 V<sub>RHE</sub></th>
+    <th>4.30 mA cm<sup>-2</sup></th>
+    <th>70 h</th>
+    <th>42</th>
+  </tr>
+  <tr>
+    <th>NiMoO<sub>3</sub>/BiVO<sub>4</sub></th>
+    <th>AM 1.5G</th>
+    <th>1.23 V<sub>RHE</sub></th>
+    <th>3.0 mA cm<sup>-2</sup></th>
+    <th>190 h</th>
+    <th>43</th>
+  </tr>
+  <tr>
+    <th>Bi<sub>2</sub>S<sub>3</sub>/NiS/NiFeO/TiO<sub>2</sub></th>
+    <th>300 W Xe lamp</th>
+    <th>1.23 V cell voltage</th>
+    <th>10 mA cm<sup>-2</sup></th>
+    <th>4 h</th>
+    <th>44</th>
+  </tr>
+  <tr>
+    <th>Bi<sub>0.6</sub>Fe<sub>0.4</sub>VO<sub>4</sub>@CNTs</th>
+    <th>AM 1.5G</th>
+    <th>1.5 V<sub>Ag/AgCl</sub></th>
+    <th>~0.1 mA cm<sup>-2</sup></th>
+    <th>1 h</th>
+    <th>45</th>
+  </tr>
+</table>

0ae031331c2d266697df1ca15bce36f1cb2333132590b2a9b295f659d1dffe9c/preprint/preprint.md

@@ -0,0 +1,229 @@
+Ultradurability of metastable β-Fe2O3 photoanodes in highly corrosive seawater
+
+Changhao Liu
+Nanjing University
+Ningsi Zhang
+Nanjing University
+Yang Li
+Nanjing University
+Rongli Fan
+Nanjing University
+Wenjing Wang
+Nanjing University
+Jianyong Feng
+Nanjing University
+C. Liu
+Institute of High Energy Physics, Chinese Academy of Sciences
+Jiaou Wang
+Chinese Academy of Sciences   https://orcid.org/0000-0002-4686-1821
+Weichang Hao
+Beihang University   https://orcid.org/0000-0002-1597-7151
+Zhaosheng Li (zsl@nju.edu.cn)
+Nanjing University   https://orcid.org/0000-0001-8114-0432
+Zhigang Zou
+Nanjing University   https://orcid.org/0000-0003-2092-8335
+
+Article
+
+Keywords:
+
+Posted Date: December 12th, 2022
+
+DOI: https://doi.org/10.21203/rs.3.rs-2096634/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on July 17th, 2023. See the published version at https://doi.org/10.1038/s41467-023-40010-9.
+Ultradurability of metastable \( \beta\)-Fe$_2$O$_3$ photoanodes in highly corrosive seawater
+
+Changhao Liu$^{1,2}$, Ningsi Zhang$^{1,2}$, Yang Li$^1$, Rongli Fan$^1$, Wenjing Wang$^1$, Jianyong Feng$^{1,*}$, Chen Liu$^3$, Jiaou Wang$^3$, Weichang Hao$^4$, Zhaosheng Li$^{1,2,*}$, Zhigang Zou$^{1,2}$
+
+Affiliations:
+$^1$ Collaborative Innovation Center of Advanced Microstructures, National Laboratory of Solid State Microstructures, College of Engineering and Applied Sciences, Nanjing University; 22 Hankou Road, Nanjing 210093, China
+$^2$ Jiangsu Key Laboratory for Nano Technology, Nanjing University; 22 Hankou Road, Nanjing 210093, China
+$^3$ Beijing Synchrotron Radiation Facility, Institute of High Energy Physics, Chinese Academy of Sciences, Beijing 100049, China.
+$^4$ School of Physics and Centre of Quantum and Matter Sciences, International Research Institute for Multidisciplinary Science, Beihang University, Beijing 100191, China
+*Corresponding authors. Email: zsli@nju.edu.cn; fengjianyong@nju.edu.cn.
+
+Abstract: Durability is one prerequisite for material application. Photoelectrochemical (PEC) decomposition of seawater is a promising approach to produce clean hydrogen by using solar energy, but it always suffers from serious Cl$^-$ corrosion. We found that the main deactivation mechanism of the photoanodes is oxide surface reconstruction accompanied by the coordination of Cl$^-$ during seawater splitting, and the stability of the photoanodes can be greatly improved by enhancing the metal-oxygen interaction. Taking the metastable \( \beta\)-Fe$_2$O$_3$ photoanode as an example, Sn added to the lattice can enhance the M–O bonding energy and hinder the transfer of protons to lattice oxygen, thereby inhibiting excessive surface hydration and Cl$^-$ coordination. Therefore, the Sn/\( \beta\)-Fe$_2$O$_3$ photoanode without any extra electrocatalyst or protective overlayer delivered a record durability for PEC seawater splitting over 1440 h.
+Main Text:
+
+The use of PEC water splitting to produce hydrogen can realize the conversion of solar energy to hydrogen energy in one step, which is a very promising solution for building a low-carbon society\( ^{1-5} \). The long-term stability of photoelectrodes is an essential prerequisite for the practical application of PEC water splitting for hydrogen production\(^6\). However, except for iron oxide, almost all bare photoelectrodes show unsatisfactory stability in water splitting for hydrogen production, let alone in highly corrosive seawater\(^{7-10}\). Some strategies, such as protective layers, electrocatalysts, and tuning electrolyte composition, have been used to improve the durability of photoelectrodes in aqueous electrolytes without Cl\(^-\) ions\(^{11-13}\). Little attention has been given to improving the stability of photoelectrodes in aqueous electrolytes with Cl\(^-\) ions\(^{14, 15}\) since Cl\(^-\) ions easily corrode photoelectrode materials and may participate in the competitive oxidation reaction to produce Cl\(_2\) or ClO\(^-\)\(^{16-19}\).
+
+Herein, we studied the effect of Cl\(^-\) ions on the stability of a photoelectrode such as \( \beta \)-Fe\(_2\)O\(_3\). Recently, \( \beta \)-Fe\(_2\)O\(_3\), as a metastable phase of iron oxide, has entered our research horizon due to a theoretical solar-to-hydrogen efficiency of 20.9%, showing good stability of PEC water splitting in aqueous electrolytes without Cl\(^-\) ions\(^{6, 7}\). We have revealed that the Cl\(^-\) ions in seawater will damage the surface hydrated layer of \( \beta \)-Fe\(_2\)O\(_3\) photoanodes, thus remarkably reducing their stability. Dispersed Sn single atoms in the lattice were found to endow the \( \beta \)-Fe\(_2\)O\(_3\) photoanodes with good inhibition of hydration and resistance to Cl\(^-\) attack in seawater. As a result, the bare Sn/\( \beta \)-Fe\(_2\)O\(_3\) shows excellent durability in seawater splitting over 1440 h and is by far the most stable
+photoanode. This study may ignite the dawn of application for PEC seawater splitting for hydrogen production and deepen the understanding of the seawater corrosion of oxides.
+
+Material characterization of the \( \beta\)-Fe$_2$O$_3$ photoanode
+
+Metastable \( \beta\)-Fe$_2$O$_3$ photoanodes doped with Sn were prepared by the spray pyrolysis method, and their phases were accurately determined (Supplementary Fig. 1). The Sn/\( \beta\)-Fe$_2$O$_3$ film is composed of blocks arranged vertically with a thickness of approximately 400 nm (Fig. 1a). A large area of lattice stripes indicates good crystallinity of \( \beta\)-Fe$_2$O$_3$ (Fig. 1b). Many bright spots with high contrast in the (1 1 0) crystal plane in Fig. 1c correspond to the Sn single atom in the \( \beta\)-Fe$_2$O$_3$ lattice. One of the regions was selected for three-dimensional modelling, which shows the contrast difference between Sn atoms and surrounding Fe atoms (Fig. 1d). It was confirmed that the lattice position of Fe was substituted by Sn. A clear atomic image of the (1 1 1) crystal plane taken from another region of \( \beta\)-Fe$_2$O$_3$ and fast Fourier transform (FFT) patterns of the (1 1 0) and (1 1 1) planes were obtained (Supplementary Fig. 1). These lattice atomic images and FFT patterns are completely consistent with the atomic arrangement of the corresponding crystal plane in the theoretical model.
+
+PEC properties and seawater splitting stability
+
+The as-prepared \( \beta\)-Fe$_2$O$_3$ photoanodes were tested for PEC simulated seawater splitting. The saturated photocurrent density of the 2% Sn/\( \beta\)-Fe$_2$O$_3$ photoanodes reaches 2.21 mA cm$^{-2}$ at 1.6 V$_{RHE}$, which is 8.5 times that of \( \beta\)-Fe$_2$O$_3$ photoanodes
+(Supplementary Fig. 2a). The Sn dopants do not affect the light absorption of the \( \beta \)-Fe\(_2\)O\(_3\) photoanodes, while the PEC performance improvement is partly due to the increased carrier concentration (Supplementary Fig. 3). Sn can simultaneously adjust the chemical field at the semiconductor/electrolyte interface, which significantly reduces the AC impedance of the interface (Supplementary Fig. 2b).
+
+The most important role of Sn dispersed in the lattice is to surprisingly promote its durability in simulated seawater with Cl\(^-\) ions. Specifically, in Fig. 2a, the stability of the \( \beta \)-Fe\(_2\)O\(_3\) photoanodes is good in 1 M KOH electrolyte within 100 h, while its photocurrent density decreased obviously in 1 M KOH + 0.5 M NaCl electrolyte. This indicates that Cl\(^-\) significantly reduced its PEC stability. In contrast, the Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes could still maintain stable performance even in a saturated NaCl electrolyte within 100 h without decay, which was much better than the \( \beta \)-Fe\(_2\)O\(_3\) photoanodes. In the stability test, the photocurrent increased slightly in a period of time after the beginning of the reaction due to the change in the state of Fe and O on the surface\(^{20-22}\). Correspondingly, the AC impedance of the Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanode in the first 50 h gradually decreases (Fig. 2b). The HADDF image of the photoanode also shows that an amorphous hydrated layer was formed on the surface of \( \beta \)-Fe\(_2\)O\(_3\) (Fig. 2c), which indicates that the FeOOH hydrated layer spontaneously formed on the surface during the reaction process. The specific process and impact of surface reconstruction will be further discussed below. Furthermore, the Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanode shows excellent stability over 1440 h in simulated seawater (Fig. 2d). After 1440 h, the photocurrent maintains 90.5% of the initial value. The Sn/\( \beta \)-Fe\(_2\)O\(_3\)
+photoanode has achieved the longest durability in research on PEC seawater splitting over the years, as shown in Fig. 2e, even without any extra electrocatalyst or protective overlayer. Additionally, the Sn/β-Fe2O3 photoanodes showed excellent stability in alkaline natural seawater (Supplementary Fig. 4).
+
+Effect of Sn on the resistance of photoanode to Cl⁻ corrosion
+
+The evolution process of the β-Fe2O3 photoanode surface in seawater splitting is explored here. XPS analysis can be used to obtain the state of the β-Fe2O3 photoanode surface elements in contact with the electrolyte during the reaction. As shown in the XPS spectrum of O in the β-Fe2O3 photoanode (Supplementary Fig. 5a), the peak of M–O at 529.4 eV decreases after the reaction. It transforms to M–OH at 531.7 eV, with a significant shift from lattice oxygen to hydroxyl oxygen on the surface\(^{23}\). This corresponds to the reconstruction of the β-Fe2O3 surface in alkaline electrolytes. In the XPS plot of Sn/β-Fe2O3 photoanodes with 100 h of reaction in Fig. 3a, a large number of O atoms can remain in the form of lattice oxygen when Sn is present on the surface. After 100 h of reaction, no Cl signal was detected on the surface of the Sn/β-Fe2O3 photoanodes (Fig. 3b). In contrast, the Cl signal was detected on the β-Fe2O3 surface (Supplementary Fig. 5b), which indicates that the process of lattice oxygen reconstruction was accompanied by the adsorption or implantation of Cl⁻ in the electrolyte. During surface hydration and lattice reformation, β-Fe2O3 slowly dissolves, which can be determined by inductive coupled plasma emission spectrometry (Supplementary Table 2). The amount of dissolved Fe atoms was also significantly reduced when Sn acted as an anchor at the surface. However, due to the intense
+hydration, the surface lattice was still continuously attacked after a long reaction time, accompanied by O remodelling and loss of metal elements. The XPS peak of the Sn 3d signal disappeared after 1000 h of reaction (Fig. 3g), indicating that Sn is also slowly lost during lattice reconstruction by surface hydration. Cations are also involved in surface hydration and embedded in the hydrated layer, such as Na\(^+\), and after 1000 h, cations were also detected on the photoanode surface. As shown in the Raman spectra of the \( \beta \)-Fe\(_2\)O\(_3\) photoanode surface (Fig. 3c), after 100 h and 1000 h of seawater splitting, there are two peaks of M–OOH at approximately 470 cm\(^{-1}\) and 550 cm\(^{-1}\) \(^{24,25}\), which echo the change in the O 1 s XPS peak. The \( \alpha \)-Fe\(_2\)O\(_3\) peak can be observed when the \( \beta \)-Fe\(_2\)O\(_3\) photoanodes are further calcined at 600 °C. Here, \( \alpha \)-Fe\(_2\)O\(_3\) was transformed from FeOOH generated by surface reconstruction during heat treatment. A NaCl peak appeared on the surface after 1000 h of reaction, indicating that the crystallization of anions and cations diffused into the hydration layer.
+
+To further confirm the reconstruction of the \( \beta \)-Fe\(_2\)O\(_3\) photoanodes and the exchange of atoms at the interface during the reaction in the electrolyte, the surface element distribution measurement was probed by time-of-flight secondary ion mass spectrometry (TOF-SIMS). In Fig. 3e, the signal of Cl can be detected on the \( \beta \)-Fe\(_2\)O\(_3\) surface after the reaction in simulated seawater for 100 h. The Cl content of the \( \beta \)-Fe\(_2\)O\(_3\) surface is much higher than that of the Sn/\( \beta \)-Fe\(_2\)O\(_3\) surface. This confirmed that the presence of Sn can significantly improve the rejection of Cl\(^-\) in the electrolyte. Meanwhile, H\(_2\)\(^{18}\)O was added to explore electrolyte participation in the surface reconstruction of the \( \beta \)-Fe\(_2\)O\(_3\) photoanode. \(^{18}\)O in the electrolyte participates in the
+formation of a hydration layer, so the signal of \(^{18}\)OH with surface m/z = 19.005 can be detected. The signal intensity of \(^{18}\)OH on the Sn/\(\beta\)-Fe\(_2\)O\(_3\) photoanode surface is much weaker than that on the \(\beta\)-Fe\(_2\)O\(_3\) surface, indicating that the Sn dopants weaken lattice oxygen reconstruction. In the depth profiling in Fig. 3f, the content of both Cl and \(^{18}\)OH in the Sn/\(\beta\)-Fe\(_2\)O\(_3\) surface decays faster with depth than without Sn dopants. This reveals that surface reconstruction and Cl\(^-\) erosion occur simultaneously. The \(^{18}\)O added to the electrolyte participates in the reconstruction of the lattice oxygen of \(\beta\)-Fe\(_2\)O\(_3\) and forms M–\(^{18}\)OH. At the same time, Cl\(^-\) in the electrolyte would also first be adsorbed on the surface and gradually infiltrate into the bulk with surface reconstruction. The \(\beta\)-Fe\(_2\)O\(_3\) photoanode undergoes excessive surface reconstruction, resulting in a thicker hydrated layer. Cl\(^-\) shuttles and infiltrates into it, which may destroy the structure of the \(\beta\)-Fe\(_2\)O\(_3\) photoanode and affect the interface water oxidation reaction. Sn inhibits the exchange of \(^{18}\)O and lattice oxygen and suppresses the erosion of Cl\(^-\), thus obtaining a more stable photoanode surface.
+
+The O K-edge of the soft X-ray absorption near-edge structure (XANES) spectrum shows the change in the lattice oxygen state before and after adding Sn to the lattice. The spectrogram of \(\beta\)-Fe\(_2\)O\(_3\) is similar to the O K-edge of standard iron oxide\(^{27}\). After adding Sn, the X-ray absorption peak of oxygen shifts to the direction of high energy, which reflects that the addition of Sn effectively improves the bonding energy of O in the lattice. A shoulder peak at 532.3 eV corresponds to the contribution of the Sn 5s orbit\(^{28,29}\). This shows that the Sn atoms dispersed in the lattice change the average chemical environment of O and play an anchor role in lattice oxygen.
+The enhanced metal–oxygen interaction in the surface chemical reaction is specifically manifested in that the lattice oxygen at the semiconductor electrolyte interface has more difficulty accepting protons, which can be confirmed by the proton-coupled electron transfer process analyzed by the H/D kinetic isotope effect\(^{30-32}\). The OER on the photoanode surface is a proton-coupled electron transfer process involving four electrons. Specifically, the reaction intermediate species *OH and *OOH transfer one electron to the semiconductor and discard one proton\(^{33,34}\). The isotope effect is particularly significant at low pH. The \( j_{\mathrm{H}_2\mathrm{O}}/j_{\mathrm{D}_2\mathrm{O}} \) value of the \( \beta \)-Fe\(_2\)O\(_3\) photoanode is always lower than that of Sn/\( \beta \)-Fe\(_2\)O\(_3\), indicating that the \( \beta \)-Fe\(_2\)O\(_3\) surface has a stronger affinity for protons. The lattice oxygen on the surface of \( \beta \)-Fe\(_2\)O\(_3\) easily acts as a proton acceptor, which to some extent accelerates the proton coupling process in the reaction process. However, lattice oxygen as a proton acceptor will bring about the problem of structural stability being destroyed. As demonstrated in Fig. 4c, protons transferred to nearby locations will combine with lattice oxygen, break the M–O bond and generate an FeOOH hydrated layer. When the M–O bond breaks, oxygen in solution will exchange with lattice oxygen, and Cl\(^-\) will also coordinate with Fe and destroy the surface structure. On the other hand, hydrated FeOOH is a loose amorphous or layered structure, which is also prone to the insertion and adsorption of Cl\(^-\), thus affecting the activity of water splitting. The Sn atoms dispersed in the lattice play a role in anchoring the lattice oxygen to prevent proton coupling between the reaction intermediate and the lattice oxygen. which shows that the proton transfer process will have a greater impact on the reaction kinetics. When the pH rises, proton transfer is no longer the rate-
+determining step. The advantages of high valence cation-doped Sn dispersed in the bulk phase in increasing the carrier concentration and conductivity can also be fully demonstrated. Therefore, the photocurrent increases to 8.5 times that of the \( \beta\)-Fe\(_2\)O\(_3\) photoanode. Although alkaline electrolytes are used in PEC tests, local pH will decrease in the water oxidation reaction, and protons with higher local concentrations will also exist. These protons attack lattice oxygen, causing surface reconstruction. Sn in the lattice enhances the metal oxygen interaction, thus inhibiting the wrong proton transfer path and avoiding surface hydration and Cl\(^-\) corrosion.
+
+The advantage of uniformly dispersed Sn in the bulk phase is that when the surface is hydrated and peeled by corrosion, the exposed Sn/\( \beta\)-Fe\(_2\)O\(_3\) is still corrosion resistant. During the annealing process of the Sn/\( \beta\)-Fe\(_2\)O\(_3\) photoanode, Sn atoms tend to diffuse to the surface\(^{26}\), but the Sn in the bulk is still relatively uniform (Supplementary Fig. 7). In contrast, we covered a layer of efficient OER electrocatalyst CoFe-LDH on the surface of Sn/\( \beta\)-Fe\(_2\)O\(_3\) as a protective passivation layer. However, a significant downwards trend of photocurrent was observed in the first 50 h of the reaction, and after a long time, the current gradually decreased to the level without loading the electrocatalyst (Supplementary Fig. 9). This shows that the surface modification of the electrocatalyst cannot resist the corrosion of Cl\(^-\) in seawater. Metal hydroxide itself will also be reconstructed in the OER reaction, which will also be accompanied by the problems of Cl\(^-\) coordination and structural collapse. Finally, the structure is destroyed and gradually dissolved and peeled off. This extra loaded electrocatalyst protective
+layer often protects the photoanode by its own corrosion and consumption, which cannot fundamentally solve the long-term stability problem.
+
+Conclusion
+
+Here, we revealed that excessive hydration reconstruction of the surface will corrode the surface of the oxide photoanode with the corrosion of Cl\(^{-}\) ions in the solution. The anchoring of the surface lattice by Sn hinders the transfer of protons to lattice oxygen, and the probability of oxygen hydrogen bonding will decrease due to the strong M–O bond, thereby suppressing the surface reconstruction and coordination of Cl\(^{-}\). The Sn/\(\beta\)-Fe\(_2\)O\(_3\) photoanode constitutes by far the most durable photoanode for seawater splitting. This strategy can also improve the durability of other photoanodes, such as \(\alpha\)-Fe\(_2\)O\(_3\) (Supplementary Fig. 10). This study will pave a new path to solving the problem of the long-term durability of photoelectrodes in energy conversion.
+Fig. 1. | Cross section and crystal structure of the Sn/β-Fe2O3 photoanode.
+
+a, b, HAADF images of the Sn/β-Fe2O3 film cross-section at different magnifications.
+c, Atomic image of the (1 1 0) plane of β-Fe2O3. d, Local enlargement near doped Sn atoms and three-dimensional modelling of surface contrast.
+Fig. 2. | PEC properties of the \( \beta \)-Fe\(_2\)O\(_3\) photoanode in simulated seawater.
+
+a, Stability test of \( \beta \)-Fe\(_2\)O\(_3\) and Sn/\( \beta \)-Fe\(_2\)O\(_3\) in 1 M KOH with and without 0.5 M NaCl for 100 h. b, AC electrochemical impedance spectra of Sn/\( \beta \)-Fe\(_2\)O\(_3\) at 1.6 V\(_{RHE}\) after the reaction. c, HAADF images of the Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanode after 100 h of reaction in simulated seawater. d, Stability test of Sn/and \( \beta \)-Fe\(_2\)O\(_3\) in saturated NaCl solution for 1440 h. e, Summary of the photoanode stability of PEC (simulated) seawater splitting over the years. Detailed information can be found in Supplementary Table 1.
+Fig. 3. | Evolution of the \( \beta\)-Fe$_2$O$_3$ surface during the long-term seawater decomposition reaction.
+
+a–c, XPS spectra of O 1s, Cl 2p and Sn 3d of Sn/\( \beta\)-Fe$_2$O$_3$ before, after 100 h, and after 1000 h of seawater splitting reaction. d, Raman spectra of \( \beta\)-Fe$_2$O$_3$ photoanodes with different reaction times and annealing treatments. e, f, TOF-SIMS of the distributions of Cl and $^{18}$OH on the surface and depth profiling of the \( \beta\)-Fe$_2$O$_3$ photoanode before and after Sn doping after the 100-h reaction in simulated seawater with 20 wt.% H$_2$$^{18}$O.
+Fig. 4. | Effect of Sn atoms in \( \beta \)-Fe\(_2\)O\(_3\) on resistance to seawater corrosion.
+
+a, XANES spectra of the O K-edge in \( \beta \)-Fe\(_2\)O\(_3\). b, The steady-state photocurrent \( j_{H_2O}/j_{D_2O} \) and \( j_{Sn}/j_{Pure} \) values at different pH values. c, Schematic diagram of doped Sn atoms against Cl\(^-\) corrosion in seawater splitting. Sn enhances the M–O bonds, prevents the hydrated surface reconstruction caused by the transfer of H\(^+\) to lattice oxygen, and weakens the coordination of Cl\(^-\).
+References
+
+1. Fujishima, A. & Honda, K. Electrochemical Photolysis of Water at Semiconductor Electrode. Nature **238**, 37–38 (1972).
+
+2. Fang, T. et al. Reactive inorganic vapor deposition of perovskite oxynitride films for solar energy conversion. Research **2019**, 9 (2019).
+
+3. De Luna, P. et al. What would it take for renewably powered electrosynthesis to displace petrochemical processes? Science **364**, eaav3506 (2020).
+
+4. Lewis, N. S. Research opportunities to advance solar energy utilization. Science **351**, aad1920 (2016).
+
+5. Zhang, Y. et al. Homogeneous solution assembled Turing structures with near zero strain semi-coherence interface. Nat. Commun. **13**, 2942 (2022).
+
+6. Kibsgaard, J. & Chorkendorff, I. Considerations for the scaling-up of water splitting catalysts. Nat. Energy **4**, 430–433 (2019).
+
+7. Jadwiszczak, M., Jakubow-Piotrowska, K., Kedzierzawski, P., Bienkowski, K. & Augustynski, J. Highly efficient sunlight-driven seawater splitting in a photoelectrochemical cell with chlorine evolved at nanostructured WO$_3$ photoanode and hydrogen stored as hydride within metallic cathode. Adv. Energy Mater. **10**, 1903213 (2020).
+
+8. Li, Y. G. *et al.*, Photoelectrochemical splitting of natural seawater with \( \alpha \)-Fe$_2$O$_3$/WO$_3$ nanorod arrays. Int. J. Hydrogen Energy **41**, 4096–4105 (2016).
+
+9. Yang, J. S. & Wu, J. J. Toward eco-friendly and highly efficient solar water splitting using In$_2$S$_3$/anatase/rutile TiO$_2$ dual-staggered-heterojunction
+nanodendrite array photoanode. ACS Appl. Mater. Interfaces **10**, 3714–3722 (2018).
+
+10. Farras, P., Strasser, P. & Cowan, A. J. Water electrolysis: Direct from the sea or not to be? Joule **5**, 1921-1923, (2021).
+
+11. Lee, D. K. & Choi, K. S. Enhancing long-term photostability of BiVO$_4$ photoanodes for solar water splitting by tuning electrolyte composition. Nat. Energy **3**, 53–60 (2018).
+
+12. Kuang, Y. B. *et al.* Ultrastable low-bias water splitting photoanodes via photocorrosion inhibition and in situ catalyst regeneration. Nat. Energy **2**, 16191 (2017).
+
+13. Hu, S. *et al.* Amorphous TiO$_2$ coatings stabilize Si, GaAs, and GaP photoanodes for efficient water oxidation. Science **344**, 1005–1009 (2014).
+
+14. Luo, W. J. *et al.* Solar hydrogen generation from seawater with a modified BiVO$_4$ photoanode. Energy Environ. Sci. **4**, 4046–4051 (2011).
+
+15. Zhong, D. K. & Gamelin, D. R. Photoelectrochemical water oxidation by cobalt catalyst ("Co-Pi")/\( \alpha \)-Fe$_2$O$_3$ composite photoanodes: oxygen evolution and resolution of a kinetic bottleneck. *J. Am. Chem. Soc.* **132**, 4202–4207 (2010).
+
+16. Li, Z. S., Luo,W. J., Zhang, M. L., Feng, J. Y. & Zou, Z. G. Photoelectrochemical cells for solar hydrogen production: current state of promising photoelectrodes, methods to improve their properties, and outlook. Energy Environ. Sci. **6**, 347–370 (2013).
+
+17. Hausmann, J. N., Schlogl, R., Menezes, P. W. & Driess, M. Is direct seawater
+splitting economically meaningful? Energy Environ. Sci. **14**, 3679–3685 (2021).
+
+18. Kuang, Y. *et al.* Solar-driven, highly sustained splitting of seawater into hydrogen and oxygen fuels. *Proc. Natl. Acad. Sci. U. S. A.* **116**, 6624–6629 (2019).
+
+19. Tong, W. M. et al. Electrolysis of low-grade and saline surface water. *Nat. Energy* **6**, 935–935 (2021).
+
+20. Feng, C. *et al.* A self-healing catalyst for electrocatalytic and photoelectrochemical oxygen evolution in highly alkaline conditions. *Nat. Commun.* **12**, 5980 (2021).
+
+21. Chung, D. Y. et al. Dynamic stability of active sites in hydr(oxy)oxides for the oxygen evolution reaction. *Nat. Energy* **5**, 550–550 (2020).
+
+22. Hunter, B. M. *et al.* Trapping an iron(VI) water-splitting intermediate in nonaqueous media. *Joule* **2**, 747–763 (2018).
+
+23. Kim, J. Y., Youn, D. H., Kang, K. & Lee, J. S. Highly conformal deposition of an ultrathin FeOOH layer on a hematite nanostructure for efficient solar water splitting. *Angew. Chem. Int. Ed.* **55**, 10854–10858 (2016).
+
+24. Tang, F. Liu, T., Jiang, W. L. & Gan, L. Windowless thin layer electrochemical Raman spectroscopy of Ni-Fe oxide electrocatalysts during oxygen evolution reaction. *J. Electroanal. Chem.* **871**, 6 (2020).
+
+25. Duan, Y. *et al.* Scaled-up synthesis of amorphous NiFeMo oxides and their rapid surface reconstruction for superior oxygen evolution catalysis. *Angew. Chem. Int. Ed.* **58**, 15772–15777 (2019).
+26. Zhang, H. M. et al. Gradient tantalum-doped hematite homojunction photoanode improves both photocurrents and turn-on voltage for solar water splitting. Nat. Commun. **11**, 4622 (2020).
+
+27. Frati, F., Hunault, M. & de Groot, F. M. F. Oxygen K-edge X-ray absorption spectra. Chem. Rev. **120**, 4056–4110, (2020).
+
+28. McLeod, J. A. et al. Band gaps and electronic structure of alkaline-earth and post-transition-metal oxides. Phys. Rev. B **81**, 245123 (2010).
+
+29. McLeod, J. A. et al. Chemical bonding and hybridization in 5p binary oxide. J. Phys. Chem. C **116**, 24248–24254, (2012).
+
+30. Burke, M. S., Kast, M. G., Trotochaud, L., Smith, A. M. & Boettcher, S. W. Cobalt-iron (Oxy)hydroxide oxygen evolution electrocatalysts: the role of structure and composition on activity, stability, and mechanism. J. Am. Chem. Soc. **137**, 3638–3648 (2015).
+
+31. Dau, H. et al. The Mechanism of water oxidation: from electrolysis via homogeneous to biological catalysis. ChemCatChem **2**, 724–761 (2010).
+
+32. Chen, J., Li, Y. F., Sit, P. & Selloni, A. Chemical dynamics of the first proton-coupled electron transfer of water oxidation on TiO$_2$ anatase. J. Am. Chem. Soc. **135**, 18774–18777 (2013).
+
+33. Iandolo, B. & Hellman, A. The role of surface states in the oxygen evolution reaction on hematite. Angew. Chem. Int. Ed. **53**, 13404–13408 (2014).
+
+34. Li, Y. F., Liu, Z. P., Liu, L. L. & Gao, W. G. Mechanism and activity of photocatalytic oxygen evolution on titania anatase in aqueous surroundings. J.
+Am. Chem. Soc. **132**, 13008–13015 (2010).
+
+35. Li, Y. G. et al. Efficient and stable photoelectrochemical seawater splitting with TiO$_2$@g-C$_3$N$_4$ nanorod arrays decorated by Co-Pi. *J. Phys. Chem. C* **119**, 20283–20292 (2015).
+
+36. Li, Y. G. et al. Construction of inorganic-organic 2D/2D WO$_3$/g-C$_3$N$_4$ nanosheet arrays toward efficient photoelectrochemical splitting of natural seawater. *Phys. Chem. Chem. Phys.* **18**, 10255–10261 (2016).
+
+37. Sharma, M. D., Mahala, C. & Basu, M. Photoelectrochemical water splitting by In$_2$S$_3$/In$_2$O$_3$ composite nanopyramids. *ACS Appl. Nano Mater.* **3**, 11638–11649 (2020).
+
+38. Sahoo, P., Sharma, A., Padhan, S. & Thangavel, R. Visible light driven photosplitting of water using one dimensional Mg doped ZnO nanorod arrays. *Int. J. Hydrogen Energy* **45**, 22576–22588 (2020).
+
+39. Gao, R. T. et al. Ultrastable and high-performance seawater-based photoelectrolysis system for solar hydrogen generation. *Appl. Catal. B-Environ.* **304**, 120883 (2022).
+
+40. Guo, X. T., Liu, X. H. & Wang, L. NiMoO$_x$ as a highly protective layer against photocorrosion for solar seawater splitting. *J Mater. Chem. A* **10**, 1270–1277 (2022).
+
+41. She, X. F. et al. Floc-like CNTs jointed with Bi$_x$Fe$_{1-x}$VO$_{(4)}$ nanoparticles for high efficient and stable photoelectrochemical seawater splitting. *J. Alloys Compd.* **893**, 162146 (2022).
+42. Seenivasan, S., Moon, H. & Kim, D. H. Multilayer strategy for photoelectrochemical hydrogen generation: new electrode architecture that alleviates multiple bottlenecks. Nano-Micro Lett. **14**, 78 (2022).
+
+Methods
+
+Preparation of photoanode
+
+Typically, in an experiment, 0.01 mol of iron acetylacetonate (AcAcFe) was dissolved in 500 mL ethanol by stirring with a magnetic force for over 48 h. Fluorine-doped tin oxide (FTO) conductive glass was cut into dimensions of 2 cm × 1 cm, wrapped with aluminum foil to make a deposition area of 1 cm × 1 cm and then placed in a tube furnace with a set temperature of 480 °C. The precursor solution was added to the injection pump and dispersed into droplets by using an ultrasonic atomizer. During the experiment, 40 mL of precursor solution was injected at a speed of 1.6 mL min\(^{-1}\), which equally matched the power of the ultrasonic atomizer. Using air as the carrier gas, the precursor was fed into a tubular furnace. After deposition, the film was annealed in a muffle furnace at 600 °C for 3 h at a heating rate of 10 °C min\(^{-1}\). The Sn/\( \beta \)-Fe\(_2\)O\(_3\) films were prepared using the same spray pyrolysis method by adding a certain amount of tetrabutyltin (C\(_{16}\)H\(_{36}\)Sn, analytical reagent, Aladdin) ethanol solution to the precursor solution so that the Sn atom concentration accounted for 1%, 2%, 3%, and 4% of the total Sn and Fe atoms. The CoFe-LDH @ Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes were prepared by a hydrothermal method. The as-prepared Sn/\( \beta \)-Fe\(_2\)O\(_3\) photoanodes were put
+into a 100 mL hydrothermal kettle, 50 mL of a solution containing 0.002 mol L\(^{-1}\) cobalt nitrate hexahydrate (Co(NO\(_3\))\(_2\)·6H\(_2\)O, Sinopharm Chemical Reagent), 0.002 mol L\(^{-1}\) iron(III) nitrate nonahydrate (Fe(NO\(_3\))\(_3\)·9H\(_2\)O, analytical reagent, Aladdin)), 0.005 mol L\(^{-1}\) urea (Aladdin) and 0.001 mol L\(^{-1}\) trisodium citrate was added, and the reaction was carried out in an oven at 120 °C for 5 h.
+
+Characterization
+
+To identify the crystal structures of the \( \beta \)-Fe\(_2\)O\(_3\) photoanodes, they were measured by powder X-ray diffraction (XRD, Rigaku Ultima III, Cu K\( \alpha \) radiation, \( \lambda = 1.54178 \) Å) at 40 kV and 40 mA. The surface morphology of the \( \beta \)-Fe\(_2\)O\(_3\) photoanodes was examined by a high-resolution scanning electron microscope (HRSEM, ZEISS ULTRA 55 at an accelerating voltage of 5 kV). Raman spectra of \( \beta \)-Fe\(_2\)O\(_3\) photoanodes were characterized with a confocal laser Raman spectrometer (Japan, Horiba, LabRAM Aramis). X-ray photoemission spectroscopy (XPS, PHI 5000 VersaProbe) was used to characterize the content and valence of Sn, O, Fe and Co, and the binding energy was calibrated by the adventitious carbon I s line at 284.8 eV. The optical absorption spectra of the photoanode were tested on a UV–Visible–NIR (near-infrared) spectrophotometer (PerkinElmer, UV3600 UV–Vis–NIR spectrophotometer). Transmission electron microscopy (TEM) and high-resolution transmission electron microscopy (HRTEM) images were obtained on an FEI Tecnai G2 F30. High-angle annular dark field (HAADF) scanning transmission electron microscopy (STEM) images were obtained
+by a JEOL JEM-ARM200F microscope incorporated with a spherical aberration correction system for STEM.
+
+PEC measurements
+
+The PEC measurements were carried out in a PEC cell with an electrochemical analyser (CHI-760E, CH Instrument, Shanghai) in a three-electrode system including a reference electrode consisting of Ag/AgCl placed in a saturated KCl solution, Pt foil as the counter electrode, and \( \beta \)-Fe\(_2\)O\(_3\) photoanodes as working electrodes. The electrolyte was a 1 M KOH aqueous solution for freshwater and 1 M KOH with 0.5 M NaCl for simulated seawater. The potential was reported vs. the reversible hydrogen electrode (RHE) with \( E_{\text{RHE}} = E_{\text{Ag/AgCl}} + 0.197 + 0.0591 \) pH. The photocurrent density was measured under AM 1.5 G light source, and the light intensity was 100 mW cm\(^{-2}\). A Newport 91150 V standard silicon cell was used as the reference standard for calibration. Mott-Schottky analysis was performed at bias potentials from 0.5 V to 1.5 V vs. RHE. AC electrochemical impedance was obtained at a bias of 1.6 V\(_{\text{RHE}}\) over the frequency range of 100 kHz to 1 Hz. The PEC stabilities were tested at a constant potential of 1.6 V\(_{\text{RHE}}\) under LED-simulated sunlight sources through illumination from the front side.
+
+In the PEC test of the H/D kinetic isotope effect, the electrolyte was measured with a pH meter to keep the concentrations of OH\(^-\) and OD\(^-\) in the solution the same (pD = pH\(_{\text{read}} + 0.4\)). D\(_2\)O was purchased from Bide Pharmatech Ltd. (99.9% atom %D). The pD values were adjusted by NaOD (Aladdin, 30 wt.% solution in D\(_2\)O, 99.5%). In the
+current time curve, the photocurrent density value after 50 s of reaction was selected as the steady-state value for the calculation of \( j_{\mathrm{H2O}}/j_{\mathrm{D2O}} \) and \( j_{\mathrm{Sw}}/j_{\mathrm{Pure}} \) (Supplementary Fig. 8).
+
+**Time-of-flight secondary ion mass spectrometry (TOF-SIMS) tests**
+
+TOF-SIMS tests were carried out by PHI nanoTOF II Time-of-Flight SIMS. Bi$_3^{++}$ with an energy of 30 eV was used in the acquisition phase in high mass resolution mode. An Ar ion gun with an energy of 4 kV was used in the sputter phase with a sputter rate of 0.4 nm/s on SiO$_2$. Before the \( \beta \)-Fe$_2$O$_3$ photoanode was tested, the reactions in the electrolyte with 1 M KOH + 0.5 M NaCl and 20 wt.% H$_2$$_{18}$O for 100 h were carried out.
+
+**X-ray absorption near-edge structure (XANES) tests**
+
+Soft X-ray absorption near-edge structure (XANES) measurements were performed at the Beijing Synchrotron Radiation Facility (BSRF), 4B9B beamline. The O-K edge and Fe-L edge spectra were collected in total electron yield (TEY) mode by measuring the sample current with an amperemeter. All spectra were normalized to the intensity of the incident beam (I0), which was measured simultaneously with the current emitted from a gold mesh located after the last optical elements of the beamline. The photon energy was calibrated using the Au-4f core level at 84.0 eV in binding energy by measuring a clean polycrystalline gold foil that is electrically connected to the sample.
+
+**Computational processing**
+The calculations on pure and Sn/β-Fe2O3 were implemented in the VASP (Vienna Ab initio Simulation Package) based on density functional theory, with a projected-augmented-wave method in the scheme of generalized-gradient approximation. The strong on-site Coulomb repulsion among the localized Fe 3d electrons was described with the generalized-gradient approximation + U approach (U is the strength of the on-site Coulomb interaction). The exchange-correlation effects were treated using the generalized gradient approximation (GGA) in the Perdew-Burke-Ernzerhof parametrization, with spin-polarized effects considered.
+
+Acknowledgements: We are indebted to Prof. Yixin Zhao (Shanghai Jiaotong University) for discussions.
+
+Funding: The authors thank the National Science Fund for Distinguished Young Scholars [No. 22025202], National Key Research and Development Program of China [Nos. 2018YFA0209303 and 2021YFA1502100], and National Natural Science Foundation of China [Nos. 51972165 and 51902153] for financial support.
+
+Author contributions: Z.L. constructed the concept and designed the project. Z.L. supervised the study. N.Z., Y.L., J.F., W.W., C.L., J.W., W.H. and Z.Z. advised on the research. C.H.L. and R.F. collected and analysed the experimental data. C.H.L. and Z.L. wrote the manuscript. Z.L. and J.F. revised the manuscript. All the authors contributed to the discussions about the manuscript.
+Competing interests: The authors declare that they have no competing interests.
+
+Data and materials availability: All data are available in the main text or the Supplementary Information.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SupportingInformationLiuCH.pdf

0b3c54e3a936d3159cf5a02b3f09843f5a92ab0f04614e74aaa5a61b2121f715/peer_review/peer_review.md

@@ -0,0 +1,430 @@
+Peer Review File
+
+A 13-Million Turnover-Number Anionic Ir-Catalyst for a Selective Industrial Route to Chiral Nicotine
+
+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/.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The paper by Yin et al. describes the use of a double anionic iridium ate complex for the asymmetric hydrogenation of aromatic ketones and aromatic ketones containing nitrogen functionalities. The product alcohols where obtained with 99% ee, but the most spectacular aspect of this work are the extremely high turnover numbers and turnover frequencies. A mechanism is proposed that was supported by control experiments and with DFT calculations. The anionic nature of the catalyst leads to an increased hydricity of the iridium hydride and at the same time the carbonyl group is activated by coordination to the sodium on the alkoxy group of the ligand. These results are very interesting and will appeal to a broad readership.
+
+The synthetic aspects of the paper as well as the large-scale application are fine. However, I see some issues in the mechanistic part that will need further work.
+
+Looking at the DFT calculations it becomes clear that the transfer of hydride to the ketone is the rate-determining step in the red mechanism which is based on involvement of the sodium alkoxide. On the contrary, in the blue mechanism the reaction of the complex with hydrogen is the rate determining step. Thus, it would stand to reason to determine the order of the reaction in hydrogen in order to gain additional experimental evidence for this mechanism.
+
+Determining the reaction order in iridium is also necessary, as this will show if the iridium complex that is actually involved in the hydrogenation is dimeric or not. Thus, I suggest the authors do a full kinetic study to determine the order in hydrogen, iridium and substrate.
+
+The authors do show a hydrogen consumption curve in the supp info. However tehre is a blue as well as a red curve and teh authors do not explain what the difefrence is between the two.
+
+The effect of the amount of solvent is rather bizarre as the product is also a secondary alcohol. Are the authors 100% sure that this effect is not caused by a minor impurity in the isopropanol? Isopropanol can even contain peroxides that could well have caused this effect. It maybe worthwhile to repeat these experiments with rigorously purified isopropanol.
+The stability of this catalyst is indeed remarkable. Is it possible to reuse the spent catalyst after a first reaction?
+
+What is entirely lacking in the manuscript is a summary of the state-of-the-art of the use of anionic complexes in (asymmetric) hydrogenation. Pez already reported the use of anionic complexes in homogeneous hydrogenation of esters in 1981. Then there is of course the work of Fujita, of Rinaldo Poli, and I am sure there will be others as well.
+
+The interesting thing about the Pez work is that other later showed that the anionic complexes were readily protonated by the alcoholic solvent. Are the authors 100% sure that this is not also the case here. It would be interesting to measure a MS of the catalyst after or better during hydrogenation to see if the catalyst is still double anionic.
+
+Some minor points.
+
+Use of acronyms in the title is usually not allowed.
+
+I also do not see the added value of the use of this triple S. I would recommend to remove this entirely from the manuscript.
+
+I would also remove the free commercial for Nicotinell. Most journals do not allow this.
+
+After these major revisions, the paper can likely be published.
+
+Reviewer #2 (Remarks to the Author):
+
+In this manuscript, Zhang and coworkers reported a highly efficient asymmetric hydrogenation of ketones catalyzed by an unprecedented Ir-ate catalyst. One of the most impressive aspects is the more than 10 million TON. This is the highest number reported in the field to date, comparable to biocatalytic hydrogen transfer. The reasons for the high activity were analyzed and the ate type Ir complex was found to be the key factor to achieve the excellent performance. This is also an important discovery. On the other hand, this strategy and method have also been used for the efficient synthesis of important chiral compounds including nicotine, and 40 tons of nicotine have
+been produced, which is also a remarkable achievement. The paper is well prepared, including the relevant theoretical calculation part. Therefore, in view of the innovation and application value of this work, I recommend this work to be published in Nature Communications.
+
+In addition, some other literatures on highly efficient asymmetric hydrogenation of ketones catalyzed by different transition metals are recommended to be cited, such as: (Ni) Angew. Chem. Int. Ed. 2022, 61, e202115983; (Ir) Nat. Catal. 2020, 3, 621–627; (Mn) Angew. Chem. Int. Ed. 2019, 58, 4973–4977; (Ir) Org. Lett., 2018, 20, 6135-6139; (Pd) Angew. Chem. Int. Ed., 2013, 52, 11632-11636.
+
+Reviewer #3 (Remarks to the Author):
+
+This communication reports on the performance of an iridium catalyst for asymmetric hydrogenation. According to this work, this is already being used in an industrial process for the synthesis of nicotine. The work is closely related to prior work by the authors (cited as refs. 23, 41 and 42, where the last one looks like a pre-print of the present communication) and it is not entirely clear to me what the focus of this manuscript is meant to be - the catalyst development, the industrial process or the accompanying mechanistic postulate and its analysis. I'm thus wondering whether this is of sufficient novelty and significance to the field, and also whether a communication is the correct format - the ESI is extensive and, in my view, necessary to make the paper accessible, such that I would favour presentation as a full paper.
+
+The writing is not very clear, with distracting errors of grammar and expression. The abstract is not written in the right style, contains multiple references and some of the information is then repeated in the introduction. In addition, abbreviations are used extensively, affecting the clarity of the writing, and figures appear overloaded and are difficult to follow. These presentational issues distract from the science and will need to be addressed.
+
+The reported work is based on multiple experiments and calculations, but these have not been presented well or logically, making it look like a set of results has been cherry-picked for the communication, while the details have been buried in the ESI. For example, I wondered about the small energy differences calculated between the mono- and diprotonated versions of the catalyst and how these figures were balanced with the sodium base; this has fed into figure 4B, but the details/critical assessment of the results are hard to come by.
+Similarly, some tests with different numbers of explicit solvent molecules seem to have been performed and are shown in Figure S38, but I struggled to locate a discussion of the different models, how solvent molecules were placed and, again, a critical assessment of the results.
+
+I also wondered whether for this type of study, optimisation in the gas phase followed by calculations of energies in solvation, with a bigger basis set, would be appropriate - at the very least the authors would need to demonstrate that the structural effects are minimal (and I fear they would not be).
+
+Calculation results have mainly been presented in terms of energy profiles in the ESI, again with minimal commentary on what was done and why - it would be sensible to expand this into data tables, discussions and the critical evaluations of whether these results can be trusted. For ee's the energy differences are small, such that computational and conformational noise need to be assessed carefully and I find it difficult to fully assess whether the interpretations of these results are reliable.
+
+Overall, then, there is a lot of work here, but the presentation needs to be improved and the key messages need to be clarified - I'm not convinced this is the right format or indeed the right journal.
+REVIEWER REPORT
+COMMENTS TO AUTHOR:
+
+Reviewer 1: The paper by Yin et al. describes the use of a double anionic iridium ate complex for the asymmetric hydrogenation of aromatic ketones and aromatic ketones containing nitrogen functionalities. The product alcohols where obtained with 99% ee, but the most spectacular aspect of this work are the extremely high turnover numbers and turnover frequencies. A mechanism is proposed that was supported by control experiments and with DFT calculations. The anionic nature of the catalyst leads to an increased hydricity of the iridium hydride and at the same time the carbonyl group is activated by coordination to the sodium on the alkoxyl group of the ligand. These results are very interesting and will appeal to a broad readership.
+
+The synthetic aspects of the paper as well as the large-scale application are fine. However, I see some issues in the mechanistic part that will need further work.
+
+Response: Thanks for your remarks and kind recommendation and as well constructive suggestions. We have thoroughly revised the manuscript and supplementary information with additional reaction kinetics experiments in dihydrogen pressure and iridium concentration. All the corrections were highlighted in yellow color.
+
+Corrections suggested are as follows.
+Looking at the DFT calculations it becomes clear that the transfer of hydride to the ketone is the rate-determining step in the red mechanism which is based on involvement of the sodium alkoxide. On the contrary, in the blue mechanism the reaction of the complex with hydrogen is the rate determining step. Thus, it would stand to reason to determine the order of the reaction in hydrogen in order to gain additional experimental evidence for this mechanism.
+Response: Thanks for this important notification. We apologize for a numerical mistake: After careful examination of the original data, we find a mistake in use of D’ instead of D” as reference in calculating the relative Gibbs free energy of IIIb. The correct value should be -2.6 kcal mol^{-1} instead of -6.7 kcal mol^{-1} in the blue mechanism. Figure 5 (shown below) and Figure S26 have also been corrected. All the original energy data are summarized in Table S11-S14 and the cartesian coordinates of all optimized structures are listed in Section 10 in the supporting information. Therefore, the transfer of hydride to acetophenone is indeed the rate/enantioselectivity-determining step for both the red mechanism and the blue mechanism. We truly appreciate your careful review that have helped us to avoid this mistake! Thanks so much!!
+Fig. 5 Predicted Gibbs free energy profile for the asymmetric hydrogenation of acetophenone via the active anionic Ir-catalyst D through ONa/MH bifunctional (in red) and NNa/MH bifunctional (in blue) paths.
+
+Importantly, for the energetically preferred red mechanism, the Gibbs free energy barriers of the hydride transfer step and dihydrogen activation step are 6.2 and 4.7 kcal mol\(^{-1}\), respectively, that are of small differences. In line with this, reaction kinetics experiments show 0.5 order in dihydrogen pressure, implying that the activation of dihydrogen is partaken in the determination of the overall reaction kinetics. The data of the kinetics experiments were presented in Table S10 and Figure S32-39 with experimental details in SI 5.2. A conclusion of the reactions kinetics experiments is added in the main text as follow:
+
+"The comparable Gibbs free energy barriers of the hydride transfer step and dihydrogen activation step are also consistent with kinetic experiments (SI 5, Table S10 and Figs. S32-47) where 0.5 order in dihydrogen pressure and 1.5 order in iridium concentration were observed."
+
+Determining the reaction order in iridium is also necessary, as this will show if the iridium complex that is actually involved in the hydrogenation is dimeric or not. Thus, I suggest the authors do a full kinetic study to determine the order in hydrogen, iridium and substrate.
+
+Response: Reaction order in iridium is done. A 1.5 order in iridium was observed, implying monomeric iridium ate catalyst is operation under catalysis conditions. The data of the kinetics experiments were presented in Figure S40-47 with experimental details in SI 5.3. A conclusion of the reactions kinetics experiments is added in the main text as follow:
+
+"The comparable Gibbs free energy barriers of the hydride transfer step and dihydrogen activation step are also consistent with kinetic experiments (SI 5, Table S10
+and Figs. 32-47) where 0.5 order in dihydrogen pressure and 1.5 order in iridium concentration were observed.".
+
+Our gas-uptake facilities only allow maximum reaction pressure of 33 bar and therefore we measured the initial reaction rates referred to Klankermayer’s method for measurement of initial turnover frequencies (DOI: 10.1021/acscatal.9b04977). In typical hydrogenation reactions, substrate coordination is assumed quasi-equilibrium and the hydride-migration-insertion or reductive elimination step is the rate determination step. In either case, a first order in substrate concentration is generally reported for hydrogenation reactions (see refs: Chem. Commun., 2009, 7447-7464; ACS Catal., 9(8), 7535-7547). From DFT data, we did not find any new mechanism regarding reaction kinetics in substrate. We therefore did not perform reaction kinetics experiments in substrate concentration, which we hope does not affect the conclusion.
+
+The authors do show a hydrogen consumption curve in the supp info. However tehre is a blue as well as a red curve and teh authors do not explain what the difefrence is between the two.
+
+Response: Thanks for your remarks. To make the figure clear, legends are now added. The black line refers to the original reaction pressure observed at specific reaction intervals. The red line simulates the reaction pressure change over time. See an exam below:
+Slope = -25.2 bar/h
+
+<table>
+  <tr>
+    <th>Model</th>
+    <th>ExpDec3</th>
+  </tr>
+  <tr>
+    <td>Equation</td>
+    <td>y = A1*exp(x*t1) + A2*exp(x*t2) + A3*exp(x*t3) + y0</td>
+  </tr>
+  <tr>
+    <td>Pct.</td>
+    <td>Pressure</td>
+  </tr>
+  <tr>
+    <td>A1</td>
+    <td>27.2856 ± 0.57919</td>
+  </tr>
+  <tr>
+    <td>A2</td>
+    <td>18.02058 ± --</td>
+  </tr>
+  <tr>
+    <td>A3</td>
+    <td>2.12387 ± 20913.19765</td>
+  </tr>
+  <tr>
+    <td>t1</td>
+    <td>18.2865 ± --</td>
+  </tr>
+  <tr>
+    <td>t2</td>
+    <td>17.88737 ± 1.7064E7</td>
+  </tr>
+  <tr>
+    <td>t3</td>
+    <td>2.12833 ± --</td>
+  </tr>
+  <tr>
+    <td>R^2</td>
+    <td>0.99424</td>
+  </tr>
+  <tr>
+    <td>Adj. R-Square</td>
+    <td>0.98847</td>
+  </tr>
+</table>
+
+\[
+\frac{n_{sub}}{\Delta_p} = \frac{61.44\ mmol}{48.0\ bar} \approx 1.28\ mmol/bar\ (equation\ 1)
+\]
+
+\[
+m_n = -m_{\Delta_p}\ \frac{n_{sub}}{\Delta_p} = 25.2\ \frac{bar}{h} \cdot 1.28\ \frac{mmol}{bar} \approx 32.26\ \frac{mmol}{h}\ (equation\ 2)
+\]
+
+\[
+TOF_{ini} = \frac{m_n}{n_{cat}} = \frac{32.26\ \frac{mmol}{h}}{0.00004\ mmol} \approx 806,400\ h^{-1}\ (equation\ 3)
+\]
+
+Figure S1. Analysis of pressure drop curve for anionic Ir-catalyst catalyzed asymmetric hydrogenation of acetophenone (S1) at 80 bar of H$_2$*
+
+The effect of the amount of solvent is rather bizarre as the product is also a secondary alcohol. Are the authors 100% sure that this effect is not caused by a minor impurity in the isopropanol? Isopropanol can even contain peroxides that could well have caused this effect. It maybe worthwhile to repeat these experiments with rigorously purified isopropanol.
+
+Response: Thanks for your valuable concerns. The amounts of solvent effects on the catalysis were reperformed using rigorously distilled isopropanol, shown in supplementary information, table S5. A higher conversion was observed upon increasing the amount of solvent from 0 to 2 mL. Further increasing the amount of solvent to 4 mL did not give negative or positive effects, implying that some amount of
+solvent is required for dissolve both the substrate and the catalyst to obtain excellent reaction rates.
+
+The stability of this catalyst is indeed remarkable. Is it possible to reuse the spent catalyst after a first reaction?
+Response: It should be possible as you can see that our catalyst can work well even after 30 days during the 13 million turnover number experiments. After catalytic experiments, the product and solvent should be easily removed under inert atmosphere and as such the spent catalyst can be reused. Unfortunately, our set-ups cannot meet the requirements. In our real industrial productions, we use only up to 10 ppm catalyst. Calculations suggest low benefits for recycling the spent catalyst as huge energy are required to remake the volatiles. The cost due to catalyst is less than 1% according to our calculations.
+
+What is entirely lacking in the manuscript is a summary of the state-of-the-art of the use of anionic complexes in (asymmetric) hydrogenation. Pez already reported the use of anionic complexes in homogeneous hydrogenation of esters in 1981. Then there is of course the work of Fujita, of Rinaldo Poli, and I am sure there will be others as well.
+Response: Great suggestion! References are added (40-43). A summary of previous work of using anionic complexes in asymmetric hydrogenation is added in the third paragraph of Introduction as below: “Inspired by highly reactive anionic reductants and multidentate Noyori-type hydrogenation catalysts, we proposed the integration of the concepts of anionic complexes and multidentate ligands for developing ultra-efficient asymmetric hydrogenation catalysts with high selectivity, stability and reactivity preeminence (Fig. 1C). The characteristic anionic complexes bearing a formal negative charge can, in principle, enable high hydricity and accordingly high catalytic reaction rates, as evidenced from the seminal anionic metal hydride catalysts for hydrogenation reaction of carbonyl compounds by Pez, Poli, and others.”
+
+The interesting thing about the Pez work is that other later showed that the anionic complexes were readily protonated by the alcoholic solvent. Are the authors 100% sure that this is not also the case here. It would be interesting to measure a MS of the catalyst after or better during hydrogenation to see if the catalyst is still double anionic.
+Response: Excellent remarks. HRMS (Figure S18-19) of a solution of Ir-precatalyst and NaOtBu in isopropyl alcohol under 30 bar H2 showed exact mass of 765.1874 [M-2Na+3H]+ and 799.1490 [M-2Na+2H+Cl], corresponding to protonated and chlorinated Ir-ate catalysts, respectively, in the positive and negative region. Therefore, we believe that we do observe the MS of the anionic Ir-catalyst. Additionally, compared to Pez work, we think that the back protonation is unlikely as the pKa of our Ir-precatalyst is much lower than the alcohols.
+
+Use of acronyms in the title is usually not allowed.
+Response: Done! Title has been changed from “Discovery of an Ir-ate Catalyst for Ultra-efficient Asymmetric Hydrogenation of Ketones with 3S Character (Stable,
+Speed and Selectivity)” to “A 13-Million Turnover-Number Anionic Ir-Catalyst for a Selective Industrial Route to Chiral Nicotine”. All “3S” have been deleted and “AH” has been changed to asymmetric hydrogenation to make the reading easier for all readers.
+
+I also do not see the added value of the use of this triple S. I also do not see the added value of the use of this triple S.
+Response: Agreed and Corrected!
+
+I would also remove the free commercial for Nicotinell. Most journals do not allow this.
+Response: Thanks for your comments. We have removed the picture of Nicotinell from Fig. 3.
+
+Reviewer 2: In this manuscript, Zhang and coworkers reported a highly efficient asymmetric hydrogenation of ketones catalyzed by an unprecedented Ir-ate catalyst. One of the most impressive aspects is the more than 10 million TON. This is the highest number reported in the field to date, comparable to biocatalytic hydrogen transfer. The reasons for the high activity were analyzed and the ate type Ir complex was found to be the key factor to achieve the excellent performance. This is also an important discovery. On the other hand, this strategy and method have also been used for the efficient synthesis of important chiral compounds including nicotine, and 40 tons of nicotine have been produced, which is also a remarkable achievement. The paper is well prepared, including the relevant theoretical calculation part. Therefore, in view of the innovation and application value of this work, I recommend this work to be published in Nature Communications.
+In addition, some other literatures on highly efficient asymmetric hydrogenation of ketones catalyzed by different transition metals are recommended to be cited, such as: (Ni) Angew. Chem. Int. Ed. 2022, 61, e202115983; (Ir) Nat. Catal. 2020, 3, 621–627; (Mn) Angew. Chem. Int. Ed. 2019, 58, 4973–4977; (Ir) Org. Lett., 2018, 20, 6135-6139; (Pd) Angew. Chem. Int. Ed., 2013, 52, 11632-11636.
+Response: Thanks for your kind comments of our work. All the recommended references were cited, as reference 15 16, 25, 33, 34 in the manuscript, which helps to make the manuscript better.
+
+Reviewer 3: This communication reports on the performance of an iridium catalyst for asymmetric hydrogenation. According to this work, this is already being used in an industrial process for the synthesis of nicotine. The work is closely related to prior work by the authors (cited as refs. 23, 41 and 42, where the last one looks like a pre-print of the present communication, it is not) and it is not entirely clear to me what the focus of this manuscript is meant to be - the catalyst development, the industrial process or the
+accompanying mechanistic postulate and its analysis. I'm thus wondering whether this is of sufficient novelty and significance to the field, and also whether a communication is the correct format - the ESI is extensive and, in my view, necessary to make the paper accessible, such that I would favour presentation as a full paper.
+
+Response: Thanks for your critical comments and concerns, which help us to make clear the focus and novelty. In refs 23 and 41 (now refs 24 and 44), we reported ligand f-amphox based iridium catalyst for asymmetric hydrogenation of aryl-alkyl or alkyl-alkyl ketones up to 1,000,000 TON and >99% ee. In ref 42 (now 45), we reported ligand f-phamidol in asymmetric hydrogenation of acetophenone up to 1,000,000 TON and >99% ee. Ligand f-phamidol was found by serendipity, where f-amphox hydrolyzed during silicon chromatography to give f-phamidol.
+
+In this manuscript, we report an ultra-efficient anionic Ir-catalyst based on f-phamidol for asymmetric hydrogenation of (hetero)aryl-alkyl ketones up to 13 million TON and >99% ee. Based on the orders of improvement of TON up to 1,000,000 (compared to the known highest 10,000 TON) in asymmetric hydrogenation of pyridyl-alkyl ketone and an industrial route for manufacture of chiral Nicotine was presented. The ONa/MH bifunctional mechanism was also firstly presented here. Therefore, we hope the referee can find it reasonable to report this new work and recognize the sufficient novelty and significance to the field. To clarify our point, please see the comparison of the work below:
+
+![Reaction schemes and product structures for amphox and phamidol catalysis](page_370_682_1002_627.png)
+
+Ref. 24   Org. Let.  2016, 18, 2938.
+Ref. 44   Org. Chem. Front.  2017, 4, 555.
+Ref. 45   Green Synth. Catal.  2022, 3, 175.
+
+Previous work
+We fully agree with the reviewer that this manuscript should be a research article as commented. For the clarity, we have corrected the title and as well some parts of the main text and supplementary information. Title has been changed from “Discovery of an Ir-ate Catalyst for Ultra-efficient Asymmetric Hydrogenation of Ketones with 3S Character (Stable, Speed and Selectivity)” to “A 13-Million Turnover-Number Anionic Ir-Catalyst for a Selective Industrial Route to Chiral Nicotine”.
+
+The writing is not very clear, with distracting errors of grammar and expression. The abstract is not written in the right style, contains multiple references and some of the information is then repeated in the introduction. In addition, abbreviations are used extensively, affecting the clarity of the writing, and figures appear overloaded and are difficult to follow. These presentational issues distract from the science and will need to be addressed.
+
+Response: Thanks for your valuable and critical comments, which we fully agree with. We have double checked the grammar and expression as far as we can, and have carefully modified the title/abstract/figures and deleted unnecessary abbreviations that are used extensively.
+The Title has been changed from “Discovery of an Ir-ate Catalyst for Ultra-efficient Asymmetric Hydrogenation of Ketones with 3S Character (Stable, Speed and Selectivity)” to “A 13-Million Turnover-Number Anionic Ir-Catalyst for a Selective Industrial Route to Chiral Nicotine”.
+The abstract is rewritten as “The development of ultra-efficient hydrogenation catalysts for reduction of organic carbonyl compounds is critical for pharmaceuticals, agrochemicals and fine chemicals. However, manufacturing practical catalysts with high selectivity, stability and reactivity preeminence remains unsolved, calling for conceptual advancement. Herein, by integration of the concepts of multidentate ligation and anionic complex, we report the first ultra-efficient anionic Ir-catalyst for highly selective construction of chiral alcohols via asymmetric hydrogenation of (nitrogen-containing) ketones. The anionic catalyst features remarkable, biocatalysis-like efficacy of 99% ee (enantiomeric excess), 13,425,000 TON (turnover number) and 224 s^{-1} TOF (turnover frequency). Quantum chemical studies reveal a novel ONa/MH bifunctional mechanism of the Ir-catalyst. With this anionic Ir-catalyst, a selective industrial route to enantiopure nicotine at 500 kg batch scale has been established, providing 40 tons scale of product.”
+
+Following the reviewer’s comment, previous Fig. 2 has been split into Fig. 2 and Fig. 3, previous Fig. 3 is split into Fig. 4 and Fig. 5, and previous Fig. 4 is split into Fig. 6
+and Fig. 7. The Figures in the SI are also reorganized. The abbreviations such as 3S are removed from the manuscript.
+
+The reported work is based on multiple experiments and calculations, but these have not been presented well or logically, making it look like a set of results has been cherry-picked for the communication, while the details have been buried in the ESI. For example, I wondered about the small energy differences calculated between the mono- and diprotonated versions of the catalyst and how these figures were balanced with the sodium base; this has fed into figure 4B, but the details/critical assessment of the results are hard to come by.
+
+Response: Thanks for your valuable and critical comments in terms of the logic and the writing style. We have substantially reorganized the manuscript to improve the presentation and make the main message clear as highlighted in the manuscript and as well in the supplementary information.
+We have reorganized the SI into several section. For example, in SI 5, we mainly present the reaction kinetics studies of the anionic Ir-catalyst catalyzed asymmetric hydrogenation of acetophenone, including measurement of initial turnover frequencies, determination of the reaction order in hydrogenation pressure and iridium.
+
+![Formation of anionic Ir-catalyst from Ir-precatalyst. Calculated relative Gibbs free energies at 298.15 K are given in brackets.](page_186_670_1077_377.png)
+
+Fig. 4 Formation of anionic Ir-catalyst from Ir-precatalyst. Calculated relative Gibbs free energies at 298.15 K are given in brackets.
+Fig. S24. Relative Gibbs free energies (values in kcal mol\(^{-1}\) at 298.15 K) of possible active anionic Ir-catalysts formation under basic condition.
+
+As depicted in Fig. 3A (now Fig 4) and main text discussion in the manuscript, compared to deprotonation of NH function, OH deprotonation is highly energetically favorable, resulting in anionic Ir-catalysts C and D (Fig. S24). Due to the small energy differences between the mono- and deprotonated versions of the catalyst, we consider that Ir-ate catalysts C and D are likely under equilibrium at the experimental condition (excess base). Such Ir-ate complexes were characterized by HRMS, in-situ high-pressure (HP) NMR and ATR-IR spectroscopy as well as DFT calculations. Thus, Ir catalysts A, C and D are all explored completely for acetophenone hydrogenation. Compared to A and anionic Ir-catalyst C (Tables S7), anionic Ir- catalyst D provides energetically favorable pathways involving NNa/MH and ONa/MH bifunctional mechanisms, as shown in Fig. 6 and S26-27. We have deleted the discussion of the role of alkali cations on the hydride transfer step in the manuscript for more clear presentation of the theoretical work.
+
+Similarly, some tests with different numbers of explicit solvent molecules seem to have been performed and are shown in Figure S38, but I struggled to locate a discussion of the different models, how solvent molecules were placed and, again, a critical assessment of the results.
+
+Response: Thanks for your concerns and critical comments. We share the same concerns. “To simplify the calculation of reaction mechanism, we didn’t consider the solvated Na+ in the model, as with other similar theoretical works (J. Am. Chem. Soc. 2014, 136, 3505–3521). However, in real condition, the Na+ may be surrounded by the solvent iPrOH. Here, considering the possible interactions, including hydrogen bonding
+and coordination interactions of solvent molecules with sodium cation, ketone substrate and ligand, two and four iPrOH molecules were added involving sufficient interactions between molecules to test the effect on the hydride transfer step. Indeed, the solvated Na+ has lower polarizability toward the substrate acetophenone, resulting in a relatively higher energy barrier of the hydride transfer step."
+We have added the above detailed description below Fig. S28 (old version in Fig. S38).
+
+<table>
+  <tr>
+    <th colspan="4">Solvent effect</th>
+  </tr>
+  <tr>
+    <th>species</th>
+    <th>II</th>
+    <th>TS1</th>
+    <th>ΔG<sub>1</sub></th>
+  </tr>
+  <tr>
+    <td>0 iPrOH</td>
+    <td>0.2</td>
+    <td>6.2</td>
+    <td>6.2</td>
+  </tr>
+  <tr>
+    <td>2 iPrOH</td>
+    <td>-18.4</td>
+    <td>-6.7</td>
+    <td>11.7</td>
+  </tr>
+  <tr>
+    <td>4 iPrOH</td>
+    <td>0.8</td>
+    <td>12.9</td>
+    <td>12.9</td>
+  </tr>
+</table>
+
+Model D
+
+Fig. S28 Structures of the transition states of the hydride transfer step upon the active anionic Ir-catalyst D with explicit solvent molecules and the Gibbs free energies
+
+I also wondered whether for this type of study, optimisation in the gas phase followed by calculations of energies in solvation, with a bigger basis set, would be appropriate - at the very least the authors would need to demonstrate that the structural effects are minimal (and I fear they would not be).
+Response: We fully agree with this concern. Unfortunately, as numerous optimized structures and corresponding energies need to be calculated in this work, calculations under large basis sets and solvation model for all the cases are not impossible, but are almost impractical because of the time consuming. In fact, we previously also tested the effects of basis set and implicit solvent model on structural optimization and the results showed that the bigger basis set or implicit solvent model gave small effects on optimized structures, implying that the previous results are not fully unreasonable. To show the difference, we have now listed the results below. Table S1 has now been added in Section 2.8 Computational methods of SI.
+
+<table>
+  <tr>
+    <th rowspan="2">Bond Length</th>
+    <th colspan="5">Table S1. Selected bond length (unit: Å) of species A under different basis sets</th>
+  </tr>
+  <tr>
+    <th>SDD-6-31G*</th>
+    <th>SDD-6-31+G*</th>
+    <th>def2-TZVP</th>
+    <th>SDD-6-31G* (SMD)</th>
+    <th>def2-TZVP-6-311++G(d,p) (SMD)</th>
+  </tr>
+  <tr>
+    <td>Ir-H5</td>
+    <td>1.692</td>
+    <td>1.692</td>
+    <td>1.685</td>
+    <td>1.693</td>
+    <td>1.684</td>
+  </tr>
+  <tr>
+    <td>Ir-H6</td>
+    <td>1.682</td>
+    <td>1.683</td>
+    <td>1.679</td>
+    <td>1.690</td>
+    <td>1.688</td>
+  </tr>
+  <tr>
+    <td>Ir-O</td>
+    <td>2.179</td>
+    <td>2.180</td>
+    <td>2.164</td>
+    <td>2.182</td>
+    <td>2.175</td>
+  </tr>
+</table>
+<table>
+  <tr>
+    <th></th>
+    <th>Ir-N3</th>
+    <th>Ir-N4</th>
+    <th>Ir-P</th>
+    <th></th>
+    <th></th>
+  </tr>
+  <tr>
+    <td></td>
+    <td>2.089</td>
+    <td>2.051</td>
+    <td>2.266</td>
+    <td>2.087</td>
+    <td>2.054</td>
+    <td>2.266</td>
+    <td>2.079</td>
+    <td>2.244</td>
+    <td>2.093</td>
+    <td>2.066</td>
+    <td>2.282</td>
+    <td>2.082</td>
+    <td>2.063</td>
+    <td>2.275</td>
+  </tr>
+</table>
+
+As the reviewer knows, studies that perform structural optimization in the gas phase followed by calculations of energies with a bigger basis set and solvation correction by implicit solvent model have been widely used ad hoc. Many times, reasonable results can be provided by such approximate approach, as shown in many research works such as J. Am. Chem. Soc. 2021, 143, 3571–3582; Nat. Chem., 2022, 14, 1233–1241; J. Am. Chem. Soc. 2023, 145(4), 2305–2314; J. Am. Chem. Soc. 2023, 145(4), 2207–2218. We therefore hope the reviewer can agree that the current approach is not ideal, but practically acceptable.
+
+Calculation results have mainly been presented in terms of energy profiles in the ESI, again with minimal commentary on what was done and why - it would be sensible to expand this into data tables, discussions and the critical evaluations of whether these results can be trusted. For ee's the energy differences are small, such that computational and conformational noise need to be assessed carefully and I find it difficult to fully assess whether the interpretations of these results are reliable.
+
+Response: Thanks for your valuable remarks. Following the suggestions, to make the data better organized and clearer, we have reorganized the related Sections and Figures in SI, and added the detailed description following the corresponding Figures. We have summarized the Gibbs free energy barriers of hydride transfer step and dihydrogen activation step on Ir-catalyst A and anionic Ir-catalyst C and D, as shown in Table S7, which can clearly show that compared to A and anionic Ir-catalyst C, anionic Ir-catalyst D provides energetically favorable pathways involving NNa/MH and ONa/MH bifunctional mechanisms.
+
+Table S7. The Gibbs free energy barriers for Ir-catalyst A and anionic Ir-catalyst C, D.
+
+<table>
+  <tr>
+    <th rowspan="2">path</th>
+    <th colspan="2">A</th>
+    <th colspan="2">C</th>
+    <th colspan="2">D</th>
+  </tr>
+  <tr>
+    <th>ΔG₁</th>
+    <th>ΔG₂</th>
+    <th>ΔG₁</th>
+    <th>ΔG₂</th>
+    <th>ΔG₁</th>
+    <th>ΔG₂</th>
+  </tr>
+  <tr>
+    <td>OX/MH-R</td>
+    <td>12.1</td>
+    <td>33.5</td>
+    <td>7.8</td>
+    <td>6.4</td>
+    <td>6.2</td>
+    <td>4.7</td>
+  </tr>
+  <tr>
+    <td>OX/MH-S</td>
+    <td>14.9</td>
+    <td>33.5</td>
+    <td>7.9</td>
+    <td>8.4</td>
+    <td>7.6</td>
+    <td>5.5</td>
+  </tr>
+  <tr>
+    <td>NX/MH-R</td>
+    <td>10.5</td>
+    <td>24.5</td>
+    <td>10.7</td>
+    <td>17.1</td>
+    <td>10.8</td>
+    <td>9.5</td>
+  </tr>
+  <tr>
+    <td>NX/MH-S</td>
+    <td>9.1</td>
+    <td>24.5</td>
+    <td>10.6</td>
+    <td>17.1</td>
+    <td>10.6</td>
+    <td>9.3</td>
+  </tr>
+</table>
+
+ΔG₁ and ΔG₂ denote the free energy barrier of the hydride transfer step and the dihydrogen addition step at 298.15K, respectively. X denotes Na or H in corresponding Ir-catalyst A, C and D. Unit: kcal mol⁻¹.
+
+General theoretical methods have been used to calculate the catalytic mechanisms as discussed in last question, and the reaction mechanisms, especially the rate-determining
+steps, are decided by the relative Gibbs free energy barriers, where the computational and conformational noise would be cancelled to a large extent, as the systematic errors would be more or less removed when considering the energy difference. Thus, we believe that the computational results of reaction mechanisms are not unreasonable. Considering the complexed solution environment and computational errors, it is difficult to exactly obtain the ee value but qualitative description of the selectivity, which is accord with the experimental results.
+
+Overall, then, there is a lot of work here, but the presentation needs to be improved and the key messages need to be clarified - I'm not convinced this is the right format or indeed the right journal.
+
+Response: Thanks for your concerns and remarks. We have carefully reorganized the manuscript to improve the presentation and make the main messages clear. Detailed modifications can be found in the highlighted manuscript. We hope the reviewer can agree that the research effort and the excellent TON result of this work is worthwhile for an multidisciplinary journal like Nat. Commun.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The paper by Zhang and co-workers has much improved. However, I am not satisfied with the kinetics calculations and the discussion of the outcome of the kinetics in relation to the DFT calculations.
+
+The authors discovered a mistake in their calculations and claim that now in both the blue and the red mechanism the reaction with hydrogen is the rate determining. However, taking a closer look at the red curve it now seems that the highest barrier is in the alcohol decomplexation step. This suggest that the reaction is inhibited by the product, which would be quite easy to prove by comparing the rate of a reaction where some product has been added right from the start with a reaction where this is not the case. The barriers of the three basic steps are rather close and so it may be too close to call. But simply stating that the rate determining step is the hydrogen addition is not correct based on this diagram.
+
+The authors have performed the kinetics, unfortunately by measuring the decline of the hydrogen pressure, which is a rather inaccurate way of doing this. I am not suggesting that the authors should redo the experiments under constant pressure. This is a lot of extra work and will not give much more information. However, the authors have made a mistake in the calculation of the order in hydrogen, which is perhaps caused by the fact that the Ln of the hydrogen pressure was the Y-axis and the Ln of the TOF was the x-axis, instead of the other way around. Thus, whereas the authors claim the order in hydrogen is 0.5 it is in reality 2.0, which was already obvious from the raw data. The order in iridium is 1.5. These are not simple kinetics and this does not quite fit the picture that is painted based on the DFT calculations. It will not be easy to get to the bottom of this story unless the calculations contain more mistakes. It will obviously need a lot more work. My proposal would be that the authors correct the obvious mistake with the hydrogen order and admit that the kinetic data do not quite fit the proposed mechanism. They could end with the statement that more research is necessary to be able to propose the definite mechanism.
+
+I personally find the hydrogenation results interesting enough to publish the paper now after these minor changes.
+
+Reviewer #3 (Remarks to the Author):
+The calculations on their own are not of the best quality but the authors make a sensible case that, in conjunction with the experimental data and practical considerations, they contribute sufficiently to the insights generated. So, while I maintain that this is not the best format for this work (a full paper in a more specialist journal seems more appropriate), I concede that it has been improved sufficiently to proceed to publication. I would suggest capturing additional parts of the reply to reviewers in the ESI, e.g. the comments around the geometry effects of a better computational approach (despite my misgivings about the length), but the paper is now more balanced.
+REVIEWER REPORT
+COMMENTS TO AUTHOR:
+
+Reviewer 1:
+The paper by Zhang and co-workers has much improved. However, I am not satisfied with the kinetics calculations and the discussion of the outcome of the kinetics in relation to the DFT calculations.
+
+Response: Thanks for your critical remarks and kindly recommendation. We have revised the manuscript and supplementary information to our best. All the corrections were highlighted in yellow color. We must admit that some dispute exists between kinetic experiments and DFT calculations which deserve more works using advanced techniques to gain future insights.
+
+The authors discovered a mistake in their calculations and claim that now in both the blue and the red mechanism the reaction with hydrogen is the rate determining. However, taking a closer look at the red curve it now seems that the highest barrier is in the alcohol decomplexation step. This suggest that the reaction is inhibited by the product, which would be quite easy to prove by comparing the rate of a reaction where some product has been added right from the start with a reaction where this is not the case. The barriers of the three basic steps are rather close and so it may be too close to call. But simply stating that the rate determining step is the hydrogen addition is not correct based on this diagram.
+
+Response: Agreed. Due to the close barriers of the three steps, the stating of the rate determining step is trick. We have corrected ‘thus facilitate hydride transfer in the rate and selectivity determining step from’ to ‘thus facilitate hydride transfer in the selectivity determining step from’. To some extent, we did not expect substrate inhibition as the catalysis reaction was performed excellent well in isopropanol.
+
+The authors have performed the kinetics, unfortunately by measuring the decline of the hydrogen pressure, which is a rather inaccurate way of doing this. I am not suggesting that the authors should redo the experiments under constant pressure. This is a lot of extra work and will not give much more information. However, the authors have made a mistake in the calculation of the order in hydrogen, which is perhaps caused by the fact that the Ln of the hydrogen pressure was the Y-axis and the Ln of the TOF was the x-axis, instead of the other way around. Thus, whereas the authors claim the order in hydrogen is 0.5 it is in reality 2.0, which was already obvious from the raw data. The order in iridium is 1.5. These are not simple kinetics and this does not quite fit the picture that is painted based on the DFT calculations. It will not be easy to get to the bottom of this story unless the calculations contain more mistakes. It will obviously need a lot more work. My proposal would be that the authors correct the obvious mistake with the hydrogen order and admit that the kinetic data do not quite fit the proposed mechanism. They could end with the statement that more research is necessary to be able to propose the definite mechanism.
+Response: Agreed. We exchanged X-Y axis of the figure S39. The reaction order in dihydrogen pressure was corrected to 1.9. Given the unusual kinetic orders in both dihydrogen pressure and iridium catalyst, and as well even complexed DFT calculations when considering the solvation, hydrogen bonding, polarization etc. effects, we fully agree with the reviewer that it is rather tricky to give a conclusive mechanism. Corrected as below:
+
+“Unexpectedly, 1.9 order in dihydrogen pressure and 1.5 order in iridium concentration were observed in kinetic experiments (SI 5, Table S10 and Figs. 32-47). Given the unusual kinetic orders in both dihydrogen pressure and iridium catalyst, and as well even complexed DFT calculations when considering the solvation, hydrogen bonding, polarization etc. effects, more research is necessary to be able to propose the definite mechanism.”
+
+I personally find the hydrogenation results interesting enough to publish the paper now after these minor changes.
+
+Response: Thanks. We have made these minor changes.
+
+Reviewer 3: The calculations on their own are not of the best quality but the authors make a sensible case that, in conjunction with the experimental data and practical considerations, they contribute sufficiently to the insights generated. So, while I maintain that this is not the best format for this work (a full paper in a more specialist journal seems more appropriate), I concede that it has been improved sufficiently to proceed to publication. I would suggest capturing additional parts of the reply to reviewers in the ESI, e.g. the comments around the geometry effects of a better computational approach (despite my misgivings about the length), but the paper is now more balanced.
+
+Response: Thanks. We have added additional information regarding the computational approach in the ESI 2.7 as “Studies that perform structural optimization in the gas phase followed by calculations of energies with a bigger basis set and solvation correction by implicit solvent model have been widely used ad hoc. Many times, reasonable results can be provided by such approximate approach, as shown in many research works such as J. Am. Chem. Soc. 2021, 143, 3571–3582; Nat. Chem., 2022, 14, 1233–1241; J. Am. Chem. Soc. 2023, 145(4), 2305–2314; J. Am. Chem. Soc. 2023, 145(4), 2207–2218. Here, we also tested the effects of basis set and implicit solvent model on structural optimization and the results showed that the bigger basis set or implicit solvent model gave small effects on optimized structures, implying that the calculated results in this work are not fully unreasonable. To show the difference, we have now listed the results below.”.

0b3c54e3a936d3159cf5a02b3f09843f5a92ab0f04614e74aaa5a61b2121f715/preprint/preprint.md

@@ -0,0 +1,228 @@
+Discovery of an Ir-ate Catalyst for Ultra-efficient Asymmetric Hydrogenation of Ketones with 3S Character (Stable, Speed and Selectivity)
+
+Congcong Yin
+Southern University of Science and Technology
+
+Yafei Jiang
+Southern University of Science and Technology
+
+Fanping Huang
+Southern University of Science and Technology
+
+Cong-Qiao Xu
+Southern University of Science and Technology https://orcid.org/0000-0003-4593-3288
+
+Yingmin Pan
+Southern University of Science and Technology
+
+Shuang Gao
+Southern University of Science and Technology
+
+Gen-Qiang Chen
+Southern University of Science and Technology https://orcid.org/0000-0003-2276-6800
+
+Xiaobing Ding
+Shenzhen Catalys Technology Co., Ltd
+
+Qiwei Lang
+Shenzhen Catalys Technology Co., Ltd
+
+Shao-Tao Bai
+Southern University of Science and Technology
+
+Jun Li
+Tsinghua University
+
+Xumu Zhang (zhangxm@sustech.edu.cn)
+Southern University of Science and Technology https://orcid.org/0000-0001-5700-0608
+
+Article
+
+Keywords:
+
+Posted Date: November 21st, 2022
+
+DOI: https://doi.org/10.21203/rs.3.rs-2214819/v1
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: Yes there is potential Competing Interest. X.D., S.G. and Q.L. are inventors on patents (WO 2021/212880 A1, EP 3 925 955 A1, CN 113527187 A, US 2022/0089564 A1, CN 202210771902.X and 202111097411.3), held and submitted by Shenzhen Catalys Technology Co., Ltd.
+
+Version of Record: A version of this preprint was published at Nature Communications on June 22nd, 2023. See the published version at https://doi.org/10.1038/s41467-023-39375-8.
+Discovery of an Ir-ate Catalyst for Ultra-efficient Asymmetric Hydrogenation of Ketones with 3S Character (Stable, Speed and Selectivity)
+
+Congcong Yin,1† Ya-Fei Jiang,2† Fanping Huang,1 Cong-Qiao Xu,2 Yingmin Pan,1 Shuang Gao,1 Gen-Qiang Chen,3 Xiaobing Ding,4 Qiwei Lang, *,4 Shao-Tao Bai, *,3 Jun Li, *,2,5 and Xumu Zhang *,1
+
+1 Department of Chemistry and Shenzhen Grubbs Institute, Southern University of Science and Technology, Shenzhen 518055, China
+2 Department of Chemistry and Guangdong Provincial Key Laboratory of Catalytic Chemistry, Southern University of Science and Technology, Shenzhen 518055, China
+3 Academy for Advanced Interdisciplinary Studies and Department of Chemistry, Southern University of Science and Technology, Shenzhen 518055, China
+4 Shenzhen Catalys Technology Co., Ltd, Shenzhen 518100, China
+5 Department of Chemistry and Engineering Research Center of Advanced Rare-Earth Materials of Ministry of Education, Tsinghua University, Beijing 100084, China
+*Corresponding authors. X.Z.: zhangxm@sustech.edu.cn; J.L.: junli@tsinghua.edu.cn; S.T.B.: baist@mail.sustech.edu.cn; Q.L.: qwlang@catalys.com.cn
+† These authors contributed equally to this work.
+
+Abstract:
+
+Reduction of organocarbonyl compounds occupies a large space in organic synthesis and mass production of fuels, chemicals and materials. 1–9 Instead of using stoichiometric reduction agents, the development of ultra-efficient homogeneous hydrogenation catalysts using green dihydrogen is one of the foremost challenges in the transition from chiral resolution toward selective production of chiral entities, as evidenced from the growing capacity and quantity of production plants. 10,11 However, despite useful enantioselectivities (>98% ee) are
+often obtained in asymmetric hydrogenation (AH) of ketones, the celebrated Noyori-type catalysts are currently limited in few million turnover numbers (TONs) and a hundred turnover frequencies (TOFs).12-24 Challenges especially remain for nitrogen-containing ketones that are relevant to construction of high-value bioactive compounds where at most 10,000 TONs are reported. Here, by integration of the concepts of multidentate ligation and ate-type complex, we report the first ultra-efficient Ir-ate catalyst for highly selective construction of chiral alcohols via AH of ketones with remarkable, biocatalysis-like 3S capacity of >99% ee (enantiomeric excess) selectivity, 13,425,000 turnover number (TON) and 253 s^{-1} turnover frequency (TOF). With this Ir-ate catalyst a selective industrial route to chiral nicotine at 500 kg batch scale has been established, already providing 40 tons of product. Mechanistic studies reveal a novel ONa/MH bifunctional mechanism of the Ir-ate catalyst. Such concept is yet to be explored and may have a major impact in applied homogeneous catalysis.
+
+Traditional stoichiometric reduction agents, such as NaBH_4 and LiAlH_4, are highly reactive ate (short for metalate) compounds compared to their predecessors, viz. neutral B_2H_6 and HAl(iBu)_2 for reduction of organocarbonyl compounds (Fig. 1A).11 However, these explosive reagents require tedious storage and usage protocols yet generating significant amounts of waste. Wilkinson and Crabtree introduced transition metal hydrogenation catalysts that are either neutral or cationic compounds for reduction catalysis using safe and green dihydrogen (Fig. 1A).25-34 Unfortunately, based on either
+monodentate or bidentate ligands, these complexes are generally less efficient for mass production due to its coordination unstable nature and facile decomposition upon contacting trace amount of oxygen or other impurities.
+
+A Reduction of carbonyl compounds: stoichiometric reduction agents and conceptual hydrogenation catalysts
+
+![Diagram showing reduction of carbonyl compounds and stoichiometric reduction agents and conceptual hydrogenation catalysts](page_324_186_1097_120.png)
+
+B State-of-the-art Noyori-type Catalysts (outer-sphere mechanism) for Asymmetric Hydrogenation of Ketones
+
+![Three molecular structures of state-of-the-art Noyori-type catalysts for asymmetric hydrogenation of ketones](page_324_355_1097_120.png)
+
+Noyori (1998) 80% ee, 2,400,000 TON, TOF 63 s^{-1}
+Zhou (2011) 98% ee, 4,550,000 TON
+Zhang (2017) >99% ee, 1,000,000 TON
+
+C Conceptually advanced 3S tetra-Ir-ate catalyst for Asymmetric Hydrogenation of Ketones
+
+![Diagram showing conceptually advanced 3S tetra-Ir-ate catalyst for asymmetric hydrogenation of ketones](page_324_524_1097_120.png)
+
+orders of improvement of TONs
+first integration of ate & multidentate concepts
+Novel ONa/MH bifunctional mechanism
+Ir-ate catalyst
+up to >99% ee
+13,425,000 TON
+253 s^{-1} TOF
+Chiral alcohols
+Nicotine (40 tons)
+✔ Selectivity
+✔ Stability
+✔ Speed
+
+Fig. 1 Reduction of organocarbonyl compounds affording high-value chemicals. A) Stoichiometric ate reduction reagents and conceptual hydrogenation catalysts. B) The state-of-the-art neutral Noyori-type catalysts for Asymmetric Hydrogenation (AH) of ketones. C) Our conceptually advanced 3S tetra-Ir-ate catalyst for ultra-efficient AH of ketones. ee: enantiomeric excess.
+
+Compared to the representative hydrogenation catalysts \(^{28-34}\) with inner-sphere mechanism, catalysts \(^{35,36}\) of the outer-sphere mechanism can avoid the contact of substrates to the metal center, thus making it possible to design highly stable,
+enantioselective and coordination saturated hydrogenation catalysts (Fig. 1A-B). Indeed, Noyori-type catalysts, i.e. Ru(bisphos)(diamine) system by Noyori \(^{12,37,38}\), Ir-PNN-complex by Zhou \(^{21-23}\) and others \(^{15-20,39}\) operating via NH/MH\(^{38,39}\) bifunction mechanism are effective ketone asymmetric reducing catalysts (Figs. 1B, S2, ST 3.1 and Table S1 of the Supplementary Information (SI)). Exceptionally, Zhou’s tridentate ligand of Ir-PNN-catalyst provides the record-high TONs of 4,550,000 at 98% ee (enantiomeric excess) selectivity. However, despite useful enantioselectivities (>98% ee) are often obtained, the celebrated Noyori-type catalysts are currently limited in few million turnover numbers (TONs) and a hundred turnover frequencies (TOFs) for production of high-value chiral alcohols and their derivatives.\(^{24}\) Challenges especially remain for nitrogen-containing ketones that are relevant to construction of high-value bioactive compounds where at most 10,000 TONs are reported (Table S2 and Fig. S3).
+
+Given the increasing capital investments, operational costs and sustainability requirements, to obtain ultra-efficient 3S (selectivity, stability and reaction speed) catalysts achieving 10-million TONs and biocatalysis-like reaction turnover frequencies (TOFs), beyond the state-of-the-art NH/MH bifunction catalysts, is highly important and has been the holy grail of the homogeneous transition metal catalysis (Fig. S1). \(^{10,13,14}\) Inspired by highly reactive ate reduction agents and multidentate Noyori-type hydrogenation catalysts, we envisioned the integration of the concepts of ate compounds and multidentate ligands for developing 3S preeminence asymmetric hydrogenation (AH) catalysts (Fig. 1C). The characteristic ate complexes bearing a formal negative charge can enable high hydricity \(^{40}\) and accordingly high catalytic
+reaction speed. The multidentate ligands can help to stabilize the metal center through coordinative saturated 18 electron complexes and form well-defined chiral environment for converting specific substrates and give rise to highly stable and selective catalysts.
+
+We demonstrate a tetradentate PNNO ligand-bearing Ir-ate catalyst that provides unprecedented up to 13 million turnover-numbers (TONs), hundreds of TOFs per second at ultra-high selectivity (>99% ee, 13,425,000 TON and 253 s\(^{-1}\) TOF, Fig 1C, and Tables S1-2) comparable to biocatalysts. With this catalyst, even ketones with awkward coordinating basic nitrogen have been hydrogenated at a million TON and >99% ee selectivity. We have thus established a selective industrial route to enantiomeric pure nicotine, already providing 40 tons of product. Compared to traditional NH/MH bifunction catalysts, the Ir-ate catalyst based on a simple and easily fabricated ligand f-phamidol at kilograms-scale demonstrates significantly improved hydricity and a novel ONa/MH bifunctional mechanism.
+Fig. 2 AH of ketones using Ir/phamidol catalyst. (A) Laboratory scale AH of representative benchmark acetophenone and nitrogen-containing aromatic ketones. (B) Industry scale AH of nitrogen-containing ketone as a selective route to chiral nicotine with schematic of our process displayed in (C).
+
+Initially, on the basis of our previous work \(^{23,41,42}\), iridium catalysts based on tetradebate PNNO ligand f-phamidol was examined at a substrate/catalyst ratio of 2,000,000 using 80 mmol benchmark acetophenone at ambient temperature (Tables S3-5). All the base gave exceptionally high enantioselectivities >99% ee, whereas the highly basic NaOtBu that leads to 99% conversions in 16 h (corresponding to 1,980,000
+TONs) is superior to the others with variation of either the alkali ions or anionic counterions. The addition of suitable amounts of solvent is important for achieving both high enantioselectivity and TON. Under the optimal conditions, high turnover experiments with a variation of the substrate/catalyst ratio from 5,000,000 to 15,380,000 were carried out. Remarkably, a steady increasing TONs from 4,835,000 to 11,535,000 was observed at excellent enantioselectivities of 99% ee and excellent conversions of 75-97%, indicating super-stable, durable and enantioselective Ir/f-phamidol catalyst at the highest loading of substrate (vide infra, Fig. 1A, Table S5). Upon increasing the substrate amount to 800 mmol at 100 bar H₂, the highest TON of 13,425,000 together with 89.5% conversion and 99% ee were observed in a 30-days’ reaction, implying outstanding averaged production rate of \( 2,846 \ \mathrm{kg}_{\text{product}} (\mathrm{kg}_{\text{precatalyst}})^{-1} \ \mathrm{h}^{-1} \). The initial TOFs were calculated based on a pressure-drop curve (Figs. S4-5). An ultra-high initial TOF of 253 s⁻¹ was recorded, which is close to the biocatalytic efficiency of (de)hydrogenase⁴³-⁴⁵.
+
+Given the successful application of the Ir/f-phamidol catalyst for AH of benchmark acetophenone, we further examined its efficiency in a more challenging conversion of nitrogen containing ketones as well as in industrial construction of enantiomeric pure nicotine (Fig. 2, S3, Table S2, SM 2.5). Ketone **S2** containing an amide function proceeds smoothly to afford the desired chiral compound at gram-scale in >99% ee, 97% conversion and 970,000 TONs. Remarkably, even though ketone **S3** contains both an amide function and a pyridine function that often results in catalysts deactivation, our
+catalyst still gave exceptionally high efficiency (>99% conv., >99% ee and 1,000,000 TONs) at gram scale.
+
+Based on these data, the Ir/f-phamidol catalyst was investigated in a selective industrial route to enantiomeric pure nicotine from easily available ketone S4. Nicotine, generally isolated from tobacco, is one of the most important bioactive natural products with estimated consumption of > 1000 tons per year with the tendency of steady increasing. After preliminary testing in laboratory scale (two routes, SM 2.6-7), a 2000 L continuously stirred tank reactor was used for the crucial AH, shown in Fig. 2B-C and SM 2.8. Given economy and safety concerns, 26 bar dihydrogen pressure and much cheaper KOH instead of NaOtBu were applied based on conditions optimization (Table S3). Stepwise scale up the feedstock from 270 to 500 kilograms at substrate/catalyst ratio of 60,000 gave full conversions, 98.9% ee in 2 h (Fig. 2C, entries 1-2). Increasing the substrate/catalyst ratio to 80,000 and 100,000 at 500 kg scale needs 3h and 4.5 h, respectively, and the reaction reached 99.6-99.7% conversion with 98.7-98.8% ee (Fig. 4C, entries 3 and 4). To our surprise, a substrate/catalyst ratio of 100,000 at 500 kg scale using recycled solvents can be operated smoothly, giving rise to a key chiral alcohol intermediate (99.7% conversions, 98.7% ee, 100,000 TON, with catalyst load 0.045 g kg\(^{-1}\) product and space-time-yield 55.6 g L\(^{-1}\) h\(^{-1}\)) for production of nicotine already 40 tons with 99% ee.
+Fig. 3 Characterization of Ir/f-phamidol catalyst. (A) Formation of Ir-ate catalyst from Ir-precatalyst. Calculated relative Gibbs free energies at 298.1 K are given in brackets.
+
+(B) Performance of modular modified ligands in comparison with f-phamidol.
+
+To understand the nature of the extremely high efficacy and the reaction mechanism, we performed detailed characterization of the Ir/f-phamidol catalyst via a combination of experimental techniques and quantum-theoretical modelling based on density functional theory (DFT, the theoretical details are given in the SI). Upon mixing f-phamidol with [Ir(COD)Cl]₂, monochloride dihydride iridium complexes were identified as the Ir-precatalyst (Fig. 3A, S6-18, Table S12), based on the evidences from HRMS, NMR, ATR-IR, Raman, XRD and DFT calculations. Briefly, DFT calculations, NMR and ATR-IR spectroscopy confirmed that the formation of mixtures of Ir-
+precatalysts with cis-configuration of hydride and carbonyl binding to Ir-metal (CO-bind cis) is slightly favorable. Two cis-hydrides instead of trans-hydrides are evidenced by ATR-IR analysis of the dry powder at 2229 and 2127 cm^{-1} and DFT modeling (Figs. S9-10). The evidence of amide-carbonyl instead of NH coordination to Ir-metal was supported by the red shifts up to 22 cm^{-1} of the amide-carbonyl group. Upon deprotonation of the alcohol donor of the Ir-precatalyst, favorable tetracoordinated dihydride Ir-ate complexes were formed via likely intermediate neutral complex A (Fig. 3A, S19-26 and Table S12). Compared to deprotonation of NH function, OH deprotonation is highly energetically favorable, resulting in Ir-ate catalysts C and D that are likely under equilibrium at the experimental condition. Such Ir-ate complexes were characterized by HRMS, in-situ high-pressure (HP) NMR and ATR-IR spectroscopy and DFT calculations. HRMS of a solution of Ir-precatalyst and NaOtBu in isopropyl alcohol under 30 bar H_2 showed exact mass of 765.1874 [M-2Na+3H]^+ and 799.1490 [M-2Na+2H+Cl]^-, corresponding to protonated and chlorinated Ir-ate catalysts, respectively, in the positive and negative region. The existence of amide-N instead of amide-carbonyl coordination to Ir-metal was supported by blue shifts observed by in-situ high pressure (HP) ATR-IR experiments and DFT calculated infrared spectra (Figs. S24-25).
+
+To confirm the crucial role of tetradentate PNNO ligand for the formation of ultra-efficient Ir-ate catalyst, we prepared slightly modified ligands for AH of acetophenone as references (Fig. 3B, S27). Ligand f-phamidol-N-Me with NH function being methylated gave a maximum 100,000 TON and 95% ee. Ligand f-phamidol-O-Me with
+OH function being methylated displayed significantly dropped TON of 33,000 and 76% ee, implying the decisive role of the extra anionic oxygen donor in creating the robust TON, reaction rate and selectivity. Despite the presence of both NH and OH, ligand f-phamidol-Nacyl-Me, which cannot form tetra-coordinated Ir-ate catalysts due to unlikely amide-NH coordination to Ir-metal originated from both steric and electronic reasons, showed a rather small TON of 7,000 and 35% ee. Those control experiments provide unequivocal evidence that the OH functional of tetradentate PNNO ligand plays a critical role in AH of acetophenone, which is different from traditional NH/MH bifunctional catalysts.
+
+To gain further insights into the ultra-efficient Ir-ate catalyst, quantum chemistry studies were performed to address the AH mechanism (see SI 3.3). We have found that upon deprotonation, Ir-complex **B** is significantly higher in energy (ca. > 20 kcal mol\(^{-1}\)) than the active Ir-ate complexes (*vide supra*, Fig. 3A, S26 and Table S12). Thus, Ir-ate complexes **C** and **D** were used as starting structures in comparison with **A**. Both the well-known NNa/MH bifunctional mechanism \(^{46}\) and ONa/MH bifunctional mechanism proposed here were extensively explored computationally. Compared to **A** and Ir-ate catalyst **C** (Figs. S28-39, Tables S12-15), Ir-ate catalyst **D** provides energetically favorable pathways involving NNa/MH and ONa/MH bifunctional mechanisms, as shown in Fig. 4A and S28-30. For both NNa/MH and ONa/MH bifunctional mechanisms, the alkali cation (Na\(^+\)) can polarize the carbonyl of the ketone substrate and thus facilitate hydride transfer in the rate and selectivity determining step from the Ir-ate catalyst to form the alkoxide intermediate **III**, which is smoothly
+converted to alcohol product and initial active catalyst by taking a dihydrogen molecule.
+
+The ONa/MH bifunctional pathway giving the desired enantiomer has a much lower free energy barrier than that of the NNa/MH bifunctional pathway (\( \text{viz.} \) 6.2 vs 10.1 kcal mol\(^{-1}\)), consistent with our experimental observations. Additionally, when explicit solvent molecules were considered in the models to simulate the actual solvated cations and hydrogen bonding environment (Fig. S38), the predicted free energy barrier of 12.9 kcal mol\(^{-1}\) is comparable to the experimental TOF (Table S15) according to the Arrhenius equation.
+
+![Proposed mechanisms from DFT calculations. (A) Predicted Gibbs free energy profile for the AH of acetophenone via the active Ir-ate catalyst D through ONa/MH. (B) Bar graph showing ΔG‡(kcal/mol) for various metals. (C) Regulating 5d orbital level of Ir atom to enhance hydricity of active catalyst.](page_184_670_1082_627.png)
+
+Fig. 4 Proposed mechanisms from DFT calculations. (A) Predicted Gibbs free energy profile for the AH of acetophenone via the active Ir-ate catalyst \( \mathbf{D} \) through ONa/MH
+bifunctional (in red) and NNa/MH bifunctional (in blue) paths. (B) Average natural population analysis (NPA) charges of hydrides in active Ir-ate catalyst **D**, distance between C atom of the carbonyl group of the substrate and Ir atom in **TS1a**, and Gibbs free energy barriers of the hydride transfer step for different alkali cations. (C) Schematic three-center-four-electron (3c-4e) orbital interactions between Ir and the hydride atoms.
+
+The origin of the significantly enhanced activity upon introduction of anionic donor can be rationalized by the orbital interactions between the ligation-tunable 5\( d \) orbitals of Ir atom and 1s orbital of hydride (Fig. 4C). The f-phamidol ligand in tetra-dentate manner stabilizes the Ir-ate catalyst (18 electron complex) under basic condition and the anionic donor greatly elevates the 5\( d \)-orbital energy of Ir atom. Higher 5\( d \)-orbital energy level of Ir atom leads to weaker orbital mixing with 1s orbitals of axial hydride based on Pimentel–Rundle three-center-four-electron (3c-4e) model \(^{47,48}\), which causes larger composition of H 1s orbitals in the bonding/nonbonding orbitals and subsequently larger electron density on hydride atoms as well as stronger hydricity of the catalysts. The Ir atom and two axial hydrides in the Ir-ate catalyst forming 3c-4e bonding is confirmed by DFT and the *ab initio* complete active space self-consistent field (CASSCF) calculations (Figs. S40-43), where the natural orbital occupation numbers (NOONs) of the three orbitals formed by 5\( d_{z^2} \) orbital of Ir atom and 1s orbitals of hydrides are 1.99, 1.97, 0.03, respectively.
+
+Based on the new ONa/MH bifunctional mechanism, the Ir-ate catalysts with proton and series of alkali cations (Li\(^+\), Na\(^+\), K\(^+\), Rb\(^+\), Cs\(^+\)) were systematically investigated via
+DFT modeling (Fig. 4B and Tables S6-10). From H+ to Na+, the Ir-hydride becomes more negatively charged, implying increasing hydricity and reactivity of the active Ir-ate catalyst. However, despite the increasing of hydricity from Na+ to Cs+, higher free energy barriers for hydride transfer are noticed due to larger distance between the carbonyl group and the hydride atom originated from the increasing radius of cations. Thus, the Ir-ate catalyst with Na+ of both favorable size and hydricity gives the highest efficiency (*vide supra*).
+
+In summary, we presented an unprecedented Ir-ate catalyst with record-high turnover-numbers of 13-million, 253 s\(^{-1}\) TOF and >99% ee for an industrial route to chiral nicotine. HRMS, *in-situ* HP ATR-IR, HP NMR, Raman, XRD characterization and quantum-theoretical modeling demonstrate highly improved metal-hydride hydricity and a novel ONa/MH bifunctional mechanism. This work will likely inspire the development of other ultra-efficient 3S homogeneous catalysts for production of chemicals, fuels and materials.
+
+Methods
+
+**Optimal conditions for 10 million turnover number experiments (at S/C = 15,000,000):** To a 20.0 mL vial was added the precatalyst (3.2 mg, 4.0×10\(^{-3}\) mmol) and anhydrous iPrOH (10.0 mL) in an argon-filled glovebox. The mixture was stirred for 0.5 h at 25 °C. And then 800 mmol of acetophenone and NaOtBu (96 mg, 1 mmol) were added into a 300 mL hydrogenation vessel. Then 20 mL anhydrous iPrOH was added and a solution of Ir-precatalyst in anhydrous iPrOH (133 μL) was added *via* an injection
+port. Then the vessel was placed in an autoclave, which was closed and moved out from glovebox. The autoclave was quickly purged with hydrogen gas for three times, and then pressurized to 100 bar H₂ (keeping the hydrogen pressure not lower than 80 bar). The reaction solution was stirred at room temperature until for 30 d, and then the pressure was released carefully. The solution was removed under reduced pressure. Conversion was determined by \( ^1 \)H NMR analysis, and ee was determined by HPLC with a chiral stationary phase. 89.5% conv., 99% ee, TON = 13,425,000.
+
+**Applied asymmetric hydrogenation procedure for construction of Nicotine at 40 kilograms-scale:** Under nitrogen atmosphere, to a 20.0 mL vial was added Ir-precatalyst (1.92 g, 2.4 mmol) and anhydrous iPrOH (10.0 mL). The mixture was stirred for 0.5 h at 25 °C. In a 200 L Hastelloy hydrogenator was charged with 40 kg compound **S4** (144 mol) in 80 L isopropanol at room temperature. 400 g KOH (7.14 mol, 5 mol %) was added to the reactor and the resulting solution was degassed by five cycles of vacuo followed by filling with nitrogen. The previously prepared solution of catalyst (S/C = 60,000) in iPrOH was transferred to the hydrogenator under a stream of nitrogen by cannula. Hydrogen was initially introduced into the autoclave at a pressure of 40 bar. The reaction mixture was stirred while maintaining a temperature range of 30-45 °C, monitored by hydrogen consumption and HPLC. The reaction was complete after 5 h. The reaction mixture was cooled to 25 °C, and hydrogen was replaced by nitrogen. The solution was transferred to a glass-lined reactor and concentrated in vacuo. Crude compound **S4-3** was obtained as red oil: 41 kg (99% yield, 99% *ee*). The crude compound **S4-3** was used in the next step without further purification.
+A 10-L, 4-neck round-bottom flask was equipped with a mechanical stirrer, a condenser with an N₂ inlet, a thermowell, and an addition funnel. The flask was charged with 400 g of compound S4-3 (1.43 mol, crude product), 232 g triethylamine (2.3 mol, 1.6 equiv.) and 2.3 L of MTBE at room temperature. The resulting reaction mixture was cooled to -10 °C, followed by dropwise addition of 229 g methanesulfonyl chloride (2 mol, 1.4 equiv.). The reaction was allowed to stir at -5 °C for 2 h. A pale-yellow suspension was observed. After that, the insoluble solids were filtered off and the solids was rinsed with MTBE (2 x 500 mL). The combined organic phase was washed by saturated aqueous NaHCO₃ (2 x 1 L). The reaction mixture was cooled to -10 °C, followed by dropwise addition of 1.86 kg sulfuric acid aqueous solution (30% by weight, 4.0 equiv.). The mixture was then allowed to gradually warm to room temperature and stirred for 2 h. The organic phase was separated to waste and the lower aqueous phase was recharged to the reactor. The reaction mixture was allowed to cool to -10 °C. Finally, 7 L of sodium hydroxide solution (2 M) was added slowly until the pH reached 10~11. The aqueous phase was extracted with ethyl acetate (3 x 5 L) at room temperature. The combined organic solvent was evaporated under reduced pressure and the resulting residue was further purified by distillation under vacuum at 70 °C to afford pure product (S)-Nicotine as colorless oil (157 g, 0.97 mol, 68% yield in 3 steps, 99% ee).
+
+Data availability
+
+The data supporting the findings of this study are available within the paper and its Supplementary Information.
+References
+
+1 de Vries, J. G. & Elsevier, C. J. The Handbook of Homogeneous Hydrogenation. (WILEY-VCH, 2006).
+2 Magano, J. & Dunetz, J. R. Large-Scale Carbonyl Reductions in the Pharmaceutical Industry. Org. Process Res. Dev. 16, 1156-1184, doi:10.1021/op2003826 (2012).
+3 Carey, F. A. & Sundberg, R. J. Advanced Organic Chemistry. (Springer, 1995).
+4 Bai, S. T. et al. Homogeneous and heterogeneous catalysts for hydrogenation of CO2 to methanol under mild conditions. Chem. Soc. Rev. 50, 4259-4298, doi:10.1039/d0cs01331e (2021).
+5 Korstanje, T. J., van der Vlugt, J. I., Elsevier, C. J. & de Bruin, B. Hydrogenation of carboxylic acids with a homogeneous cobalt catalyst. Science 350, 298-302, doi:10.1126/science.aaa8938 (2015).
+6 Pritchard, J., Filonenko, G. A., van Putten, R., Hensen, E. J. & Pidko, E. A. Heterogeneous and homogeneous catalysis for the hydrogenation of carboxylic acid derivatives: history, advances and future directions. Chem. Soc. Rev. 44, 3808-3833, doi:10.1039/c5cs00038f (2015).
+7 Dub, P. A. & Ikariya, T. Catalytic Reductive Transformations of Carboxylic and Carbonic Acid Derivatives Using Molecular Hydrogen. ACS Catal. 2, 1718-1741, doi:10.1021/cs300341g (2012).
+8 Whittall, J. & Sutton, P. W. Practical Methods for Biocatalysis and Biotransformations 2. (John Wiley & Sons, 2012).
+9 Li, A. Y. & Moores, A. Carbonyl Reduction and Biomass: A Case Study of Sustainable Catalysis. ACS Sustain. Chem. Eng. 7, 10182-10197, doi:10.1021/acssuschemeng.9b00811 (2019).
+10 Asymmetric Catalysis on Industrial Scale. (Wiley-VCH, 2010).
+11 Reductions by the Alumino- and Borohydrides in Organic Synthesis 2edn, (Wiley-VCH, 1997).
+12 Doucet, H. et al. trans-[RuCl2(phosphane)2(1,2-diamine)] and Chiraltrans-[RuCl2(diphosphane)(1,2-diamine)]: Shelf-Stable Precatalysts for the Rapid, Productive, and Stereoselective Hydrogenation of Ketones. Angew. Chem. Int. Ed. 37, 1703-1707, doi:10.1002/(sici)1521-3773(19980703)37:12<1703::Aid-anie1703>3.0.Co;2-i (1998).
+13 Wen, J., Wang, F. & Zhang, X. Asymmetric hydrogenation catalyzed by first-row transition metal complexes. Chem. Soc. Rev. 50, 3211-3237, doi:10.1039/d0cs00082e (2021).
+14 Wang, H., Wen, J. & Zhang, X. Chiral Tridentate Ligands in Transition Metal-Catalyzed Asymmetric Hydrogenation. Chem. Rev. 121, 7530-7567, doi:10.1021/acs.chemrev.1c00075 (2021).
+15 Hu, A., Ngo, H. L. & Lin, W. 4,4'-Disubstituted BINAPs for highly enantioselective Ru-catalyzed asymmetric hydrogenation of ketones. Org. Lett. 6, 2937-2940, doi:10.1021/ol048993j (2004).
+16 Li, W. et al. Highly efficient and highly enantioselective asymmetric hydrogenation of ketones with TunesPhos/1,2-diamine-ruthenium(II) complexes. J. Org. Chem. **74**, 1397-1399, doi:10.1021/jo802372w (2009).
+17 Wu, W. et al. Iridium Catalysts with f-Amphox Ligands: Asymmetric Hydrogenation of Simple Ketones. Org. Lett. **18**, 2938-2941, doi:10.1021/acs.orglett.6b01290 (2016).
+18 Zheng, Z. et al. Chiral cyclohexyl-fused spirobiindanes: practical synthesis, ligand development, and asymmetric catalysis. J. Am. Chem. Soc. **140**, 10374 (2018).
+19 Wang, Y. et al. Structure, reactivity and catalytic properties of manganese-hydride amidate complexes. Nat. Chem., doi:10.1038/s41557-022-01036-6 (2022).
+20 Ratovelomanana-Vidal, V. & Phansavath, P. Asymmetric Hydrogenation and Transfer Hydrogenation. (WILEY-VCH GmbH, 2021).
+21 Xie, J. H., Liu, X. Y., Xie, J. B., Wang, L. X. & Zhou, Q. L. An additional coordination group leads to extremely efficient chiral iridium catalysts for asymmetric hydrogenation of ketones. Angew. Chem. Int. Ed. **50**, 7329-7332, doi:10.1002/anie.201102710 (2011).
+22 Xie, J. H. et al. Chiral iridium catalysts bearing spiro pyridine-aminophosphine ligands enable highly efficient asymmetric hydrogenation of beta-aryl beta-ketoesters. Angew. Chem. Int. Ed. **51**, 201-203, doi:10.1002/anie.201105780 (2012).
+23 Wu, W. et al. Asymmetric hydrogenation of \( \alpha \)-hydroxy ketones with an iridium/f-amphox catalyst: efficient access to chiral 1,2-diols. Org. Chem. Front. **4**, 555-559, doi:10.1039/c6qo00810k (2017).
+24 Arai, N. & Ohkuma, T. Design of molecular catalysts for achievement of high turnover number in homogeneous hydrogenation. Chem. Rec. **12**, 284-289, doi:10.1002/tcr.201100019 (2012).
+25 Crabtree, R. H., Felkin, H., Fillebeen-Khan, T. & Morris, G. E. Dihydridoiridium diolefin complexes as intermediates in homogeneous hydrogenation. J. Organomet. Chem. **168**, 183-195, doi:10.1016/s0022-328x(00)83274-x (1979).
+26 Young, J. F., Osborn, J. A., Jardine, F. H. & Wilkinson, G. Hydride intermediates in homogeneous hydrogenation reactions of olefins and acetylenes using rhodium catalysts. Chem. Commun., doi:10.1039/c19650000131 (1965).
+27 Montelatici, S., van der Ent, A., Osborn, J. A. & Wilkinson, G. Further studies on the homogeneous hydrogenation of olefins by use of tris (tertiary phosphine)chlororhodium(I) complexes. J. Chem. Soc. A, doi:10.1039/j19680001054 (1968).
+28 Halpern, J. Mechanism and stereoselectivity of asymmetric hydrogenation. Science **217**, 401-407, doi:10.1126/science.217.4558.401 (1982).
+29 Friedfeld, M. R., Zhong, H., Ruck, R. T., Shevlin, M. & Chirik, P. J. Cobalt-catalyzed asymmetric hydrogenation of enamides enabled by single-electron reduction. Science **360**, 888-893, doi:10.1126/science.aar6117 (2018).
+30 Friedfeld, M. R. *et al.* Cobalt precursors for high-throughput discovery of base metal asymmetric alkene hydrogenation catalysts. *Science* **342**, 1076-1080, doi:10.1126/science.1243550 (2013).
+31 Bell, S. *et al.* Asymmetric hydrogenation of unfunctionalized, purely alkyl-substituted olefins. *Science* **311**, 642-644, doi:10.1126/science.1121977 (2006).
+32 Li, B., Chen, J., Liu, D., Gridnev, I. D. & Zhang, W. Nickel-catalysed asymmetric hydrogenation of oximes. *Nat. Chem.* **14**, 920-927, doi:10.1038/s41557-022-00971-8 (2022).
+33 Liu, W., Sahoo, B., Junge, K. & Beller, M. Cobalt Complexes as an Emerging Class of Catalysts for Homogeneous Hydrogenations. *Acc. Chem. Res.* **51**, 1858-1869, doi:10.1021/acs.accounts.8b00262 (2018).
+34 Peters, B. B. C. & Andersson, P. G. The Implications of the Bronsted Acidic Properties of Crabtree-Type Catalysts in the Asymmetric Hydrogenation of Olefins. *J. Am. Chem. Soc.* **144**, 16252-16261, doi:10.1021/jacs.2c07023 (2022).
+35 Mas-Rosello, J., Smejkal, T. & Cramer, N. Iridium-catalyzed acid-assisted asymmetric hydrogenation of oximes to hydroxylamines. *Science* **368**, 1098-1102, doi:10.1126/science.abb2559 (2020).
+36 Zuo, W., Lough, A. J., Li, Y. F. & Morris, R. H. Amine(imine)diphosphine iron catalysts for asymmetric transfer hydrogenation of ketones and imines. *Science* **342**, 1080-1083, doi:10.1126/science.1244466 (2013).
+37 Ohkuma, T., Ooka, H., Hashiguchi, S., Ikariya, T. & Noyori, R. Practical Enantioselective Hydrogenation of Aromatic Ketones. *J. Am. Chem. Soc.* **117**, 2675-2676, doi:10.1021/ja00114a043 (2002).
+38 Dub, P. A. & Gordon, J. C. The role of the metal-bound N–H functionality in Noyori-type molecular catalysts. *Nat. Rev. Chem.* **2**, 396-408, doi:10.1038/s41570-018-0049-z (2018).
+39 Zhao, B., Han, Z. & Ding, K. The N-H functional group in organometallic catalysis. *Angew. Chem., Int. Ed.* **52**, 4744 (2013).
+40 Wiedner, E. S. *et al.* Thermodynamic Hydricity of Transition Metal Hydrides. *Chem. Rev.* **116**, 8655-8692, doi:10.1021/acs.chemrev.6b00168 (2016).
+41 Wu, W. *et al.* Iridium Catalysts with f-Amphox Ligands: Asymmetric Hydrogenation of Simple Ketones. *Organic Letters* **18**, 2938-2941, doi:10.1021/acs.orglett.6b01290 (2016).
+42 Yu, J. *et al.* Discovery and development of ferrocene-based tetradentate ligands for Ir-catalysed asymmetric hydrogenation of ketone. *Green Synth. Catal.* **3**, 175-178, doi:10.1016/j.gresc.2022.03.004 (2022).
+43 Madden, C. *et al.* Catalytic turnover of [FeFe]-hydrogenase based on single-molecule imaging. *J. Am. Chem. Soc.* **134**, 1577-1582, doi:10.1021/ja207461t (2012).
+44 Faber, K. *Biotransformations in Organic Chemistry.* (Springer Berlin, Heidelberg, 2011).
+45 Hall, M. & Bommarius, A. S. Enantioenriched compounds via enzyme-catalyzed redox reactions. Chem. Rev. **111**, 4088-4110, doi:10.1021/cr200013n (2011).
+46 Dub, P. A., Scott, B. L. & Gordon, J. C. Why Does Alkylation of the N-H Functionality within M/NH Bifunctional Noyori-Type Catalysts Lead to Turnover? *J. Am. Chem. Soc.* **139**, 1245-1260, doi:10.1021/jacs.6b11666 (2017).
+47 Pimentel, G. C. The Bonding of Trihalide and Bifluoride Ions by the Molecular Orbital Method. *J. Chem. Phys.* **19**, 446-448, doi:10.1063/1.1748245 (1951).
+48 Rundle, R. E. Electron Deficient Compounds. II. Relative Energies of "Half-Bonds". *J. Chem. Phys.* **17**, 671–675, doi:10.1063/1.1747367 (1949).
+
+Acknowledgement.
+
+This work was supported by National Natural Science Foundation of China (grant 22033005, 22038002, and 21991113), National Key R&D Program of China (grant 2021YFA1500200), Stable Support Plan Program of Shenzhen Natural Science Fund (grant No. 20200925161222002) and Guangdong Provincial Key Laboratory of Catalysis (grant No. 2020B121201002). Computational resources are supported by the Center for Computational Science and Engineering (SUSTech) and Tsinghua National Laboratory for Information Science and Technology.
+
+Author contributions
+
+X.Z., J.L., S.T.B, Q.L. conceived the idea and directed the project. C.Y. designed and conducted the experiments with inputs and support from Q.L., F.H. and G.Q.C. Y.F.J. performed all the theoretical computations with guidance of J.L. and C.Q.X. S.T.B. performed the characterizations with support from Y.P. using HRMS, in-situ HP ATR-IR, HP NMR, Raman, XRD spectroscopy. S.T.B. and Y.F.J. analyzed the data and drafted the manuscript. All authors contributed to the manuscript.
+
+Competing interests: X.D., S.G. and Q.L. are inventors on patents (WO 2021/212880 Al, EP 3 925 955 A1, CN 113527187 A, US 2022/0089564 Al, CN 202210771902.X and 202111097411.3), held and submitted by Shenzhen Catalys Technology Co., Ltd.
+Additional information
+
+Supplementary information is available for this paper at Additional information
+
+Supplementary information is available for this paper at https://doi.org/....
+
+Correspondence and requests for materials should be addressed to X.Z., J.L., S.T.B, Q.L.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SupplementaryInformation.pdf

0b8f1f6be4795d439f43f4c9625e8c3c0efbb434a42fa2851f1539502182ecc0/peer_review/peer_review.md

@@ -0,0 +1,337 @@
+Peer Review File
+
+Effect of trade on global aquatic food consumption patterns
+
+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/.
+Reviewers' Comments:
+
+Reviewer #1:
+Remarks to the Author:
+The authors cover an important topic that is worthy of publication. However, in my opinion there are some errors that need correcting.
+
+1. The title of the paper is misleading (and is the opinion of the authors and not of most NGOs) and I would suggest it be changed to "Effect of trade on global aquatic food consumption patterns";
+
+2. The paper excludes shelled mollusks (bivalves, clams, mussels, oysters etc) and aquatic plants that are also consumed as food - so really it only covers selected aquatic animal foods;
+
+3. It is a pity that the authors did not consult with the FAO Food Balance Sheets (latest data for 2020) in their analysis as this includes all aquatic food products on a live weight basis, including all mollusks and aquatic plants - these food products would bring down the trophic level of global aquaculture production even further;
+
+4. In view of the above, the paper is not a review of global aquatic food consumption patterns, but of selected food products - this should be clearly stated;
+
+5. Statements implying that higher trophic level aquatic species have a better protein quality are not true and should stated as such (line 67, 74, 149, 152 etc) - the implication being that lower trophic level species have a lower nutritional quality - in fact the reverse is probably true as wild-caught high trophic level species are more likely to be contaminated with environmental contaminants such as POPs, heavy metals, and microplastics;
+
+6. The authors investigate the contribution of aquatic animal food products to total animal protein supply;
+
+7. Trade impacts on aquatic food consumption patterns - no mention is given to the costs of aquatic food products and their affordability or not to consumers - in most South American countries poultry is the cheapest source of animal protein whereas in Asia it is the reverse - dried fish usually being the cheapest source of animal protein to the rural poor, including small captured pelagic fish in West African countries;
+
+8. Finally, the exclusion of the Peruvian anchovy from the analysis (Methods - 263, 289 etc) is not recommended - an increasing proportion being directly consumed as food (it is a food grade product and as such not labeled as forage fish in my opinion), and also have a low trophic level which would influence final trophic level calculations;
+
+Reviewer #2:
+Remarks to the Author:
+This paper connects production with consumption of aquatic foods and explores how trade influenced HATL and consumption patterns. I found the analysis and paper interesting and well written, further increasing our understanding of how trade within the aquatic food system contribute to alleviating nutritional insecurity. As such, it is an important contribution to the scientific community and policymakers. Further, I believe this paper should be accepted for publication, pending some major comments I provide below. These are concerned mostly with the interpretation of the results and conclusions.
+
+The authors used changes in consumption per capita and trophic levels to infer on changes to food
+security. However, such an inference is only partial. For one, fish’s nutritional contribution is more a product of its micronutrient contribution (concentrations) than its protein. The variation in protein concentration among fish (teleost) is very small, but as the authors rightly mentioned at the beginning there are small fish (lower trophic level) with phenomenal micronutrient concentrations. So, I would suggest amending the text in the cases where high trophic level is used synonymously with “higher quality”. The importance of tracking trophic levels is well noted but it’s difficult to infer on the nutritional quality as a whole just by looking at trophic levels.
+
+The authors note positive contributions of trade to consumption per capita amongst most of the population in the world, but looking closely reveals very trivial contributions (Fig 3) for most continents (except North and South America). If the authors would have highlighted the changes to consumption of the ‘before trade’ and the ‘after trade’ in percentages, I would assume we would see very small values in many cases, maybe within the range of uncertainties (which are not noted here). The authors note that >60% of countries experienced an increase in food availability, but how much of those are above, say, 20% change or any other uncertainty level? It is important to note that food availability is only one dimension of food security and other important factors, especially within countries, such as access, affect food security too. Moreover, the authors track food availability, not true consumption at household level. Given the increasing rates of food loss and waste as the industrialization and income of countries increase some of those food gains they noted may be offsetted by supply chain leakages, resulting in smaller gains or perhaps even reduction in actual consumptions. Therefore, the trade analysis offers a glimpse to possible food security gains, but other factors, not addressed here, are as crucial in order to draw a strong conclusion on how trade effects food security. The authors must acknowledge this in their conclusions and limitations more explicitly (not only in the Methods). Perhaps even changing the title of the paper.
+
+Lastly, the authors highlight the aquatic food production and consumption amongst nations, but maybe adding another figure that notes the supply chain flows – i.e., connection of aquatic foods between specific countries (circular flow chart) – can also be revealing in this analysis.
+
+General comments
+Line 129 might be necessary to define what trade deficient/surplus means
+Fig 2a and b – please make sure the colors of export and import are similar on both subplots
+
+Fig 2d,f I would use the same color code as in fig 1 for HATL for consistency and easiness of comparison.
+Line 168: Is “before trade” actually the domestic consumption per capita of what is produced locally? Maybe change the wording to make it clearer.
+Line 169 maybe add a sentence that says that you’ve explored the differences between “before” and “after trade” on countries’ consumption per capita and HATL to identify patterns of changes due to trade. It’s not clear enough.
+Line 170 unnecessary comma after especially.
+Line 172: please write “after/before trade” as a phrase using parenthesis, otherwise its confusing.
+Line 235-237 This is inaccurate. Energy cascade among trophic levels of 10% is generally correct, but GHG of wild capture fisheries is dictated by fishing efforts and when it comes to aquaculture with specific management practices, feeding practices (prices, access) and regions, there are exceptions to trophic level being a predictor for environmental impacts. I would cite the Gephart et al, Nature 2021 for the environmental impacts of aquatic foods.
+Line 242-243 only in the past? It’s still an on-going question in the present. This present research adds another piece to the puzzle, but it certainly does not resolve this. Trade impacts food security on multiple dimensions, not all of them addressed here.
+Line 252 imbalances of what? Not clear
+Reviewer #3:
+Remarks to the Author:
+This paper uses seafood production and trade data from the FAO statistical database and FishBase to develop species-level mass balance data and a trophic level dataset. The authors use the datasets to calculate per capita aquatic food consumption and an aquatic food trophic level at the country level. The authors examine how aquatic food consumption and the aquatic food trophic level is impacted by trade of aquatic food. The work is important given global levels of traded aquatic food and the mixed literature with regards to the impacts of trade on aquatic food security. The results show that the mean aquatic food trophic level has declined. The results also show that international trade has increased aquatic food consumption and the trophic level of most countries as well as reduced geographic differences in trophic levels.
+
+The paper is well written, the methodology is sound and described in great detail, and I expect the paper will be of high interest to those in and out of the field. My main concern is related the authors’ conclusion that a decline in aquatic trophic level “is potentially good news from the perspective of the environmental footprint of food production” (line 232-233). The relationship between trophic level and environmental performance is much more nuanced. Gephart et al. (2021, Nature) show that several low trophic level species, like tilapia and shrimp, have higher environmental impacts in many dimensions compared to higher trophic level species like salmon. More so, the environmental performance of aquacultured species is intricately linked to feed inputs and conversion efficiencies which doesn’t necessarily correspond to trophic level (Cottrell et al. 2021, Reviews in Aquaculture). I would like to see the authors interpret their results in light of this.
+
+The authors should also compare the per capita aquatic food consumption estimates in this paper with the FAO food balances and discuss thoughts on why their estimates are in some cases very different from the food balances (e.g., the authors’ estimates of per capita consumption for China and Indonesia are much lower).
+REVIEWER COMMENTS
+
+NOTE: We address each comment individually below with the original comment in black text and our response in blue text. To help locating our changes in the manuscript, we indicate the lines with “L-”.
+
+Reviewer #1:
+
+The authors cover an important topic that is worthy of publication. However, in my opinion there are some errors that need correcting.
+
+We are grateful for your thorough review and your constructive feedback, which have helped improving our manuscript. We hope to have addressed all your concerns and that you find our revision satisfactory.
+
+1. The title of the paper is misleading (and is the opinion of the authors and not of most NGOs) and I would suggest it be changed to "Effect of trade on global aquatic food consumption patterns";
+
+Agreed. We have changed the title to "Effect of trade on global aquatic food consumption patterns" as suggested.
+
+2. The paper excludes shelled mollusks (bivalves, clams, mussels, oysters etc) and aquatic plants that are also consumed as food - so really it only covers selected aquatic animal foods;
+
+Thank you for your comment. Our primary consideration for not including these groups is that the edible proportion in live weight of shelled mollusks and algae is very low, and there is a significant variation among different mollusk species. Moreover, they comprised a very small proportion in edible weight of total aquatic food production and consumption\( ^{1-3} \). For instance, on an edible-weight basis, mollusks and algae comprised only 6% and 7.6%, respectively, of total aquaculture output\( ^{1,4} \). We have added some text to the corresponding section of the Methods to clarify this point (L301-308).
+
+“Algae, aquatic plants, mollusks (Bivalvia, Gastropod, Barnacle, and Ascidiacea), echinoderm, cnidaria, miscellaneous aquatic animals (such as turtles, frogs, and mammals), and reported inedible species were not considered for this study. Although mollusks and algae accounted for a significant proportion of fisheries output in live weight, especially in aquaculture, they comprised a very small proportion in edible weight \(^{15,18,53,54}\). Further, we were unable to find publicly available preservation factors (i.e., the ratio of edible portions to final product live weight) for the different processing methods of these groups.”
+
+3. It is a pity that the authors did not consult with the FAO Food Balance Sheets (latest data for 2020) in their analysis as this includes all aquatic food products on a live weight basis, including all mollusks and aquatic plants - these food products would bring down the trophic level of global aquaculture production even further;
+The FAO Food Balance Sheets only contain live weight of broad taxonomic groups5. There are significant variations in the edible proportions among different mollusk and aquatic plant species, such as Bivalvia, Gastropod, Barnacle, and Ascidiacea. Therefore, if we only have the live weight of mollusks and aquatic plants without more detailed information, the calculated trophic levels of production and consumption would become unreliable.
+
+We agree that the trophic level of global aquaculture production and HATL will further decrease if we combine all aquatic food products including mollusks and aquatic plants. However, as explained in our response to your previous comment, given their very low edible proportions, including these food products would result in excessively low trophic levels in both production and consumption especially for aquaculture. Nonetheless, we have included statements to acknowledge that inclusion of mollusks and aquatic plants would result in a lower trophic level of production and consumption compared to the actual estimated values, which suggests that our estimates of HATL reductions are conservative (L247-251):
+
+“In this study, we did not consider mollusks and aquatic plants. Given that these groups have a lower trophic level than most other aquatic foods and considering the recent increase in the consumption of mollusks and aquatic plants2,15,29,44, our estimate of the current consumption trophic level can be considered conservative (i.e., the inclusion of these groups will likely lower our estimated HATL).”
+
+4. In view of the above, the paper is not a review of global aquatic food consumption patterns, but of selected food products - this should be clearly stated;
+
+Thank you for your comment. We have edited the text in the Introduction to make clear the scope of the study (L79-81):
+
+“We then calculate the HATL and per capita consumption across different countries and regions to analyze global aquatic food (i.e., fish, cephalopods, and crustaceans) consumption patterns, trade characteristics, and impacts.”
+
+We also provide a clear definition of what we mean by aquatic foods in the context of this study in the Methods (L 299-300):
+
+“The term “aquatic foods” is used throughout this study to denote all freshwater and marine fish, cephalopods, and crustaceans.”
+
+5. Statements implying that higher trophic level aquatic species have a better protein quality are not true and should stated as such (line 67, 74, 149, 152 etc) - the implication being that lower trophic level species have a lower nutritional quality - in fact the reverse is probably true as wild-caught high trophic level species are more likely to be contaminated with environmental contaminants such as POPs, heavy metals, and microplastics;
+
+Thank you for this important comment. We strongly agree and have modified the statements regarding the extended implications of trophic levels to clarify this point (L63-76):
+
+“They not only represent a synthetic metric of species’ diets, which is an important indicator
+of different aspects of the environmental footprint of food production for aquaculture and wild caught aquatic foods\(^{28,29}\), but they are also widely recognized as an appropriate indicator of aquatic food value (i.e., higher trophic level generally corresponding to higher price)\(^{30-32}\). Although the trophic level of food items in the human diet (human trophic level, HTL) has been considered a simple composite metric that synthetically reflects global patterns of human diet\(^{33}\), there is currently no quantitative assessment of the human aquatic food trophic level (HATL) and the impacts of trade on it. Nonetheless, we note that some small low-trophic level pelagic and inland fish are also nutrient-rich (e.g., calcium, iron, zinc, long-chain omega-3 polyunsaturated fatty acids)\(^{6,8,34}\), and that wild-capture high-trophic level species are more likely to be contaminated with biomagnifying substances such as persistent organic pollutants (POPs), heavy metals, and microplastics\(^{35-37}\). Therefore, the trophic level of aquatic foods can indicate the value of aquatic foods based on price, but it does not predictably reflect the concentration of any nutrients or contaminants status."
+
+We corrected all the text in the cases where trophic level is used synonymously with nutritional or protein quality (L26, 28, 67, 75, 153, 155, 204, 245, 266).
+
+6. The authors investigate the contribution of aquatic animal food products to total animal protein supply;
+
+Research regarding the contribution of aquatic animal food products to total animal protein supply has been conducted by FAO\(^{6-9}\). Our primary focus lies in analyzing global patterns of aquatic food consumption (live weight availability and HATL), trade characteristics, and impacts.
+
+7. Trade impacts on aquatic food consumption patterns - no mention is given to the costs of aquatic food products and their affordability or not to consumers - in most South American countries poultry is the cheapest source of animal protein whereas in Asia it is the reverse - dried fish usually being the cheapest source of animal protein to the rural poor, including small captured pelagic fish in West African countries;
+
+We thank the reviewer for this important comment. We have edited the relevant discussion in the 'Discussion' section to reflect these geographical differences in the access to aquatic food in particular to vulnerable communities (L266-273):
+
+"However, the direct contribution of trade to the food system in vulnerable population groups is limited because the beneficiaries tend to be high-income groups as most exported products consist of high-trophic level species (high-value species) for international markets\(^{48}\). Furthermore, differences in dietary habits, income levels, natural resource conditions, and other aspects among different regions can lead to variation in the cost of aquatic products and consumers' affordability. These situations highlight the need for free, transparent, and adaptive trade and market policies to ensure that all segments of society benefit from international trade\(^{47}\)."
+
+8. Finally, the exclusion of the Peruvian anchovy from the analysis (Methods - 263, 289 etc) is not recommended - an increasing proportion being directly consumed as food (it is a food grade product and as such not labeled as forage fish in my opinion), and also have a
+low trophic level which would influence final trophic level calculations;
+
+Thank you for your suggestion. We further consulted relevant information about this issue. First, the State of World Fisheries and Aquaculture 2018’s Fish utilization and processing section describes how Peruvian anchovy is used as follows:
+
+“Many different species are used for fishmeal and fish oil production, small pelagic species predominating. Many of the species used, such as Engraulis ringens, have comparatively high oil yields but are rarely used for direct human consumption.”
+
+“In 2016, landings from fisheries directed for fishmeal production were down to less than 15 million tonnes (live weight equivalent) because of reduced catches of Engraulis ringens.”
+
+Second, IFFO - The Marine Ingredients Organisation also published an article titled “Peruvian Anchovy Why feed, not food?” (https://www.iffo.com/case-study-peruvian-anchovy-why-feed-not-food). This article lists a number of reasons why Peruvian anchovies are not used for direct consumption, and the following is a summary of the article.
+
+“Although eaten as whole fish, the majority of Peruvian anchovy are turned into fish oil for feed and capsules, as well as fishmeal, mainly used in aquafeeds. The comparatively low rate of direct human consumption has led some to accuse the industry and the Peruvian Government of depriving local communities of a valuable food source. However, although much effort is, and has been, devoted to promoting the consumption of anchovy in fresh, canned and frozen state, that market remains very small. Despite the efforts of the Peruvian Government and the private sector, direct human consumption of Peruvian anchovy remains at a very low level in the country.”
+
+This article also said humans have not been very keen on eating anchovies. Between 2005 and 2011, there was a slight increase in anchovies’ consumption, reaching a peak of 1.75% of all anchovies caught in 2011, but it dropped significantly when the purchase obligation for public entities was removed. In recent years, consumption levels have been at half of the 2011 peak.
+
+As we explain in the manuscript (L486-489), we nonetheless could not remove the effect of all other forage species, which to some extent should offset the possible impact in our trophic level calculations from not considering the very small proportion of consumption of the Peruvian anchovy. In any case, we have edited the ‘Preprocessing of fisheries datasets’ section in the Methods to reflect the reasons behind this choice (L315-319):
+
+“Engraulis ringens had comparatively a massive production but is rarely used for direct human consumption in Chile and Peru55. Additionally, due to the absence of accurate or even approximate long-term time series data on Engraulis ringens consumption proportions, accurate country-level estimations of the theoretical maximum live weight availability for this species were not possible.”
+Reviewer #2:
+
+This paper connects production with consumption of aquatic foods and explores how trade influenced HATL and consumption patterns. I found the analysis and paper interesting and well written, further increasing our understanding of how trade within the aquatic food system contribute to alleviating nutritional insecurity. As such, it is an important contribution to the scientific community and policymakers. Further, I believe this paper should be accepted for publication, pending some major comments I provide below. These are concerned mostly with the interpretation of the results and conclusions.
+
+We are grateful for your careful review of our manuscript and for your thoughtful and constructive feedback that helped improving our manuscript. Please see below our point-by-point response to your comments. We hope you find our revision satisfying.
+
+The authors used changes in consumption per capita and trophic levels to infer on changes to food security. However, such an inference is only partial. For one, fish’s nutritional contribution is more a product of its micronutrient contribution (concentrations) than its protein. The variation in protein concentration among fish (teleost) is very small, but as the authors rightly mentioned at the beginning there are small fish (lower trophic level) with phenomenal micronutrient concentrations. So, I would suggest amending the text in the cases where high trophic level is used synonymously with “higher quality”. The importance of tracking trophic levels is well noted but it’s difficult to infer on the nutritional quality as a whole just by looking at trophic levels.
+
+Thank you for this important comment, which we strongly agree with.
+
+First, we modified the sentences related to food security, changing “food security changes” to “changes in aquatic food availability” (L191-194):
+
+“Finally, the fact that Africa is notably the only region where both post-trade per capita consumption of aquatic foods and HATL have increased underscores the importance of trade in reducing food deficiency and malnutrition in Africa.”
+
+(L264-266):
+
+“Our study supports the conclusion that global trade can improve food availability by allowing most countries to access larger quantities and higher trophic level aquatic foods that otherwise are domestically unavailable.”
+
+(L279-282):
+
+“Based on our findings, we believe that aquatic food trade can play substantial and diverse roles in global transformations toward more sustainable and equitable food systems and healthy diets to address multiple forms of food deficiency and malnutrition5,8.”
+
+Second, following your suggestion, we have revised the text to avoid making those direct, unidirectional links between trophic level and food quality. We revised the statements regarding the extended implications of trophic levels (L63-76):
+
+“They not only represent a synthetic metric of species’ diets, which is an important indicator
+of different aspects of the environmental footprint of food production for aquaculture and wild caught aquatic foods\(^{28,29}\), but they are also widely recognized as an appropriate indicator of aquatic food value (i.e., higher trophic level generally corresponding to higher price)\(^{30-32}\). Although the trophic level of food items in the human diet (human trophic level, HTL) has been considered a simple composite metric that synthetically reflects global patterns of human diet\(^{33}\), there is currently no quantitative assessment of the human aquatic food trophic level (HATL) and the impacts of trade on it. Nonetheless, we note that some small low-trophic level pelagic and inland fish are also nutrient-rich (e.g., calcium, iron, zinc, long-chain omega-3 polyunsaturated fatty acids)\(^{6,8,34}\), and that wild-capture high-trophic level species are more likely to be contaminated with biomagnifying substances such as persistent organic pollutants (POPs), heavy metals, and microplastics\(^{35-37}\). Therefore, the trophic level of aquatic foods can indicate the value of aquatic foods based on price, but it does not predictably reflect the concentration of any nutrients or contaminants status."
+
+Third, we corrected the text in the cases where trophic level is used synonymously with nutritional or protein quality (L26, 28, 67, 75, 153, 155, 204, 245, 266).
+
+The authors note positive contributions of trade to consumption per capita amongst most of the population in the world, but looking closely reveals very trivial contributions (Fig 3) for most continents (except North and South America). If the authors would have highlighted the changes to consumption of the ‘before trade’ and the ‘after trade’ in percentages, I would assume we would see very small values in many cases, maybe within the range of uncertainties (which are not noted here). The authors note that >60% of countries experienced an increase in food availability, but how much of those are above, say, 20% change or any other uncertainty level? It is important to note that food availability is only one dimension of food security and other important factors, especially within countries, such as access, affect food security too. Moreover, the authors track food availability, not true consumption at household level. Given the increasing rates of food loss and waste as the industrialization and income of countries increase some of those food gains they noted may be offsetted by supply chain leakages, resulting in smaller gains or perhaps even reduction in actual consumptions. Therefore, the trade analysis offers a glimpse to possible food security gains, but other factors, not addressed here, are as crucial in order to draw a strong conclusion on how trade effects food security. The authors must acknowledge this in their conclusions and limitations more explicitly (not only in the Methods). Perhaps even changing the title of the paper.
+
+Sorry for making this unclear. Indeed, as seen in Figure 3, the impact of aquatic food trade appears negligible for many countries. However, even minor changes in per capita consumption volume resulting from trade are significant for countries with low production and consumption levels. To reflect this better, and after updating population data for certain countries, we recalculated the changes in per capita consumption after trade and found that 66.1% of countries experienced an increase in per capita consumption of aquatic foods (Supplementary Table 2). Next, we calculated the average annual change in per capita consumption of aquatic foods due to trade as a ratio to the average annual per capita production. For reference, we found that only 37.9% of countries had a change rate below
+20%, with the majority of countries (62.1%) having a rate exceeding 20%. Nevertheless, we do acknowledge that the precise range of uncertainties in this study cannot be determined due to complex factors. Therefore, we have included reference to the shortage and uncertainties of trade impacts in the new section “Limitations and uncertainties” (L489-495):
+
+“Third, given the increasing rates of food loss and waste linked to the increasing industrialization and income of countries, it is likely that the effects of trade may have been offset to some extent by supply chain leakages, resulting in some cases in smaller gains than those estimated here or perhaps even reduction in actual consumptions. Therefore, our trade analysis provides a baseline for potential food security gains, but we recognize that other factors, not addressed here, are also crucial to draw a strong conclusion on how trade affects food security and should be subject of future studies.”
+
+We also agree that availability is just one aspect of food security. We have modified the descriptions related to food security accordingly as explained in our answered to your previous question.
+
+Finally, we have followed your suggestion and changed the title to "Effect of trade on global aquatic food consumption patterns".
+
+Lastly, the authors highlight the aquatic food production and consumption amongst nations, but maybe adding another figure that notes the supply chain flows – i.e., connection of aquatic foods between specific countries (circular flow chart) – can also be revealing in this analysis.
+
+We appreciate your valuable suggestion. Due to data limitations and lack of a precise method, we have not been able to obtain detailed information on the complex trade network relationships in imports and exports of aquatic foods among countries. The most applied database (United Nations Comtrade database) contains bilateral aquatic food trade value information (i.e., data on trade from country A to country B), but does not include quantitative data on trade flows for all aquatic food trade\(^{10}\). In this study, therefore, we did not focus on trade flows, but we think this is a new very important yet very complex topic. Your suggestion is worthy and we continue to strive building trade flows and supply chains through new method in the process of data collection and analysis. This represents our future direction of effort.
+
+General comments
+Line 129 might be necessary to define what trade deficient/surplus means
+
+Agreed. We have added the definitions of trade deficient and surplus to the corresponding section (L132-133).
+
+Fig 2a and b – please make sure the colors of export and import are similar on both subplots
+
+We have changed the colors of export and import subplots to be the same.
+
+Fig 2d,f I would use the same color code as in fig 1 for HATL for consistency and easiness
+of comparison.
+
+Corrected. We have changed the colors for all trophic level figures to the same color code.
+
+Line 168: Is “before trade” actually the domestic consumption per capita of what is produced locally? Maybe change the wording to make it clearer.
+
+Thank you. We have revised this sentence to make it clearer (L170-174):
+
+“Differences in per capita consumption of aquatic foods and HATL by country between ‘before trade’ (i.e., consumption stage before trade transactions) and ‘after trade’ (i.e., apparent consumption patterns after completing trade transactions) on reveal the rapidly increasing volume and shifting trade features in various regions that have affected both per capita consumption of aquatic foods and HATL in most parts of the world.”
+
+Line 169 maybe add a sentence that says that you’ve explored the differences between “before” and “after trade” on countries’ consumption per capita and HATL to identify patterns of changes due to trade. It’s not clear enough.
+
+Thank you for your suggestion. We have revised several sentences accordingly (L170-176):
+
+“Differences in per capita consumption of aquatic foods and HATL by country between ‘before trade’ (i.e., consumption stage before trade transactions) and ‘after trade’ (i.e., apparent consumption patterns after completing trade transactions) on reveal the rapidly increasing volume and shifting trade features in various regions that have affected both per capita consumption of aquatic foods and HATL in most parts of the world. Continentally, the per capita consumption of aquatic foods has decreased in Asia and especially South America after trade (Fig. 3a).”
+
+Line 170 unnecessary comma after especially.
+
+Corrected.
+
+Line 172: please write “after/before trade” as a phrase using parenthesis, otherwise its confusing.
+
+Corrected.
+
+Line 235-237 This is inaccurate. Energy cascade among trophic levels of 10% is generally correct, but GHG of wild capture fisheries is dictated by fishing efforts and when it comes to aquaculture with specific management practices, feeding practices (prices, access) and regions, there are exceptions to trophic level being a predictor for environmental impacts. I would cite the Gephart et al, Nature 2021 for the environmental impacts of aquatic foods.
+
+We strongly agree with this comment. Indeed, we can’t directly associate trophic levels with environmental performance, as the latter encompasses multiple dimensions beyond just energy transfer efficiency between trophic levels. The environmental performance of farmed species is also intricately linked to feed inputs and conversion efficiencies which don’t necessarily correspond to trophic level. We also agree with Cottrell, et al. 11 in that the trophic level in aquaculture can no longer be viewed as a trait of the farmed species,
+but rather is a dynamic feature of the production system. Consequently, we have interpreted the implications of HATL decline as follows (L252-260):
+
+“Nonetheless, the continuous decrease in the effective trophic level of the majority of farmed species29 coupled with the increasing proportion of low trophic level species in the diet is encouraging for progressing towards reducing dependence on multiple marine ingredients (e.g., fishmeal and oil). Further improvements in resource conversion efficiency on this basis will yield even greater results. Furthermore, aquatic foods not only provide comparatively higher nutrient richness across multiple micronutrients, vitamins, and long-chain polyunsaturated fatty acids relative to terrestrial animal-source foods6, but aquatic foods also typically have lower environmental footprint compared to other animal-sourced foods45,46. The observed trends of increasing global contributions from aquatic foods suggest a promising future for more sustainable global diets.”
+
+Line 242-243 only in the past? It’s still an on-going question in the present. This present research adds another piece to the puzzle, but it certainly does not resolve this. Trade impacts food security on multiple dimensions, not all of them addressed here.
+
+Thank you for your comments. We have modified this sentence and revised the text related to impacts on food security, narrowing down the implications of this research (please refer to our response to your first two comments).
+
+Line 252 imbalances of what? Not clear
+
+We have modified this sentence to make it clear (L274-275):
+
+“Until now, the problem of imbalances in the growth of demand and aquatic food supply remains prevalent across regions, countries, and income groups18,49,50.”
+
+Reviewer #3:
+
+This paper uses seafood production and trade data from the FAO statistical database and FishBase to develop species-level mass balance data and a trophic level dataset. The authors use the datasets to calculate per capita aquatic food consumption and an aquatic food trophic level at the country level. The authors examine how aquatic food consumption and the aquatic food trophic level is impacted by trade of aquatic food. The work is important given global levels of traded aquatic food and the mixed literature with regards to the impacts of trade on aquatic food security. The results show that the mean aquatic food trophic level has declined. The results also show that international trade has increased aquatic food consumption and the trophic level of most countries as well as reduced geographic differences in trophic levels.
+
+Thank you for your constructive feedback and useful comments. We hope you find our responses satisfying.
+
+The paper is well written, the methodology is sound and described in great detail, and I expect the paper will be of high interest to those in and out of the field. My main concern is related the authors’ conclusion that a decline in aquatic trophic level “is potentially good
+news from the perspective of the environmental footprint of food production” (line 232-233). The relationship between trophic level and environmental performance is much more nuanced. Gephart et al. (2021, Nature) show that several low trophic level species, like tilapia and shrimp, have higher environmental impacts in many dimensions compared to higher trophic level species like salmon. More so, the environmental performance of aquacultured species is intricately linked to feed inputs and conversion efficiencies which doesn’t necessarily correspond to trophic level (Cottrell et al. 2021, Reviews in Aquaculture). I would like to see the authors interpret their results in light of this.
+
+Thank you for your comment. As explained in our response to a similar comment from the previous reviewer, we agree that we can’t directly associate trophic levels with environmental performance. The environmental performance of farmed species is also intricately linked to feed inputs and conversion efficiencies which don’t necessarily correspond to trophic level. We also agree with Cottrell, et al. 11 in that the trophic level in aquaculture can no longer be viewed as a trait of the farmed species, but rather is a dynamic feature of the production system. Consequently, we have interpreted the implications of HATL decline as follows (L252-260):
+
+“Nonetheless, the continuous decrease in the effective trophic level of the majority of farmed species29 coupled with the increasing proportion of low trophic level species in the diet is encouraging for progressing towards reducing dependence on multiple marine ingredients (e.g., fishmeal and oil). Further improvements in resource conversion efficiency on this basis will yield even greater results. Furthermore, aquatic foods not only provide comparatively higher nutrient richness across multiple micronutrients, vitamins, and long-chain polyunsaturated fatty acids relative to terrestrial animal-source foods6, but aquatic foods also typically have lower environmental footprint compared to other animal-sourced foods45,46. The observed trends of increasing global contributions from aquatic foods suggest a promising future for more sustainable global diets.”
+
+The authors should also compare the per capita aquatic food consumption estimates in this paper with the FAO food balances and discuss thoughts on why their estimates are in some cases very different from the food balances (e.g., the authors’ estimates of per capita consumption for China and Indonesia are much lower).
+
+There are indeed some differences in per capita consumption of aquatic foods for certain countries, which primarily arise from a slight disparity between our equations. We have only calculated the theoretical maximum available volume (i.e., production + imports - exports). FAO’s aquatic food supply calculations take also into account the post-balancing non-food uses (Including utilization of aquatic products for reduction to meal and oil, for feed and bait, for ornamental purposes, withdrawals from markets and any other non-food use of fish production) and stock variations. However, FAO acknowledges that information on changes in stocks occurring between the production and the retail levels, or in levels of inventories, is very incomplete3. In most instances data indicated are the minimum required to avoid a negative balance.
+
+Another reason is our exclusive focus on fish, cephalopods, and crustaceans. When we narrow down FAO’s data to these categories, the average annual difference in per capita
+consumption for China is 1.61 kg/year, while Indonesia's difference is just 0.1 kg/year. When we calculate the average annual per capita consumption relative to FAO's average annual per capita supply, 66.7% of countries have an average change magnitude below 20%, and a substantial majority, 73.6%, fall below the 30% threshold. We also verified the differences between theoretical exports and actual export volume in all 174 countries, which indicate the validity and robustness of our mass balancing method (L456-461).
+
+To summarize all this, we now compare and discuss the discrepancies with FAO Food Balance Sheets in the ‘Limitations and uncertainties’ section of the Methods, and acknowledge the possible implications and uncertainties resulting from not having accounted for non-food uses and variations in stocks in our balanced data (L480-489):
+
+“Second, to see the magnitude of the differences between both estimates, we compared our balanced average annual per capita consumption to the average annual per capita supply in the FAO Food Balance Sheets. A total of 66.7% of countries had an average difference magnitude below 20%, and a large majority (73.6%) fell below 30%. These differences can mainly be attributed to the fact that the FAO accounts for non-food uses and variations in stocks, as well as differences in live weight conversion factors. Although we removed the world’s most important forage fish (i.e., Engraulis ringens) from the capture dataset, we couldn’t remove the effect of all other existing fish species used in the production of fishmeal, fish oil, and other non-food uses, as well as variations in stocks on aquatic food consumption in all countries.”
+
+References
+
+1 Naylor, R. L. et al. A 20-year retrospective review of global aquaculture. Nature **591**, 551–563, doi:10.1038/s41586-021-03308-6 (2021).
+2 Naylor, R. L. et al. Blue food demand across geographic and temporal scales. Nat. Commun. **12**, 1–14, doi:10.1038/s41467-021-25516-4 (2021).
+3 Fisheries and Aquaculture Software. FishStatJ: Software for Fishery and Aquaculture Statistical Time Series (FAO Fisheries Division, 2023).
+4 Edwards, P., Zhang, W., Belton, B. & Little, D. C. Misunderstandings, myths and mantras in aquaculture: Its contribution to world food supplies has been systematically over reported. Marine Policy **106**, 103547 (2019).
+5 Fishery and Aquaculture Statistics. Food balance sheets of fish and fishery products 1961–2017 (FishstatJ) (FAO Fisheries Division, Rome, 2020).
+6 FAO. The State of World Fisheries and Aquaculture. Vol. 4 (FAO, 2022).
+7 FAO. The State of World Fisheries and Aquaculture. Vol. 4 (FAO, 2020).
+8 FAO. The State of World Fisheries and Aquaculture. Vol. 4 (FAO, 2018).
+9 FAO. The State of World Fisheries and Aquaculture. Vol. 4 (FAO, 2016).
+10 Gephardt, J. A. & Pace, M. L. Structure and evolution of the global seafood trade network. Environmental Research Letters **10**, 125014, doi:10.1088/1748–9326/aae065 (2015).
+11 Cottrell, R. S. et al. Time to rethink trophic levels in aquaculture policy. Rev. Aquac., doi:10.1111/raq.12535 (2021).
+Reviewers' Comments:
+
+Reviewer #1:
+Remarks to the Author:
+All my comments and suggestions have been addressed by the authors
+
+Reviewer #2:
+Remarks to the Author:
+Dear Authors,
+Thank you for adequately addressing all my comments. The paper has indeed improved and I believe it is ready for publication.
+
+Reviewer #3:
+Remarks to the Author:
+The authors have addressed many of the major questions raised by reviewers, and the manuscript has improved.
+
+Please find some additional comments for further improvement.
+
+There are still indications especially in the results and discussion that changes in HATL infers changes to food security and environmental impacts. The authors should remove the positive and negative connotations when reporting changes in HATL. There is not a clear argument that a higher HATL is better or worse for nutritional or environmental outcomes. For instance, Line 204 states "...trade has improved the availability and trophic level of aquatic foods...". Similarly, line 197 says "trade has positively impacted HATLs..."
+
+There are also some weak, unsupported statements in the discussion. For instance, Line 279, "Based on our findings, we believe that aquatic food trade can play substantial and diverse roles in global transformations toward more sustainable and equitable food systems and healthy diets to address multiple forms of food deficiency and malnutrition." The analysis provides no evidence that trade of aquatic foods increases environmentally sustainability, and trade appears to create inequities given that 33% of countries have a decline in per capita consumption after trade.
+
+Also, “the need for free, transparent, and adaptive trade and market policies to ensure that all segments of society benefit from international trade (line 272-273). How will free trade ensure all segments benefit? Free trade tends to result in unequal distribution of wealth.
+
+In its current form, the discussion fails to summarize and highlight the study's contribution to the broader literature and the study's implications.
+REVIEWER COMMENTS
+
+NOTE: We address each comment individually below with the original comment in black text and our response in blue text. To help locating our changes in the manuscript, we indicate the lines with “L-”.
+
+Reviewer #1:
+All my comments and suggestions have been addressed by the authors.
+
+We are grateful for your thorough review.
+
+Reviewer #2:
+Thank you for adequately addressing all my comments. The paper has indeed improved and I believe it is ready for publication.
+
+We are grateful for your thorough review.
+
+Reviewer #3:
+The authors have addressed many of the major questions raised by reviewers, and the manuscript has improved. Please find some additional comments for further improvement.
+
+We are grateful for your careful review and your thoughtful and constructive feedback that helped improving our manuscript. Please see below our point-by-point response to your comments. We hope you find our revision satisfying.
+
+There are still indications especially in the results and discussion that changes in HATL infers changes to food security and environmental impacts. The authors should remove the positive and negative connotations when reporting changes in HATL. There is not a clear argument that a higher HATL is better or worse for nutritional or environmental outcomes. For instance, Line 204 states “…trade has improved the availability and trophic level of aquatic foods…”. Similarly, line 197 says “trade has positively impacted HATLs.”
+
+Thank you for your comment. We agree that we can’t directly involve the HATL to nutritional or environmental outcomes. We removed the positive and negative connotations when reporting changes in HATL following your suggestion.
+
+(L189-191):
+“Finally, the fact that Africa is notably the only region where both post-trade per capita consumption of aquatic foods and HATL have increased underscores the importance of trade in improving aquatic food availability in Africa.”
+
+(L195-196):
+“Similarly, trade has also increased HATLs in most countries across continents except Oceania (Fig. 3c).”
+
+(L201-203):
+“Globally, international trade has increased the availability and trophic level of aquatic
+foods in most (>60%) countries over the past decades (Fig. 3b, d).”
+
+There are also some weak, unsupported statements in the discussion. For instance, Line 279, “Based on our findings, we believe that aquatic food trade can play substantial and diverse roles in global transformations toward more sustainable and equitable food systems and healthy diets to address multiple forms of food deficiency and malnutrition.” The analysis provides no evidence that trade of aquatic foods increases environmentally sustainability, and trade appears to create inequities given that 33% of countries have a decline in per capita consumption after trade.
+
+Thank you. We strongly agree with this comment and have modified the referred statements in the discussion accordingly (L277-284):
+
+“Based on our findings, it is evident that trade has played an important role in harmonizing the global aquatic food consumption. Nevertheless, there are still some countries where per capita consumption of aquatic foods has not improved after trade and remains quite low (Figs. 1e and 3b). Heightened attention and concerted efforts for context-specific mitigation should therefore be given in the future to these countries. To this end, attaining globally equitable trade distribution patterns as well as more harmonized trade environment and policies should be a priority.”
+
+Also, “the need for free, transparent, and adaptive trade and market policies to ensure that all segments of society benefit from international trade (line 272-273). How will free trade ensure all segments benefit? Free trade tends to result in unequal distribution of wealth.
+
+Thank you for your comments. We have modified this sentence to make it more reasonable (L269-271):
+
+“These situations highlight the need for fair, transparent, sustainable and adaptive trade and market policies to ensure that more segments of society benefit from international trade47”
+
+In its current form, the discussion fails to summarize and highlight the study’s contribution to the broader literature and the study's implications.
+
+Thank you for your comments. We revised the last paragraph in discussion to better summarize and highlight our work’s contribution and implications (L284-292):
+
+“Despite the important progress attained in research on aquatic food production and trade, significant challenges persist in achieving a comprehensive understanding of the outcomes of aquatic food trade5,17-21. Our work adds another piece to this puzzle by identifying the implications of trade for contemporary changes in global aquatic food consumption patterns, highlighting the increased availability and trophic level of consumed aquatic foods in a majority of countries with reduced differences in HATL despite the important remaining inequalities. These results provide an important foundation to guide future research on the globalization of aquatic food systems and the impacts of trade on food security.”

0b8f1f6be4795d439f43f4c9625e8c3c0efbb434a42fa2851f1539502182ecc0/preprint/preprint.md

@@ -0,0 +1,276 @@
+Trade is improving global aquatic food consumption patterns
+
+Jun XU (xujun@ihb.ac.cn)
+Institute of Hydrobiology, Chinese Academy of Sciences
+
+Kangshun Zhao
+Institute of Hydrobiology, Chinese Academy of Sciences
+
+Steven Gaines
+UC Santa Barbara   https://orcid.org/0000-0002-7604-3483
+
+Jorge Molinos
+Arctic Research Center, Hokkaido University
+
+Min Zhang
+Huazhong Agricultural University
+
+Article
+
+Keywords:
+
+Posted Date: June 30th, 2023
+
+DOI: https://doi.org/10.21203/rs.3.rs-3085251/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on February 15th, 2024. See the published version at https://doi.org/10.1038/s41467-024-45556-w.
+Trade is improving global aquatic food consumption patterns
+
+Kangshun Zhao1,2,3, Steven D. Gaines3, Jorge García Molinos4, Min Zhang5, Jun Xu1,6,7*
+
+1 Donghu Experimental Station of Lake Ecosystems, State Key Laboratory of Freshwater Ecology and Biotechnology of China, Institute of Hydrobiology, Chinese Academy of Sciences, Wuhan, China
+2 University of Chinese Academy of Sciences, Beijing, China
+3 Bren School of Environmental Science & Management, University of California, Santa Barbara, Santa Barbara, CA, USA
+4 Arctic Research Center, Hokkaido University, Sapporo, Japan
+5 College of Fisheries, Huazhong Agricultural University, Hubei Provincial Engineering Laboratory for Pond Aquaculture, Freshwater Aquaculture Collaborative Innovation Center of Hubei Province, Wuhan, China
+6 State Key Laboratory of Marine Resource Utilization in South China Sea, Hainan University, Haikou, China
+7 Laboratory for Marine Fisheries Science and Food Production Processes, Qingdao National Laboratory for Marine Science and Technology, Qingdao, China
+* Corresponding author: JX (xujun@ihb.ac.cn)
+Abstract
+
+Globalization of fishery products is playing a significant role in shaping the harvesting and use of aquatic foods, but the vigorous debate has focused on whether the trade is a driver of the inequitable distribution of aquatic foods. Here, we develop species-level mass balance and trophic level identification datasets for 174 countries and territories to analyze global aquatic food consumption patterns, trade characteristics, and impacts from 1976 to 2019. We find that per capita consumption of aquatic foods has increased significantly at the global scale, but the human aquatic food trophic level (HATL), i.e., the average trophic level of aquatic food items in the human diet, is declining (from 3.42 to 3.18) because of the considerable increase in low-trophic level aquaculture species output relative to that of capture fisheries since 1976. Moreover, our study finds that trade can improve food security by contributing to increasing the availability and quality of aquatic foods in >60% of the world’s countries. Trade has also reduced geographic differences in the quality of aquatic food consumption among countries over recent decades. We suggest that there are important opportunities to widen the current focus on productivity gains and economic outputs to a more equitable global distribution of aquatic food quantity and quality.
+Introduction
+
+The Sustainable Development Goals (SDGs) agenda puts food security and ending malnutrition as a global priority ¹. Aquatic systems have a significant role to play in meeting these objectives. Fisheries, aquaculture, and their trade are critical to the achievement of food security, and sustainable economic, social, and environmental development goals ²,³. In recent decades, global fisheries and aquaculture production has grown substantially; aquatic foods are among the most highly traded commodities in the global food system and are becoming increasingly globalized ⁴,⁵.
+
+As a highly diverse food group, aquatic foods are now widely recognized in global food systems and can supply critical nutrients and improve overall human health ⁶-⁹. However, accelerating climate change, overfishing, industrial pollution, and coastal urbanization challenge the ocean’s ability to meet growing aquatic food demands ¹⁰-¹³. The percentage of fishery stocks with biologically unsustainable levels has increased from 10% in 1974 to 35.4% in 2019 ¹⁴. Promisingly, global aquaculture has rapidly developed over the past few decades and is thought to be the only reliable way to meet the growing future demand for aquatic foods ¹⁵,¹⁶. Meanwhile, given the geographic patchiness of wild fish and aquaculture production, trade will be increasingly essential for the redistribution of global aquatic products and food security.
+
+Our understanding of the wide diversity of aquatic species produced and traded worldwide, and the impacts of aquatic food products trade across geographies on food security goals remain ambiguous and the evidence is mixed ¹⁷-²⁰. While previous studies have provided essential insights into general aquatic food trade and consumption characteristics and trade impacts in several countries and regions ⁵,¹⁷,¹⁹-²¹, our collective understanding of the outcomes of aquatic food globalization is still limited by a fundamental gap between detailed production and trade data. Reconciling aquatic food production and trade data has remained a challenge due to mismatches in species- versus product-level reporting and weight losses during processing ²²,²³. Compared with a deeper understanding of the role of trade in land-based food systems (e.g., agricultural and livestock products) ⁵,²⁴,²⁵, insights into global aquatic food consumption patterns and the impact of trade continue to lag far behind ²⁶.
+
+The trophic levels of animal or plant species, representing the relative positions in the aquatic food chains, are a primary metric used in ecological studies for a wide range of
+applications \(^{27}\). They not only represent a synthetic metric of species’ diets, which is an important indicator of aspects of the environmental footprint of food production (for both aquaculture and wild caught aquatic foods) \(^{28,29}\), but are also widely recognized as an appropriate indicator of aquatic food quality (i.e., higher trophic level corresponding to generally higher price and better protein quality) \(^{30-32}\). Hicks, et al. \(^{8}\) also indicated that protein concentrations were greater in marine fish species from higher trophic levels and those with a pelagic feeding pathway. Although the trophic level of food items in the human diet (human trophic level, HTL) has been considered a composite metric that simply and synthetically reflects global patterns of human diet \(^{33}\), there is currently no quantitative assessment of the human aquatic food trophic level (HATL). Given that some small pelagic and inland fish are also rich in nutrients \(^{6,8,34}\), the trophic level of aquatic foods only indicates the quality of aquatic products based on price and protein quality, and does not reflect the concentration of other nutrients (e.g., calcium, iron, zinc, long-chain omega-3 polyunsaturated fatty acids). Here, we first use the FAO national fisheries and aquaculture production and trade data (1979–2019) to develop a species-level mass balance dataset and a trophic level identification dataset for 174 countries and territories (hereafter called countries). We then calculate the HATL and per capita consumption across different countries and regions to analyze global aquatic food (i.e., fish, cephalopods, and crustaceans) consumption patterns, trade characteristics, and impacts.
+Results
+
+Aquatic food consumption patterns
+
+Population, income growth, and associated changes in dietary habits are the main influential drivers of the increase in global fish demand in recent decades \(^{35,36}\). From 1976 to 2019, global per capita consumption of aquatic foods has increased significantly, but the HATL has declined from 3.42 to 3.18 (Fig. 1a), contrary to the global trend of HTL that also includes land based foods \(^{33}\). This declining trend for aquatic foods can be primarily explained through a combination of two factors. First, the global aquaculture trophic level is significantly lower than that of capture fisheries (nearly 0.8 lower than capture fisheries on average) (Fig. 1b). Second, while global capture fisheries production has experienced only a marginal increase in recent decades, aquaculture output has experienced a sustained and very rapid increase over the entire study period, especially in Asia (Fig. 1b, Supplementary Figs. 1 and 2). The proportion of aquatic foods originating from aquaculture production rose from 6% in the 1960s to 56% in 2020 \(^{14}\). This finding indicates that aquaculture is driving the decreasing trend in global HATL, despite increasing consumption of aquatic food (Fig. 1a) and the rapid growth in the production of high-trophic level species driven by globalized trade and favorable economic conditions for large-scale intensive farming \(^{37}\). Asia, and China in particular, plays a crucial role not only because of their significant contribution to global aquaculture production but also because it accounts for the world’s largest quantity of farmed low-trophic level species (Supplementary Figs. 1–4 and 5b). It should be noted that the consumption of aquatic foods does not refer to the quantity effectively eaten in this study but to the live weight available before consumption.
+
+Regional consumption and HATL trends show large variation (Fig. 1c, d). In line with global trends, Asia’s per capita consumption of aquatic foods has increased rapidly, mainly driven by China, whereas the HATL declined at an approximate rate of 0.08 per decade; approximately 1.4 times the global rate (Fig. 1a, c, d and Supplementary Fig. 4c, d). By contrast, the per capita consumption of aquatic foods in Europe and South America rose rapidly for a short time and then fell sharply from the 1990s, coinciding with the increase in their HATLs. North America experienced a slight overall increase in per capita consumption of aquatic foods but a decrease in HATL. By contrast, Oceania was the only region where both per capita
+consumption of aquatic foods and HATL increased over the study period, although recent years suggest a decline in per capita consumption (Fig. 1c, d). Currently, Asia has the lowest HATL, Africa has the lowest per capita consumption of aquatic foods and a low HATL, and Europe and Oceania are the regions with the highest per capita consumption of aquatic foods and HATL (Fig. 1c-f). In general, we find regions with more developing countries to have lower per capita consumption and lower trophic level of aquatic foods than regions with more developed countries. Nonetheless, trade seems to mediate these apparent strong imbalances in aquatic food consumption across regions and countries (see next section).
+Fig. 1 Global trends of aquatic food production and apparent consumption from 1976 to 2019. a, The global change in per capita consumption of aquatic foods and HATL. b, The global change in production and trophic level of aquaculture and capture fisheries. The line refers to trophic level, and the envelope refers to production. c, Trends of per capita consumption of aquatic foods in different continents. d, Trends of HATL in different continents. e, The mean country-level per capita consumption of aquatic foods. f, The median country-level HATL. Countries in grey: No data available.
+Aquatic food trade
+
+In the past decades, international trade in aquatic foods has risen across all continents, especially in Asia and Europe, which represent the two major trading regions (Fig. 2a). Historically, aquatic food trade has been dominated by a few countries, such as China, USA, Norway, Thailand, and Japan (Fig. 2c, e and Supplementary Table 1). Since the World Trade Organization (WTO) was founded in 1995, Asia and South America have been the major trade surplus regions, whereas Africa and North America have been the major trade deficit regions (Fig. 2a). The share of imports in total aquatic food consumption has been rising in developed countries, which have good supply chain infrastructures; and more consumers can afford to buy imported high-value species \(^{2,14}\). Developing countries are becoming increasingly prominent in the supply of aquatic products and becoming increasingly important as supply chain intermediaries, importing raw materials and re-exporting processed or value-added products \(^{14}\). For example, although China is also one of the largest importers and exporters (in terms of live weight), more than two-thirds of these imports are raw materials to be processed and re-exported \(^{23}\).
+
+The trophic level of continental trade from 1976 to 2019 was generally above 3.3 across all continents (Fig. 2b), while the country-level median import and export trophic level was above 3.0 for most countries (Fig. 2d, f). This figure suggests that most of the exported aquaculture and capture fisheries products consist of higher-trophic level species for international markets, particularly from Europe and Oceania. Interestingly, in recent decades, Asia and South America’s aquatic food import trophic levels have surpassed those of exports, whereas North America and Europe show the opposite trend (Fig. 2b). Meanwhile, although the trophic level of imports and exports has remained similar in Africa, the import volume is gradually increasing faster than the export volume (Fig. 2a, b). Together, these results indicate that the trade structure and consumption features of aquatic foods in these regions are changing. First, a more robust demand for aquatic species with higher trophic level (i.e., higher quality) is apparent in Asia and South America as production and incomes rise, gradually redirecting products once produced mainly for exports toward domestic markets \(^{15,35,38,39}\). Second, the trend in developing countries to export high-quality aquatic foods in exchange for low-quality aquatic foods from industrial fisheries is being reversed. While significant quantities of high-
+trophic level species (e.g., salmonids) are traded and continue to grow, the trade volume of low-trophic level species (e.g., tilapia and shrimp) has also increased drastically \(^{14}\), which helps the global import trophic level to remain largely stable between 3.5 to 3.6 (Supplementary Fig. 6). Asian and South American countries, especially in East Asia, have been the central supply regions for relatively low-trophic level species in recent decades (Fig. 2a, b, d, f).
+
+![Trends and maps showing aquatic food trade volume and trophic level from 1976 to 2019](page_184_370_1080_1012.png)
+
+Fig. 2 Global aquatic food trade volume and trophic level from 1976 to 2019. a, Trends of aquatic food import and export volume in different continents. b, Trends of aquatic food import and export trophic level in different continents. The mean country-level aquatic food import volume (c) and export volume (e). The median country-level aquatic food import trophic level (d) and export trophic level (f). Countries in grey: No data available.
+Trade impacts on aquatic food consumption patterns
+
+The rapidly increasing volume and different trade features in various regions have affected both per capita consumption of aquatic foods and HATL in most parts of the world. Specifically, before trade refers to the production stage, while after trade refers to the apparent consumption patterns after completing trade transactions. The per capita consumption of aquatic foods has decreased in Asia and especially, South America after trade (Fig. 3a). Meanwhile, the HATL of these two regions remained nearly unchanged (Fig. 3c). After 2000, Asia and South America were two major trade surplus regions (i.e., aquatic food consumption after trade was lower than before trade), supplying increasing quantities of low-trophic level species and playing an essential role in boosting per capita consumption of aquatic foods in the rest of the world (Figs. 2a, b, and 3a, b). North America has maintained its per capita consumption of aquatic foods over time, despite producing less, by a gradually increased reliance on imports (Figs. 2a and 3a). For example, the import share of total aquatic food consumption in the USA rose from one-third in 1961 to nearly three-quarters in 2019 \(^{14}\). Conversely, import and export volumes in Europe have grown at similar rates, resulting in only slight differences in per capita consumption of aquatic foods before and after trade in recent years. Countries in both continents are mostly developed, and post-trade HATL has declined slightly over the past decade (Fig. 3c). Meanwhile, although the difference in Oceania’s per capita consumption of aquatic foods before and after trade is small, the HATL has decreased considerably after trade (Fig. 3a, c), because the trophic level of exports is significantly higher than that of imports (Fig. 2d). Finally, Africa is notably the only region where both post-trade per capita consumption of aquatic foods and HATL have increased, underscoring the importance of trade in reducing food insecurity and malnutrition in Africa.
+
+At the national scale, although Asia is a trade surplus region, more than 60% of Asian countries have benefited from trade in their aquatic food consumption (Fig. 3a, c). Except for South America, per capita consumption of aquatic foods increased in most countries of all continents (Fig. 3a). Similarly, trade has also positively impacted HATLs in most countries of all the continents except Oceania (Fig. 3c). In particular, more than 70% of countries in Africa and Europe benefit from trade in both aspects of aquatic food consumption. Indeed, the dominance of certain countries in international fish trade masks the importance of trade for
+Africa, where aquatic food demand has grown faster than supply, resulting in an increase in the import share of consumption from 16% in 1970 to 39% in 2017 as production from domestic fish capture has either stagnated or been exported \(^{18,40}\). Globally, international trade has improved the availability and quality of aquatic foods in most countries (>60% of the world’s countries) over the past decades (Fig. 3b, d). Furthermore, the heterogeneity in HATL of countries declines globally after trade, especially in Europe, North America, and Africa; and the mean HATL has also increased in most continents (Fig. 4). Overall, our findings suggest that international trade has both reduced geographic differences in HATL and improved the aquatic food consumption in most parts of the world and therefore will be an important part of a transition to sustainable fisheries.
+Fig. 3 Impact of trade on aquatic food consumption per capita and HATL. a, Trends of continental per capita consumption of aquatic foods before and after trade. b, The mean country-level change in annual per capita consumption of aquatic foods after trade from 1976 to 2019. c, Trends of continental HATL before and after trade. d, The mean country-level change in annual HATL after trade from 1976 to 2019. All country-level changes in per capita consumption of aquatic foods and HATL are the post-trade value minus the pre-trade value year by year. Countries in grey: No data available. Percentage values indicate the proportion of countries affected by trade (positively or negatively) in each region (a, c) and globally (b, d) (for details, see Supplementary Table 2). Before trade refers to the production stage, while after trade refers to the apparent consumption patterns after completing trade transactions.
+Fig. 4 Global and regional HATL dispersion before and after trade from 1976 to 2019. In all plots, each point shows the mean estimate, and error bar shows 95% reference range (mean±1.96 SD) for each country. The shaded blue column (before trade) and green column (after trade) indicates the 95% reference range (mean±1.96 SD) for all countries in different continents. The dotted lines indicate the HATL averages of all countries in different continents. Numbers indicate the number of countries in each region (N = 174 total). Before trade refers to the production stage, while after trade refers to the apparent consumption patterns after completing trade transactions.
+Discussion
+
+The need to increase aquatic food diversity and supply to help achieve global food and nutrition security goals within environmental boundaries is a global consensus. In recent decades, the rapid growth in aquatic food globalization and consumption has been driven by increased trade liberalization and facilitated by advances in food processing and transportation technologies \(^{2,14}\). Aquatic foods are among the most highly traded commodities \(^{4,5}\), comprising nearly 10% of all food trade (by value) \(^{21}\). Obviously, international trade in fishery products is playing a significant role in shaping the harvesting and global aquatic food consumption \(^{4}\). However, aquatic foods have often been excluded from previous studies on detailed global food trade due to the difficulty in reconciling species-level production and trade data \(^{18,22}\).
+
+By developing a species-level mass balance dataset and a trophic level identification dataset for 174 countries, we first reveal precise country-level aquatic food consumption patterns, identify the features of aquatic food trade, and quantify the effect of trade on aquatic food consumption globally. From 1976 to 2019, global per capita consumption of aquatic foods has increased significantly, but the HATL is declining (from 3.42 to 3.18). This is potentially good news from the perspective of the environmental footprint of food production. In the natural food web, the average energy transfer efficiency between trophic levels is only 10%, with the majority of energy being lost through energy expenditure \(^{28,41-43}\). This implies that the lower the trophic level of aquatic foods consumed by humans, the less ecosystem energy is utilized in their production, resulting in a lower environmental footprint. Meanwhile, international trade has played an important role in harmonizing the global consumption of aquatic products, increasing the availability and quality of aquatic foods in most countries (especially for countries in Africa and Europe), and reducing the heterogeneity in HATL worldwide.
+
+In the past, the impacts of the aquatic food trade on food security and well-being have been a subject of debate \(^{3,20}\). While some claimed that the fish trade has a pro-poor effect, others denounced the negative impact of fish exports on local poor populations’ food security \(^{20,44}\). Our study supports the conclusion that global trade can improve food security by allowing most countries to access larger quantities and higher trophic level aquatic foods that are domestically unavailable. However, the direct contribution of trade to food system in vulnerable population
+groups is limited because the beneficiaries tend to be high-income groups as most exported products consist of high-trophic level species (high-value species) for international markets \(^{45}\). This situation highlights the need for free and transparent trade and market policies to ensure that all segments of society benefit from international trade \(^{44}\).
+
+Until now, apart from the aquatic food supply, the problem of imbalances also exists in aquatic food demand growth across regions, countries, and income groups \(^{18,46,47}\). Many people remain under multiple forms of malnutrition and per capita consumption of aquatic foods is far below the world average \(^{6,14}\). Geography plays a major role in explaining these differences \(^{14}\). As the world may be approaching the constraints of a finite, global, aquatic food production capacity \(^{48,49}\), sourcing trajectories from all nations must be considered together \(^{35}\). Based on our findings, we believe that aquatic food trade can play substantial and diverse roles in global transformations toward more sustainable and equitable food systems and healthy diets to address multiple forms of food insecurity \(^{6,8}\). To this end, much depends on attaining globally equitable distribution patterns as well as more harmonized trade environment and economic policies oriented toward social equality and environmental sustainability.
+Methods
+
+Preprocessing of fisheries datasets
+
+Three original global fisheries statistic datasets (aquaculture, capture, and trade) were taken from the FishStatJ software \(^{50}\). The original trade dataset contained more than 100,000 commodities, each mixed by species/species group, preservation, and preprocessing method, such as ‘Catfish fillets, frozen’. The original aquaculture and capture datasets also had more than 25,000 items. The earliest coincident year of these three datasets was 1976. We removed all items of negligible importance (i.e., total volume <100t) from 1976 to 2019 in all original datasets to facilitate the definition of the live weight conversion factor for trade commodities and the trophic level of species or species groups in aquaculture and capture datasets according to the fishing area in each country (see next two sections). Removed items accounted for a total of 0.001% of aquaculture production, 0.004% of capture production, and 0.044% of trade volume.
+
+The term “aquatic foods” is used throughout this study to denote all freshwater and marine fish, cephalopods, and crustaceans. Brackish fish were identified as freshwater fish or marine fish according to the major fishing area in the aquaculture and capture datasets. Although mollusks and algae accounted for a significant proportion of fisheries output in live weight, especially in aquaculture, they comprised a very small proportion in edible weight \(^{15,50}\). Algae, aquatic plants, mollusks (Bivalvia, Gastropod, Barnacle, and Ascidiacea), echinoderm, cnidaria, miscellaneous aquatic animals (such as turtles, frogs, and mammals), and reported inedible species were not considered for this study.
+
+In the trade dataset, we deleted commodities for which no live weight conversion factors were available, such as fish sausage and fish cake. We have limited our focus solely to edible aquatic foods, and have therefore discarded commodities such as fishmeal and fish oil that are unsuitable for direct human consumption. To mitigate the impact of specific fish used in the production of fishmeal and fish oil on aquatic food consumption in several countries, we removed the world’s most important forage fish (i.e., Engraulis ringens) from the capture dataset. Because Engraulis ringens had comparatively massive production but was rarely used for direct human consumption in Chile and Peru \(^{51}\). While many other small pelagic fish (e.g., herrings and sardines) are also partly reduced to fishmeal and fish oil, their volumes were
+relatively low, and some were used directly for human consumption. Therefore, we retained them for analysis. Following these data processing steps, 174 countries with available aquaculture, capture, and trade data accounting for 95.8% of the world’s fisheries and aquaculture production from 1976 to 2019 were included for subsequent analyses.
+
+To determine whether the removal of production and trade items with a total volume < 100t from 1976 to 2019 would result in highly unequal impacts on small and large countries, we examined the proportion of removed items in production and trade for all 174 countries. We discovered that in 97% of countries, the ratio of excluded items to total production is below 1%. Additionally, in 95% of countries, the proportion of excluded items to total trade volume is less than 3%. Therefore, the removal of items with a total volume <100t would not have significant unequal impact on different countries.
+
+Live weight conversion of traded commodities
+
+Trade data were reported at the species-level and broader commodity groups (e.g., ‘Salmons nei, Eels nei, and Tunas nei’) and labeled based on processing (e.g., fresh, frozen, dried, fillets, etc.). In order to back-transform processed product weight into whole-animal live weight equivalents, we first extracted commodity species or species groups, preprocessing and preservation methods from original commodities in trade dataset. We then used conversion factors from Fluet-Chouinard, et al. \(^{52}\) Hortle \(^{53}\), expressed as live weight (kg) of aquatic animals required to make 1 kg of product, calculated as the product of the preprocessing factor and the preservation factor. The preprocessing factor represents the ratio of live weight to edible portions after cleaning (beheading, gutting, etc.). The preservation factor is defined as the ratio of edible portions to final product weight. Aquatic products for which no preservation and/or preprocessing methods were reported were assumed to refer to fresh units and/or whole-animal. All conversion factors are listed in Supplementary Table 4. After conversion, the average live weight ratio of the total imports to the total exports was \(0.99 \pm 0.04\):1 over the study period (1976–2019), suggesting that the conversion factors were relatively reliable, because in theory the world’s total import volume should equal total export volume. All processing details and results were available in Supplementary Data 1.
+Trophic level identification
+
+Items in the aquaculture and capture production datasets were often not provided at the species-level but were more generically referred to by a species-group name (e.g., ‘Groupers nei’). In such cases, we searched each species group in Fishbase (www.fishbase.org)54 using the “Common name is” function to screen out all species identified by that common name and with clear economic value according to the reported main fishing area by country. In a few cases, no species were found. In those rare cases, we used the “Common name ends with” function instead to identify the species involved. The trophic level of each species group was then calculated as the mean trophic level of all identified species for that group (Supplementary Data 2 and 3). The trophic level of specific species was extracted directly from Fishbase.
+
+For unidentified ‘Freshwater fish nei’ and ‘Marine fish nei’ items in production dataset, the trophic level was obtained by the production-weighted average of trophic levels of identified fish species or species groups according to the fishing area in each country. For those few countries that did not have any identified species or species groups, we weighted and averaged the trophic level of identified ‘Freshwater fish nei’ and ‘Marine fish nei’ items from all countries in the same fishing area. The trophic levels of marine ‘Pelagic fish nei’ and ‘Demersal fish nei’ were considered the same as ‘Marine fish nei’. The trophic level of unidentified items was based on the trophic level and total production of identified species or species groups:
+
+\[
+TL = \frac{\sum TL_n * W_n}{\sum W_n}
+\]
+
+where \( TL \) is the trophic level of unidentified items, \( TL_n \) is the trophic level of involved species or species group to be weighted, and \( W_n \) is the total production of involved species or species group from 1976 to 2019.
+
+In the import dataset, the trophic level of specific species was also extracted directly from Fishbase. For generic import commodities, such as ‘Tunas nei’, the trophic level was calculated as the weighted average of trophic levels based on global production data for all involved species in the group. It was assumed that the most productive species were the most likely to enter the trade flow. When several trade commodities included more than three species or species groups (e.g., ‘Herring, anchovy, sardine, sardinella, brisling/sprat, mackerel, Indian mackerel, seerfish, jack & horse mackerel, jack, crevalle, cobia, silver pomfret, pacific.saury,
+scad, capelin, etc.'), we kept the first three species or species groups (i.e., ‘Herring, anchovy, and sardine’) based on the importance ranking assumption. The trophic level of this commodity type was the weighted average of trophic levels of all involved species in these three groups. Moreover, the trophic level of import ‘marine fish nei’ and ‘freshwater fish nei’ commodities were obtained as the weighted average of trophic levels of all ‘marine fish nei’ and ‘freshwater fish nei’ items in the global production dataset. Since we subtracted exports and reexports from production and imports after identifying the trophic level of items in production and import datasets, the trophic level of exported commodities could be acquired when the species-level mass balance was finished (see ‘Species-level mass balance from FAO statistics’ below).
+
+In the production and trade datasets, ‘Crustacean’ was given a trophic level of 2.5, ‘Cephalopod’ of 3.0, and ‘Demersal percomorphs nei’ of 4.0. Although the trophic level of cultured species is related to the feed composition and diverges in effective trophic level from their wild counterparts \(^{29,31}\), we did not consider this due to the lack of sufficient data. Similarly, the trophic level of the same wild capture species was considered constant across different seas and time periods.
+
+**Human aquatic food trophic level**
+
+In this study, human aquatic food trophic level (HATL) was considered a composite metric that reflects human aquatic food diet patterns simply and synthetically. We calculated the HATL using trophic level and live weight data of consumed species or species groups \(^{55,56}\):
+
+\[
+HATL = \frac{\sum TL_i * W_{ij}}{\sum W_{ij}}
+\]
+
+(2)
+
+where \(TL_i\) is the trophic level of species or species group \(i\), and \(W_{ij}\) is the live weight of species or species group \(i\) in year \(j\). HATL is the quantity-weighted average of trophic levels of species or species groups consumed in a particular year by country. Similarly, the trophic level of aquaculture, capture, imports, and exports were calculated using equation (2) as defined above.
+
+**Species-level mass balance from FAO statistics**
+
+We subtracted export weights from production in four sequential steps and present the details of species-level mass balance from FAO statistics in Supplementary Fig. 7. Once the exports and reexports were subtracted from production and imports separately, the remaining weight was assumed to represent apparent consumption per commodity group in each country.
+
+**Mass balance principle 1:**
+Each reexported commodity was matched one-to-one with the imported commodity with the same common name. All unmatched commodities were matched with each country’s ‘Fish nei’ or ‘Freshwater fish nei’ or ‘Marine fish nei’ items. Step 1 produced the remaining imports data.
+
+Mass balance principle 2:
+We combined aquaculture, capture, and remaining imports data (produced in step 1) into one dataset. The same species or species groups in each country were summed, and the trophic levels were simultaneously weighted and averaged.
+
+Each exported commodity was matched one-to-one with the combined data items with the same common name (i.e., aquaculture + capture + remaining imports). All unmatched commodities were matched with each country's ‘Fish nei’ or ‘Freshwater fish nei’ or ‘Marine fish nei’ items. This step was deemed necessary, because we believed that global reexport volumes were underestimated. For example, Asche, et al. \(^{23}\) estimated that 74.9% of China’s seafood imports were reexported, but there were very few records of China’s reexports in the FAO trade dataset, so some reexports in the trade dataset must be only roughly marked for exports or omitted. Thus, we combined aquaculture, capture, and remaining imports data to subtract exports. Step 2 produced the remaining exports 1 (negative value) and consumption volume 1 (positive value).
+
+Mass balance principle 3:
+Species-level remaining exports 1 commodity was matched with the generic item (e.g., ‘Yellow tuna’ can be matched with ‘Tuna nei), and generic commodity was matched with all contained species (e.g., ‘Tunas nei’ could be matched with all tuna species) in consumption volume 1. Some detailed matching information between several specific exported commodity groups and production items can be searched in Supplementary Table 3.
+
+All unmatched remaining exports 1 commodities were matched with the ‘Fish nei’ or ‘Freshwater fish nei’ or ‘Marine fish nei’ items in each country. A few remaining unmatched ‘Fish nei’ or ‘Freshwater fish nei’ or ‘Marine fish nei’ were matched with the three most productive items in each country. Each remaining exports 1 commodity was deducted from consumption volume 1 according to the proportion of the output of all matching items each year. Step 3 produced the remaining exports 2 (negative value) and consumption volume 2
+(positive value).
+
+Mass balance principle 4:
+Except for each country’s ‘Freshwater fish nei’, ‘Marine fish nei’, ‘Fish nei’, ‘Pelagic fish nei’, and ‘Demersal fish nei’ were matched with the three most productive items, all remaining export 2 commodities were matched with the ‘Freshwater fish nei’ or ‘Marine fish nei’ or ‘Fish nei’ items in consumption volume 2. Each remaining exports 2 commodities was deducted from consumption volume 2 according to the proportion of the output of all matching items every year. Step 4 produced the remaining unexplained exports (negative value) and final consumption volume (positive value).
+
+Theoretical reexported commodities and volumes can be obtained by comparing the original imports with the subtracted imports. Likewise, the theoretical exports are the difference between the final consumption volume and the original data (i.e., aquaculture + capture + remaining imports).
+
+Mass balance rationality analysis and limitations
+Theoretical exports plus theoretical reexports were hereafter called exports. After the process described above, the total exports accounted for 95.3 ± 0.9% (mean ± SD) of the total original exports (including exports and reexports), and the ratio to imports was 0.96 ± 0.03:1 from 1976 to 2019 (Supplementary Fig. 6 a). We also calculated the proportion of export volume to original export volume for all countries and regions in each year, with a minimum average of 72.6% from 1979 to 2019, and over 90% of countries and regions exceeding 85%. The remaining unmatched exports might be due to imprecise conversion factors, imperfect matching, and reporting errors 23, which only account for 1.1 ± 0.4% of total consumption from 1976 to 2019 (Supplementary Fig. 6 b). Due to these inevitable errors mentioned above, it is almost impossible for global import and export trophic levels to be theoretically identical. Similarly, Asche, et al. 23 Kroetz, et al. 57 also indicated that it was difficult to reconcile aquatic food production and trade data because of mismatches in species- versus product-level reporting and weight losses during processing. In this study, the average difference between trophic levels of imports and exports is only 0.02 (Supplementary Fig. 6c), and global human aquatic food trophic level trends before and after trade almost coincided (Supplementary Fig. 6d). Thus, theoretical exports could be a good proxy for original exports, and remaining
+unmatched exports would not affect global human aquatic food consumption patterns.
+
+Although we provided more detailed information on aquatic food production, trade, and consumption than the FAO Food Balance Sheets, which contain live weight of broad taxonomic groups, we acknowledge that our study has some limitations due to a lack of sufficient data. It would be inevitable that production and trade data in some countries are inconsistent or not perfectly processed, leading to over- or under-estimated apparent consumption and HATL. First, we couldn’t eliminate the impact of all fish species used in the production of fishmeal and fish oil on aquatic food consumption in all countries, although we had removed the world’s most important forage fish (i.e., Engraulis ringens) from the capture dataset. Thus, we may overestimate the apparent consumption of aquatic foods in several countries. Second, we didn’t handle the difference between the wild trophic level of farmed species versus actual trophic level based on what they were actually eating. These patterns not only exhibited strong temporal trends, but also showed significant spatial variations. Third, live weight conversion factors could vary over time and geographies due to differences in processing technologies, even for the same species.
+
+Data availability
+
+All datasets analyzed in this study are either included in the Supplementary Information or available in a figshare repository: https://doi.org/10.6084/m9.figshare.21692186.
+
+Code availability
+
+All codes used to conduct the study are available in a GitHub repository: https://github.com/zhaokangshun/Trade-is-improving-global-aquatic-food-consumption-.git.
+
+Acknowledgements
+
+This research was supported by the National Key R&D Program of China (grant no. 2018YFD0900904). K.Z. was funded by the China Scholarship Council. J.X. acknowledges the support received from the International Cooperation Project of the Chinese Academy of Sciences (grant no. 152342KYSB20190025) and the National Natural Science Foundations of China (grant no. 31872687).
+
+Author contributions
+
+K.Z. and J.X. conceived the idea. S.D.G. contributed to the study design. K.Z., S.D.G., M.Z.
+and J.X. contributed to the acquisition and analysis of data. K.Z., S.D.G. and J.G.M. contributed to the interpretation of results. K.Z., S.D.G., M.Z. and J.G.M. wrote and edited the manuscript.
+
+Competing interests
+All authors have no competing interests.
+
+Correspondence and requests for materials should be addressed to K.Z. or J.X.
+
+References
+1 Nations, U. Transforming our world: the 2030 agenda for sustainable development. New York: United Nations, Department of Economic and Social Affairs (2015).
+2 FAO. The State of World Fisheries and Aquaculture. Vol. 4 (FAO, 2020).
+3 Gephart, J. A. et al. Scenarios for global aquaculture and its role in human nutrition. Reviews in Fisheries Science & Aquaculture 29, 122-138, doi:10.1080/23308249.2020.1782342 (2020).
+4 Bellmann, C., Tipping, A. & Sumaila, U. R. Global trade in fish and fishery products: An overview. Marine Policy 69, 181-188, doi:10.1016/j.marpol.2015.12.019 (2016).
+5 Gephart, J. A. & Pace, M. L. Structure and evolution of the global seafood trade network. Environmental Research Letters 10, 125014, doi:10.1088/1748-9326/aae065 (2015).
+6 Golden, C. D. et al. Aquatic foods to nourish nations. Nature 598, 315-320, doi:10.1038/s41586-021-03917-1 (2021).
+7 Golden, C. D. et al. Nutrition: Fall in fish catch threatens human health. Nature 534, 317-320, doi:10.1038/534317a (2016).
+8 Hicks, C. C. et al. Harnessing global fisheries to tackle micronutrient deficiencies. Nature 574, 95-98, doi:10.1038/s41586-019-1592-6 (2019).
+9 Youn, S.-J. et al. Inland capture fishery contributions to global food security and threats to their future. Global Food Security 3, 142-148, doi:10.1016/j.gfs.2014.09.005 (2014).
+10 Free, C. M. et al. Expanding ocean food production under climate change. Nature 605, 490-496, doi:10.1038/s41586-022-04674-5 (2022).
+11 Lotze, H. K. et al. Global ensemble projections reveal trophic amplification of ocean biomass declines with climate change. Proceedings of the National Academy of Sciences 116, 12907-12912, doi:10.1073/pnas.1900194116 (2019).
+12 Tigchelaar, M. et al. Compound climate risks threaten aquatic food system benefits. Nature Food 2, 673-682, doi:10.1038/s43016-021-00368-9 (2021).
+13 Lam, V. W. et al. Climate change, tropical fisheries and prospects for sustainable development. Nature Reviews Earth & Environment 1, 440-454, doi:10.1038/s43017-020-0071-9 (2020).
+14 FAO. The State of World Fisheries and Aquaculture. Vol. 4 (FAO, 2022).
+15 Naylor, R. L. et al. A 20-year retrospective review of global aquaculture. Nature 591, 551-563, doi:10.1038/s41586-021-03308-6 (2021).
+16 Cressey, D. Future fish: the only way to meet the increasing demand for fish is through aquaculture. Daniel Cressey explores the challenges for fish farmers and what it means for dinner plates in 2030. Nature 458, 398-401, doi:10.1038/458398a (2009).
+17 Nash, K. L. et al. Trade and foreign fishing mediate global marine nutrient supply. Proceedings of
+the National Academy of Sciences **119**, e2120817119 (2022).
+Naylor, R. L. *et al.* Blue food demand across geographic and temporal scales. *Nat. Commun.* **12**, 1–14, doi:10.1038/s41467-021-25516-4 (2021).
+Xu, Z. *et al.* Impacts of international trade on global sustainable development. *Nature Sustainability* **3**, 964–971 (2020).
+Béné, C., Lawton, R. & Allison, E. H. “Trade matters in the fight against poverty”: Narratives, perceptions, and (lack of) evidence in the case of fish trade in Africa. *World Development* **38**, 933–954, doi:10.1016/j.worlddev.2009.12.010 (2010).
+Asche, F., Bellemare, M. F., Roheim, C., Smith, M. D. & Tveteras, S. Fair enough? Food security and the international trade of seafood. *World Development* **67**, 151–160, doi:10.1016/j.worlddev.2014.10.013 (2015).
+Kroetz, K. *et al.* Consequences of seafood mislabeling for marine populations and fisheries management. *Proceedings of the National Academy of Sciences* **117**, 30318–30323, doi:10.1073/pnas.2003741117 (2020).
+Asche, F. *et al.* China’s seafood imports—Not for domestic consumption? *Science* **375**, 386–388, doi:10.1073/pnas.2003741117 (2022).
+Burkholz, R. & Schweitzer, F. International crop trade networks: The impact of shocks and cascades. *Environmental Research Letters* **14**, 114013, doi:10.1088/1748-9326/ab4864 (2019).
+Chatellier, V. International trade in animal products and the place of the European Union: main trends over the last 20 years. *Animal* **15**, 100289, doi:10.1016/j.animal.2021.100289 (2021).
+Desiere, S., Hung, Y., Verbeke, W. & D’Haese, M. Assessing current and future meat and fish consumption in Sub-Saharan Africa: Learnings from FAO Food Balance Sheets and LSMS household survey data. *Global Food Security* **16**, 116–126, doi:10.1016/j.gfs.2017.12.004 (2018).
+Elton, C. *Animal Ecology* (*Sidgwick and Jackson, London*). (1927).
+Pauly, D. & Christensen, V. Primary production required to sustain global fisheries. *Nature* **374**, 255–257 (1995).
+Cottrell, R. S. *et al.* Time to rethink trophic levels in aquaculture policy. *Rev. Aquac.*, doi:10.1111/raq.12535 (2021).
+Zhao, K., Jorge Garcia, M., Zhang, H., Zhang, M. & Xu, J. Contemporary changes in structural dynamics and socioeconomic drivers of inland fishery in China. *Science of the Total Environment* **648**, 1527–1535, doi:10.1016/j.scitotenv.2018.08.196 (2018).
+Tacon, A. G., Metian, M., Turchini, G. M. & De Silva, S. S. Responsible aquaculture and trophic level implications to global fish supply. *Reviews in Fisheries Science* **18**, 94–105, doi:10.1080/10641260903325680 (2009).
+Zhang, W. *et al.* Aquaculture will continue to depend more on land than sea. *Nature* **603**, E2–E4, doi:10.1038/s41586-021-04331-3 (2022).
+Bonhommeau, S. *et al.* Eating up the world’s food web and the human trophic level. *Proceedings of the National Academy of Sciences* **110**, 20617–20620, doi:10.1073/pnas.1305827110 (2013).
+Roos, N., Wahab, M. A., Hossain, M. A. R. & Thilsted, S. H. Linking human nutrition and fisheries: incorporating micronutrient-dense, small indigenous fish species in carp polyculture production in Bangladesh. *Food and Nutrition Bulletin* **28**, S280–S293 (2007).
+Crona, B. *et al.* China at a crossroads: An analysis of China’s changing seafood production and consumption. *One Earth* **3**, 32–44, doi:10.1016/j.oneear.2020.06.013 (2020).
+Abbott, J. K., Willard, D. & Xu, J. Feeding the dragon: The evolution of China’s fishery imports. *Marine Policy* **133**, 104733, doi:10.1016/j.marpol.2021.104733 (2021).
+Bostock, J. et al. Aquaculture: global status and trends. Philosophical Transactions of the Royal Society B: Biological Sciences **365**, 2897–2912 (2010).
+
+Belton, B., Bush, S. R. & Little, D. C. Not just for the wealthy: Rethinking farmed fish consumption in the Global South. *Global Food Security* **16**, 85–92, doi:10.1016/j.gfs.2017.10.005 (2018).
+
+Fabinyi, M., Liu, N., Song, Q. & Li, R. Aquatic product consumption patterns and perceptions among the Chinese middle class. *Regional Studies in Marine Science* **7**, 1–9, doi:10.1016/j.rsma.2016.01.013 (2016).
+
+Liverpool-Tasie, L. S. O., Sanou, A., Reardon, T. & Belton, B. Demand for imported versus domestic fish in Nigeria. *Journal of Agricultural Economics* **72**, 782–804, doi:10.1111/1477-9552.12423 (2021).
+
+Kercher, J. & Shugart Jr, H. Trophic structure, effective trophic position, and connectivity in food webs. *The American Naturalist* **109**, 191–206 (1975).
+
+Kozlovsky, D. G. A critical evaluation of the trophic level concept. I. Ecological efficiencies. *Ecology* **49**, 48–60 (1968).
+
+Lindeman, R. L. The trophic-dynamic aspect of ecology. *Ecology* **23**, 399–417 (1942).
+
+Kurien, J. *Responsible fish trade and food security.* (Food & Agriculture Org., 2005).
+
+Golden, C. D. *et al.* Does aquaculture support the needs of nutritionally vulnerable nations? *Frontiers in Marine Science* **4**, 159, doi:10.3389/fmars.2017.00159 (2017).
+
+Cai, J. & Leung, P. Short-term projection of global fish demand and supply gaps. *FAO Fisheries and Aquaculture technical paper* (2017).
+
+Kidane, D. G. & Brækkan, E. H. Global seafood demand growth differences across regions, income levels, and time. *Marine Resource Economics* **36**, 289–305 (2021).
+
+Costello, C. *et al.* The future of food from the sea. *Nature* **588**, 95–100, doi:10.1038/s41586-020-2616-y (2020).
+
+Troell, M. *et al.* Does aquaculture add resilience to the global food system? *Proceedings of the National Academy of Sciences* **111**, 13257–13263, doi:10.1073/pnas.1404067111 (2014).
+
+Fisheries and Aquaculture Software. FishStatJ: Software for Fishery and Aquaculture Statistical Time Series (FAO Fisheries Division, 2021).
+
+FAO. *The State of World Fisheries and Aquaculture*. Vol. 4 (FAO, 2018).
+
+Fluet-Chouinard, E., Funge-Smith, S. & McIntyre, P. B. Global hidden harvest of freshwater fish revealed by household surveys. *Proceedings of the National Academy of Sciences* **115**, 7623–7628, doi:10.1073/pnas.1721097115 (2018).
+
+Hortle, K. G. Consumption and the yield of fish and other aquatic animals from the Lower Mekong Basin. 1–88 (Mekong River Commission, Vientiane, Lao PDR, 2007).
+
+Froese, R. & Pauly, D. *Fishbase*, <www.fishbase.org> (2021).
+
+Pauly, D. & Watson, R. Background and interpretation of the 'Marine Trophic Index'as a measure of biodiversity. *Philosophical Transactions of the Royal Society B: Biological Sciences* **360**, 415–423 (2005).
+
+Branch, T. A. *et al.* The trophic fingerprint of marine fisheries. *Nature* **468**, 431–435, doi:10.1038/nature09528 (2010).
+
+Kroetz, K. *et al.* Consequences of seafood mislabeling for marine populations and fisheries management. *Proceedings of the National Academy of Sciences* **117**, 30318–30323, doi:10.1073/pnas.2003741117 (2020).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SMdata1Liveweightandconversionfactorsoftradedcommodities.xlsx
+• SMdata2Trophiclevelidentificationproductiondata.xlsx
+• SMdata3Data2continued.xlsx
+• Supplementarymaterial.pdf
+• NCOMMS2327307rs.pdf

0be9eeae9b556d120e798bfcebf02432e565fe87c2a307aed440d75c79f4497a/peer_review/peer_review.md

@@ -0,0 +1,198 @@
+Peer Review File
+
+scMINER: a mutual information-based framework for clustering and hidden driver inference from single-cell transcriptomics data
+
+Corresponding Author: Dr Jiyang Yu
+
+This manuscript has been previously reviewed at another journal. This document only contains information relating to versions considered at Nature Communications.
+
+This file contains all reviewer reports in order by version, followed by all author rebuttals in order by version.
+
+Version 0:
+
+Reviewer comments:
+
+Reviewer #2
+
+(Remarks to the Author)
+The manuscript introduces scMINER, a framework that integrates existing tools for single-cell transcriptomic data analysis. While the paper does not propose fundamentally new methods, the integration of established approaches into a coherent pipeline is a valuable contribution, particularly in the context of unsupervised cell clustering, TF network inference, and SIG network analysis. By leveraging information theory to improve network inference, the approach offers an interesting alternative to traditional methods. Additionally, the authors provide comprehensive software documentation and tutorials, which significantly enhance the usability and accessibility of their approach for researchers. Nonetheless, despite the manuscript's overall clarity and the valuable contributions it makes, a few minor issues remain that warrant further attention.
+
+Minor Comments:
+
+1. Line 604: The equation for H(C_i,C_j) is missing.
+
+2. Line 608: This manuscript integrates several methods, and validating the optimality of each component in the combination is inherently difficult and nearly impossible. However, I am particularly concerned about the setting of the clustering modes. While the two modes, MICA-MDS and MICA-GE, offer distinct advantages for small and large datasets respectively, a direct comparison between the two strategies on a moderately sized dataset (e.g., ~5,000 cells) would be valuable to assess their performance trade-offs in terms of both accuracy and efficiency.
+
+3. Line 613: “Extended Data Fig. 2a” may be “Extended Data Fig. 3a”
+
+4. Line 730-734: Duplicated from Line 724-727.
+
+5. Line 740: In benchmarking analysis of cell clustering, the authors swept the resolution of Louvain from 0.1 to 4 to ensure that the predicted number of clusters matched the ground-truth labels. A question arises: how can the resolution be determined in real data analysis when ground-truth labels are unavailable, or when only the number of clusters is known? One potential approach could be to perform a binary search to determine the resolution, which might be more efficient than the traditional grid search. Additionally, it would be valuable to know if the authors used parallel processing to accelerate the grid search process from 0.1 to 4. If so, how does this compare in terms of efficiency to binary search in both benchmarking and real data analyses?
+
+6. Line 766, 784: While other methods use Louvain for benchmarking analysis of clustering, why did Scanpy opted for Leiden over Louvain? What advantages does Leiden offer in this context, and how does it compare to Louvain in terms of clustering performance?
+
+7. Fig. 2e: The legend is missing.
+8. Extended Data Fig. 4a: The y-axis should be adjusted to clearly display the performance of datasets excluding Klein.
+
+(Remarks on code availability)
+
+Reviewer #5
+
+(Remarks to the Author)
+I think the authors have addressed most of the comments from Reviewers 1 and 4, with the following points remaining:
+
+1. Reviewer 1 has a comment "The choices of algorithms need to be explained more clearly and justified." The authors responses to this comment, in terms of the choice of clustering methods, is rather high-level: "The clustering engine, MICA, incorporates methods that were selected after thorough evaluation of available algorithms for cell clustering, particularly in dimensionality reduction." It would be more clear if the authors can elaborate and present the specific evaluation that was performed. The manuscript does not provide more details on this. Although two benchmarking papers were cited, it is not straightforward which top performing methods were adapted in scMINER. Specific information needs to be provided to justify the choice of methods.
+
+2. In the response to Reviewer 4's comments, the authors stated that "Due to the size limitation of the R package, it's not feasible to include the input data, codes and documentation for all benchmarked datasets. Instead, we have used the PBMC14K dataset as a showcase and have integrated its input data and codes in the scMINER R package." I suggest that the authors explore the possibility of depositing raw or preprocessed data in Zenodo, for datasets presented in the paper.
+
+3. If submitting another revision, it would be helpful if the authors can highlight changes in the manuscript and in the responses refer to line numbers in the manuscript so that reviewers can quickly locate the changes corresponding to each response.
+
+(Remarks on code availability)
+I tested the interactive platform and basic functions are working.
+
+Reviewer #6
+
+(Remarks to the Author)
+Pan et al., detailed the development of a very useful tool, scMINER, that can perform cell clustering, transcription factor and signaling protein network inference analyses, and identify hidden drivers from single cell transcriptomic data analyses that not only have a much better performance compared to other existing methods that have some of the similar functions, but also provide novel functions, such as signaling protein network analysis. The most challenging tests are the applications on T cell subsets that normally can only be distinguished by transcriptional factors, mostly lowly expressed. The authors thoroughly addressed the reviewers’ critiques raised previously. I do not have any major concerns and only some minor points:
+1. A few abbreviations need to be defined.
+2. Grammatic errors need to be corrected.
+3. The abstract and the last paragraph of the intro missed an opportunity to highlight innovation and extensive validation that this study provides.
+
+(Remarks on code availability)
+Open Access This Peer Review 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 cases where reviewers are anonymous, credit should be given to 'Anonymous Referee' and the source.
+The images or other third party material in this Peer Review 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 https://creativecommons.org/licenses/by/4.0/.
+Point by point responses to reviewer’s comments:
+
+We sincerely appreciate the insightful and constructive comments from all reviewers. Below, we provide our detailed, point-by-point responses to their comments.
+
+Reviewer #2 (Remarks to the Author)
+The manuscript introduces scMINER, a framework that integrates existing tools for single-cell transcriptomic data analysis. While the paper does not propose fundamentally new methods, the integration of established approaches into a coherent pipeline is a valuable contribution, particularly in the context of unsupervised cell clustering, TF network inference, and SIG network analysis. By leveraging information theory to improve network inference, the approach offers an interesting alternative to traditional methods. Additionally, the authors provide comprehensive software documentation and tutorials, which significantly enhance the usability and accessibility of their approach for researchers. Nonetheless, despite the manuscript's overall clarity and the valuable contributions it makes, a few minor issues remain that warrant further attention.
+
+Minor Comments:
+
+1. Line 604: The equation for H(C_i,C_j)is missing.
+
+Response: We thank the reviewer for catching this. We have added the equation for \( H(C_i, C_j) \) in the revised manuscript (Line 602-605).
+
+2. Line 608: This manuscript integrates several methods, and validating the optimality of each component in the combination is inherently difficult and nearly impossible. However, I am particularly concerned about the setting of the clustering modes. While the two modes, MICA-MDS and MICA-GE, offer distinct advantages for small and large datasets respectively, a direct comparison between the two strategies on a moderately sized dataset (e.g., ~5,000 cells) would be valuable to assess their performance trade-offs in terms of both accuracy and efficiency.
+
+Response: We appreciate the reviewer’s insightful comments regarding this aspect of our study. As described in the manuscript (Line 120-122), we developed two operational modes to optimize clustering accuracy while ensuring scalability for large datasets. Our benchmarking results indicate that the MICA-MDS mode achieves superior clustering accuracy, whereas the MICA-GE mode excels in handling large datasets with efficiency.
+
+Following the reviewer’s suggestion, we conducted a comparative analysis of these two modes using a dataset of ~5,000 cells. This dataset was derived from the PBMC14K dataset by randomly selecting 750 cells for each of the seven cell types. Clustering analysis was performed under both modes using default parameters. The result, summarized in the table below, demonstrated that the MICA-MDS mode achieved an ARI of 0.833, slightly outperforming the MICA-GE mode with an ARI of 0.828. On the other hand, the MICA-GE mode exhibited significantly better computational efficiency
+than MICA-MDS, with a runtime of 734 seconds compared to 4,332 seconds for the MICA-MDS mode.
+
+<table>
+  <tr>
+    <th colspan="5">Response Table 1. Comparison of MICA-MDS and MICA-GE modes on a medium-size dataset.</th>
+  </tr>
+  <tr>
+    <th>Dataset</th>
+    <th>Mode</th>
+    <th>ARI</th>
+    <th>Running time</th>
+    <th>Max Memory</th>
+  </tr>
+  <tr>
+    <td>PBMC cells (750 cells * 7 cell types = 5250 cells)</td>
+    <td>MICA-MDS</td>
+    <td>0.833</td>
+    <td>4332 sec.</td>
+    <td>1235 MB</td>
+  </tr>
+  <tr>
+    <td></td>
+    <td>MICA-GE</td>
+    <td>0.828</td>
+    <td>734 sec.</td>
+    <td>3242 MB</td>
+  </tr>
+</table>
+
+We would also like to clarify that the 5,000-cell threshold for mode selection is an empirical guideline rather than a strict cutoff. The scMINER framework allows users to adjust this threshold based on their specific dataset characteristics and computational requirements.
+
+3. Line 613: “Extended Data Fig. 2a” may be “Extended Data Fig. 3a”
+
+Response: We thank the reviewer for catching this error. It has been corrected in the revised manuscript (Line 617).
+
+4. Line 730-734: Duplicated from Line 724-727.
+
+Response: We apologize for the confusion. Now the redundant text has been removed.
+
+5. Line 740: In benchmarking analysis of cell clustering, the authors swept the resolution of Louvain from 0.1 to 4 to ensure that the predicted number of clusters matched the ground-truth labels. A question arises: how can the resolution be determined in real data analysis when ground-truth labels are unavailable, or when only the number of clusters is known? One potential approach could be to perform a binary search to determine the resolution, which might be more efficient than the traditional grid search. Additionally, it would be valuable to know if the authors used parallel processing to accelerate the grid search process from 0.1 to 4. If so, how does this compare in terms of efficiency to binary search in both benchmarking and real data analyses?
+
+Response: We thank the reviewer for raising this fundamental question and for suggesting potential solutions. Determining the optimal resolution is indeed a critical and challenging task, as it requires balancing clustering granularity with biological interpretability. This balance usually cannot be achieved without subsequent analyses, such as marker gene expression and functional annotation. However, there are several approaches to address this challenge, and we have equipped scMINER with most of them, including but not limited to the following:
+
+1) Iterative Resolution Testing
+scMINER provides a user-friendly way to perform iterative resolution testing, simpler than what is offered by tools like Seurat and Scanpy. Users can define a range of resolutions by specifying the minimum, maximum and step-size. scMINER processes all resolutions in parallel to ensure computational efficiency
+and generates clustering results for all test resolutions, making it easier to compare and evaluate the outcomes.
+
+2) Clustering Evaluation Metrics
+scMINER supports various metrics for clustering evaluation. For example, the Silhouette Score can be calculated for each resolution, which helps assess how well the clusters were defined.
+
+3) Biological Interpretability Tools
+scMINER embeds several functions to enhance biological interpretability, such as marker gene identification and visualization. These features are designed to help users better understand the biological relevance of the clusters. More details can be found here: https://jyyulab.github.io/scMINER/bookdown/cell-type-annotation.html.
+
+For real-world datasets where ground-truth labels or the number of clusters are not available, we recommend users to leverage these functionalities to generate clustering results across multiple resolutions. By comparing the results for both granularity and biological interpretability, users can select the resolution that best balances these two aspects.
+
+Regarding the binary search algorithm, while it is an efficient method for finding a target value in a sorted array, it may not be suitable in this context. This is because the relationship between clustering resolution and the number of clusters is not always linear, making binary search less applicable for this specific problem.
+
+6. Line 766, 784: While other methods use Louvain for benchmarking analysis of clustering, why did Scanpy opted for Leiden over Louvain? What advantages does Leiden offer in this context, and how does it compare to Louvain in terms of clustering performance?
+
+Response: We thank the reviewer for raising these questions. The Leiden algorithm was introduced as an improvement over the Louvain algorithm (PMID: 30914743). It addresses several known limitations of the Louvain algorithm, including its inability to guarantee well-connected communities, suboptimal modularity, and the resolution limit. Theoretically, the Leiden algorithm offers better clustering accuracy than the Louvain method and is recommended in the standard tutorial of Scanpy (https://scanpy.readthedocs.io/en/stable/tutorials/basics/clustering.html). At the reviewer’s suggestion, we compared the clustering performance of these two methods using all 10 ground-truth datasets. For these analyses, all inputs and parameters were kept identical, except for the clustering method. As shown in Response Figure 1 below, the Leiden algorithm outperformed the Louvain method in clustering accuracy for 7 out of 10 datasets (left panel), while the Louvain method demonstrated slightly better clustering efficiency than the Leiden algorithm across most dataset (right panel). These results align with findings on clustering accuracy and efficiency from other benchmarking studies (PMID: 30914743).
+![Bar plots indicating the ARI values (left) and CPU times (right) of cell clustering across 10 ground-truth datasets by the Louvain(blue) and Leiden (brown) methods in Scanpy.](page_184_120_1207_393.png)
+
+Response Figure 1. Comparison of clustering algorithms in Scanpy. Bar plots indicating the ARI values (left) and CPU times (right) of cell clustering across 10 ground-truth datasets by the Louvain(blue) and Leiden (brown) methods in Scanpy.
+
+7. Fig. 2e: The legend is missing.
+
+Response: We thank the reviewer for catching this. We have added the legend (Line 1298-1301) in our revised manuscript.
+
+8. Extended Data Fig. 4a: The y-axis should be adjusted to clearly display the performance of datasets excluding Klein.
+
+Response: We appreciate the reviewer’s valuable suggestion. In response, we have updated the y-axis scale using the log10-transformation method to enhance the clarity and interpretability of the data. The revised figure can now be found in Supplementary Fig. 4a. For the reviewer’s convenience, we have also included the updated figure below (Response Figure 2).
+![Bar chart comparing running time (in seconds) of cell clustering analysis by scMINER and other five benchmarked algorithms across 10 ground-truth datasets.](page_232_186_1147_393.png)
+
+Response Figure 2 (new Supplementary Fig. 4a). Running time of cell clustering analysis by scMINER and other five benchmarked algorithms across 10 ground-truth datasets.
+Reviewer #5 (Remarks to the Author)
+
+I think the authors have addressed most of the comments from Reviewers 1 and 4, with the following points remaining:
+
+1. Reviewer 1 has a comment "The choices of algorithms need to be explained more clearly and justified." The authors responses to this comment, in terms of the choice of clustering methods, is rather high-level: "The clustering engine, MICA, incorporates methods that were selected after thorough evaluation of available algorithms for cell clustering, particularly in dimensionality reduction." It would be more clear if the authors can elaborate and present the specific evaluation that was performed. The manuscript does not provide more details on this. Although two benchmarking papers were cited, it is not straightforward which top performing methods were adapted in scMINER. Specific information needs to be provided to justify the choice of methods.
+
+Response: We thank the reviewer for bring this to our attention and apologize for any confusion caused by the previous description, which may have been unclear or insufficient. Here, we provide more details about the algorithms incorporated in scMINER and the specific evaluations conducted for each.
+
+• Algorithm #1: MICA, the cell clustering engine of scMINER
+When developing MICA, we focused on two key aspects of cell clustering analysis: cell-cell distance estimation and dimensionality reduction. For cell-cell distance estimation, we evaluated four widely-used metrics: mutual information (MI), Pearson correlation coefficient, Spearman rank correlation coefficient and Euclidean distance. We performed cell clustering analysis using these four metrics on four gold-standard ground-truth datasets (Yan, Pollen, Kolod, and Buettner), while keeping the other steps and parameters the same. As shown in Response Figure 3a,b (same to Fig. 2f and Supplementary Fig. 3a), MI achieved the highest ARI. Similarly, we assessed four common dimensionality reduction methods: multidimensional scaling (MDS), PCA, Laplacian eigenmaps and a combination of PCA and Laplacian (PCA/Laplacian) using the same ground-truth datasets. As shown in Response Figure 3c,d (same to Fig. 2g and Supplementary Fig. 3a), MDS outperformed the other methods in clustering accuracy. Based on these specific evaluations, we choose to integrate MI-based cell-cell distance estimation method and MDS-based dimensionality reduction method into MICA.
+Response Figure 3 (Fig. 2e,f and Supplementary Fig. 3a). Comparisons of cell-cell distance metrics (a,b) and dimensionality reduction methods (c,d) across four gold-standard ground-truth datasets.
+
+• Algorithm #2: SJARACNe, the network inference engine of scMINER.
+SJARACNe is re-engineered version of ARACNe, optimized to efficiently reconstruct gene networks from large datasets (PMID: 30388204). The accuracy of network inference by scMINER has been thoroughly benchmarked against three well-established tools-GENIE3, GRNBoost2 and PIDC-using datasets from different modalities, including ATAC-seq, Perturb-seq and CITE-seq. Due to space constrains, we have not included these results here, but they can be found in Fig. 4, Fig. 6, Supplementary Fig. 8 and Supplementary Fig. 9.
+
+We would also like to clarify that the top-performing methods discussed in the two benchmarking papers (PMID: 35135612; PMID: 31907445) are included for
+benchmarking purposes only. scMINER does not incorporate any components of these methods.
+
+2. In the response to Reviewer 4's comments, the authors stated that "Due to the size limitation of the R package, it's not feasible to include the input data, codes and documentation for all benchmarked datasets. Instead, we have used the PBMC14K dataset as a showcase and have integrated its input data and codes in the scMINER R package. " I suggest that the authors explore the possibility of depositing raw or preprocessed data in Zenodo, for datasets presented in the paper.
+
+Response: We thank the reviewer for this instructive suggestion. Following this suggestion, we have uploaded the raw and preprocessed data of all datasets used in clustering benchmarks to the Zenodo (DOI: 10.5281/zenodo.15040179).
+
+3. If submitting another revision, it would be helpful if the authors can highlight changes in the manuscript and in the responses refer to line numbers in the manuscript so that reviewers can quickly locate the changes corresponding to each response.
+
+Response: We thank the reviewer for this thoughtful suggestion and sincerely apologize for any inconvenience caused the absence of highlighted changes in the previous revised manuscript. The decision not to highlight changes was made because over 60% of the text had been refined, which would have resulted in excessive and potentially distracting annotations. However, in line with the reviewer’s suggestion, we have now specified the line numbers in our responses and highlighted all changes in a red font in the newly revised manuscript for easier review and clarity.
+
+4. Remarks on code availability: I tested the interactive platform and basic functions are working.
+
+Response: We thank the reviewer for testing and confirming the functions of our scMINER portal.
+Reviewer #6 (Remarks to the Author)
+
+Pan et al., detailed the development of a very useful tool, scMINER, that can perform cell clustering, transcription factor and signaling protein network inference analyses, and identify hidden drivers from single cell transcriptomic data analyses that not only have a much better performance compared to other existing methods that have some of the similar functions, but also provide novel functions, such as signaling protein network analysis. The most challenging tests are the applications on T cell subsets that normally can only be distinguished by transcriptional factors, mostly lowly expressed. The authors thoroughly addressed the reviewers’ critiques raised previously. I do not have any major concerns and only some minor points:
+
+1. A few abbreviations need to be defined.
+2. Grammatic errors need to be corrected.
+
+Response (to 1 and 2): We thank the reviewer for the reminders on abbreviations and grammatic errors. In response, we have thoroughly reviewed the manuscript to ensure that all abbreviations are clearly defined wherever necessary and that any grammatical errors have been carefully corrected. The relevant modifications have been made and can be found at the following locations: Line 118-119, Line 174-175, Line 1316-1318, Line 1327-1330, Line 1344, Line 1367-1368, Line 1382. For clarity and ease of review, these changes have been highlighted in red font in the revised manuscript.
+
+3. The abstract and the last paragraph of the intro missed an opportunity to highlight innovation and extensive validation that this study provides.
+
+Response: We sincerely thank the reviewer for this thoughtful suggestion and agree that emphasizing the innovation and extensive validations in the Abstract and Introduction would strengthen the manuscript. However, due to the space constrains, providing detailed descriptions in both sections might make them overly text-heavy. Instead, we have included a comprehensive discussion of these aspects in the first paragraph of the Discussion section (Line 471-483). We believe this placement effectively conveys the significance of our work while adhering to the formatting requirements.

0c0fdd1f1980a187e682cbf3e5e869fcf82f48c1168f402cd0132e7122adb5fe/preprint/preprint.md

@@ -0,0 +1,157 @@
+Direct bandgap quantum wells in hexagonal Silicon Germanium
+
+Erik Bakkers
+e.p.a.m.bakkers@tue.nl
+
+Eindhoven University of Technology   https://orcid.org/0000-0002-8264-6862
+Wouter Peeters
+Eindhoven University of Technology
+Victor Lange
+Eindhoven University of Technology
+Abderrezak Belabbes
+Friedrich Schiller University Jena
+Max van Hemert
+Eindhoven University of Technology   https://orcid.org/0009-0005-4791-2204
+Marvin Jansen
+Eindhoven University of Technology
+Riccardo Farina
+Eindhoven University of Technology
+Marvin Tilburg
+Eindhoven University of Technology   https://orcid.org/0000-0002-6177-4339
+Marcel Verheijen
+Eindhoven University of Technology   https://orcid.org/0000-0002-8749-7755
+Silvana Botti
+Friedrich Schiller University Jena
+Friedhelm Bechstedt
+Friedrich Schiller University Jena
+Jos Haverkort
+Eindhoven University of Technology   https://orcid.org/0000-0003-3051-673X
+DOI: https://doi.org/10.21203/rs.3.rs-3875137/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on June 19th, 2024. See the published version at https://doi.org/10.1038/s41467-024-49399-3.
+Direct bandgap quantum wells in hexagonal Silicon Germanium
+
+Wouter H.J. Peeters,1,* Victor T. van Lange,1,* Abderrazek Belabbes,2,3,* Max C. van Hemert,1 Marvin Marco Jansen,1 Riccardo Farina,1 Marvin A.J. van Tilburg,1 Marcel A. Verheijen,1,4 Silvana Botti,3,5 Friedhelm Bechstedt,3 Jos. E.M. Haverkort,1 and Erik P.A.M. Bakkers1,†
+
+1Department of Applied Physics, Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands
+2Department of Physics, Sultan Qaboos University, P.O. Box 123, Muscat, Oman
+3Institut für Festkörpertheorie und -optik, Friedrich-Schiller-Universität Jena, Jena, Germany
+4Eurofins Materials Science Netherlands BV, 5656 AE Eindhoven, The Netherlands
+5Faculty of Physics, Ruhr University Bochum, Universitätsstraße 150, D-44801 Bochum, Germany.
+
+(Dated: January 16, 2024)
+
+Silicon is indisputably the most advanced material for scalable electronics, but it is a poor choice for active photonic applications, due to its indirect band gap. The recently developed hexagonal (hex-)Si1−xGex semiconductor features a direct bandgap at least for x > 0.65, and the realization of quantum heterostructures would unlock new opportunities for advanced optoelectronic devices based on the SiGe system. Here, we demonstrate the synthesis and characterization of direct bandgap quantum wells (QWs) realized in the hex-Si1−xGex system. Photoluminescence experiments on hex-Ge/Si0.2Ge0.8 QWs demonstrate quantum confinement in the hex-Ge segment with type-I band alignment, showing light emission up to room temperature. Moreover, the tuning range of the QW emission energy can be extended using hex-Si1−xGex/Si1−yGey QWs with additional Si in the well. These experimental findings are supported with ab initio bandstructure calculations. A direct bandgap with type-I band alignment is pivotal for the development of novel low-dimensional light emitting devices based on hex-Si1−xGex alloys, which have been out of reach for this material system until now.
+
+Electronic devices based on silicon have been the driver for the revolution in information technology witnessed today. However, with their standard cubic-diamond crystal structure silicon, germanium and SiGe-alloys are all indirect band gap semiconductors, impeding the use of silicon-based materials for lasers and optical amplifiers for integrated photonics [1]. Several strategies have been investigated for integrating light emitting materials on silicon, including III-V [2, 3], GeSn [4–6], strained Ge [7, 8], and SiGe quantum wells and dots [9–15], but remain challenging due to various reasons. When transformed into the hexagonal crystal structure, the hex-Si1−xGex alloys [16] are direct bandgap semiconductors with the fundamental bandgap at the Γ-point. The hex-Si1−xGex compositional family shows tuneable light emission from 1.5 μm to 3.4 μm and features a nanosecond radiative lifetime. As such, hex-Si1−xGex stands out in the field of group IV photonics as a semiconductor with a relatively large energy difference between the direct and indirect conduction band minima, up to 0.3 eV for hex-Ge [17, 18].
+
+Quantum confinement in direct bandgap semiconductors has stood at the cradle of many photonic devices such as, single photon quantum dot (QD) emitters [19–22], quantum well (QW) lasers [23, 24] and colloidal QD LED display technology [25–27]. These direct bandgap low dimensional structures have been responsible for major advances in science, and constitute a toolbox for many optoelectronic, and quantum photonic devices[28, 29], allowing for tunable and narrow band emission, and the concentration of charge carriers.
+
+Here, we show the synthesis of hex-SiGe quantum wells and we demonstrate quantum confinement of the energy levels with type-I band alignment between the hex-Si1−xGex well (0.9 < x < 1.0) and the hex-Si1−yGey barrier (0.7 < y < 0.8). We observe broad tunability of the QW emission from 3.4 μm for hex-Ge/Si0.2Ge0.8 to 2.0 μm for hex-Si0.1Ge0.9/Si0.3Ge0.7, which may be further extended down towards 1.5 μm, the limits of which are a subject of future investigations. Most notably, we confirm direct bandgap emission by observing a sub-nanosecond photoluminescence lifetime, comparable with direct bandgap emission in bulk hex-SiGe.
+
+First, we calculate the electronic band structure of hex-Ge/Si0.25Ge0.75 multi-quantum well (MQW) structures, with (1100) interfaces, as superlattices (see Fig. 1a). The ab initio calculations are based on Density Functional Theory (DFT) for optimized atomic geometries and an approximate quasiparticle (QP) electronic structure approach to the bandstructures. The bandstructures are aligned employing the branch points (BPs) [30]. The Si0.25Ge0.75 barrier thickness is kept constant at 2 nm, i.e. 12 monolayers along the [1100] direction, while the Ge well thickness is varied between 4 and 15 nm. This barrier thickness is sufficient to prevent tunneling of electron and hole wavefunctions through the barriers [31]. As a consequence, the Ge layers in the MQW are electronically decoupled. The Ge/Si0.25Ge0.75 MQW structure is biaxially strained by −0.6% and −0.91% along the [1120] and [0001] directions respectively to more closely
+FIG. 1. Band structure calculations of hex-Ge/Si_{0.25}Ge_{0.75}. a) Hexagonal Ge/Si_{0.25}Ge_{0.75} heterostructure with (1100) interfaces. b) Bulk hexagonal Brillouin zone (BZ) and its projection onto the two-dimensional BZ of the (1\overline{1}00) interface. c) Direct bandgap band structure of hexagonal 4 nm Ge/2 nm Si_{0.25}Ge_{0.75} multiple quantum well structure (black lines) and bulk Si_{0.25}Ge_{0.75} (grey area) projected onto the two-dimensional Brillouin zone. The horizontal red line indicates the branching points of the two systems used as energy zero for alignment. d) Energies of the lowest electron and highest hole subband at \( \Gamma \) versus Ge thickness in the Ge/Si_{0.25}Ge_{0.75} heterostructures studied. They are compared with the lowest conduction and highest valence band of the bulk Si_{0.25}Ge_{0.75} barrier material. Dashed lines indicate the extrapolated band-states at infinite Ge well thickness. For comparison, also the energy position of the lowest indirect conduction band minimum outside \( \Gamma \) (dot-dashed line) is given.
+
+resemble pseudomorphic growth of a Ge QW between thick Si_{0.25}Ge_{0.75} barriers. These strain values have been obtained from X-ray diffraction experiments presented in the Extended data. The QP bandstructure of a (superlattice with a) 4 nm thick Ge layer is displayed in Fig. 1b-c, clearly showing a direct bandgap with a \( \Gamma \) minimum approximately 0.3 eV below the lowest indirect conduction band minimum which appears near the corner point \( \overline{M} \) of the Brillouin zone boundary. We plot the bandstructure of the MQW together with a background illustrating the projected bandstructure of the strained Si_{0.25}Ge_{0.75} bulk. The two bandstructures are aligned by their BPs. The bands of the Ge/Si_{0.25}Ge_{0.75} MQW, within the fundamental gap of the projected Si_{0.25}Ge_{0.75} bandstructure, describe subbands of electrons and holes, whose wavefunctions are both localized in the Ge layers. The localization of both the electron and hole wavefunctions in the Ge well (Extended data Fig. E2a) clearly indicates type-I band alignment. The type-I behaviour is confirmed by the energies for the highest hole subbands and lowest-energy electron subbands at the \( \Gamma \) point, which are presented versus the Ge layer thickness in Fig. 1d; corresponding bandstructures for MQW structures with thicker Ge layers are displayed in Extended data Fig. E2b of the Supplementary Material. The band offsets in the conduction band and the valence band of 0.14–0.15 eV are nearly equal (Extended data Fig. E2c). The band offsets can be employed as barrier heights in simplified rectangular finite QW models for electrons and holes. The confinement energy of electrons (holes) in the QW vary from 72 (36) to 31 (8) meV for thicknesses of 4–15 nm respectively. These values are much smaller than the offsets, and we therefore may approximate the system as an infinite QW. This approximation suggests effective electron (hole) masses of \( \approx 0.01\ m_0 \) (\( \approx 0.02\ m_0 \)), being much smaller than masses which are 0.076 \( m_0 \) (0.055 \( m_0 \)) that have been calculated for unstrained bulk hex-Ge [17].
+
+We have embedded coaxial hex-Ge quantum wells in hex-Si_{0.2}Ge_{0.8} barriers, grown epitaxially on the {1100} m-plane facets of wurtzite (WZ) GaAs core nanowires (NWs) [16, 32], as shown in Fig. 2a. The Scanning Electron Microscopy image in Fig.2b illustrates the dimensions of the resulting structures. The Ge QW thickness is varied between 10 and 30 nm by changing the growth time, while the SiGe barrier thickness exceeds 50 nm. Details about the material synthesis can be found in the extended data. The Ge/Si_{0.2}Ge_{0.8} QWs are characterized by cross-sectional Scanning Transmission Electron Microscopy (STEM) along two different zone-axes. When imaged along the [0001] zone axis, the Ge/Si_{0.2}Ge_{0.8} QW is visible as a hexagon; an example is given in Fig. 2c, and other data is shown in Extended data Fig. E3a. As illustrated, the thickness of the Ge QW varies between the different facets, which has been observed in all samples. Fluctuations in QW thickness have also been reported for other material systems [33, 34], possibly resulting in charge carrier localization in the thickest well [35]. An overview of the QW thicknesses per sample is presented in Extended data Fig. E3b. Moreover, the Si_{0.2}Ge_{0.8} has composition fluctuations, Si-rich spokes connect the corners of the GaAs with the outer corners of the NW [36]. When imaged along the [1\overline{1}20] zone axis, the Ge/Si_{0.2}Ge_{0.8} QW is visible as a vertical stripe in TEM (Fig. 2d). The thickness of the QW is not constant along the length of the NW (Extended data Fig. E4), and the roughness on the {1100} interface between Ge/Si_{0.2}Ge_{0.8} is estimated from Fig. 2d to be a few nm. Additionally, the [1\overline{1}20] zone axis allows to distinguish between hexagonal and cubic stacking. The hexagonal stacking is not continuous along the [0001] direction, but is segmented due to the inclusion of cubic defects. These I3 defects nucleate either on the GaAs-Si_{0.2}Ge_{0.8} interface, or at random positions in the shell [37]. An example is indicated with the arrow in (Fig. 2d). A statistical analysis of the atomic stacking shows a broad distribution
+FIG. 2. Structural properties of the studied Ge/Si_{0.2}Ge_{0.8} QWs, all analyses are from NWs from the same sample with (12 ± 3) nm (2.0 min) Ge/Si_{0.2}Ge_{0.8} QW. a) Schematic illustration of the GaAs/Si_{0.2}Ge_{0.8}/Ge/Si_{0.2}Ge_{0.8} core-multishell nanowires. All interfaces are orthogonal to ⟨110⟩ directions. b) 30-degree tilted scanning electron micrograph of a nanowire array. c) False coloured HAADF-STEM image of a cross-sectional lamella, viewing the Ge QW along the [0001] zone-axis. Inset shows that Ge QWs on neighbouring facets have different thicknesses. d) False coloured HAADF-STEM image of a cross-sectional lamella, viewing the QW along the [1120] zone axis. The core of the NW is on the left. Locations with local hexagonal (ABABA, blue), cubic (ABCA, green), and twinned cubic (ABCBA, pink) stacking are indicated with circles. The pink arrow highlights a defect that is nucleated in the Ge QW. e) X-ray diffraction reciprocal space map around the hexagonal [10\overline{1}5] reflection. The peak position does not match Vegard’s rule (dashed line), indicating pseudomorphic strain relaxation.
+
+in the length of segments with the hexagonal stacking. In contrast, only narrow segments of coherent cubic stacking\textsuperscript{185} are observed (Extended data Fig. E5a). X-ray diffraction (XRD) is used to study the crystalline quality and lattice constants from a large ensemble of NWs. The diffraction spectra of all samples are very similar, indicating comparable crystalline quality between samples (Extended data\textsuperscript{190} Fig. E5b). A reciprocal space map around the hexagonal [10\overline{1}5] reflection shows a single peak (Fig. 2e), indicating that there is pseudomorphic strain relaxation in the structures, despite the 0.8% lattice mismatch between Ge and Si_{0.2}Ge_{0.8}. Increasing the Ge thickness does not significantly influence the lattice parameters of the NWs (Extended data Fig. E6a), so all studied samples have comparable c-lattice constants in the Ge QW. However, the Ge is compressed along the (1120) and (0001) directions, and pseudomorphic strain relaxation in the Ge QW\textsuperscript{200} results in an increased radial relaxation with increasing thickness, as confirmed by the Geometric Phase Analysis (GPA) of TEM images (Extended data Fig. E6b).
+
+The optical properties of the QW samples have been studied by low-temperature photoluminescence (PL) as\textsuperscript{205} a function of the QW thickness in Fig. 3. We observe that the emission energy consistently blue-shifts with decreasing QW thickness demonstrating increasing quantum confinement with decreasing thickness. Moreover, all QW emission peaks are positioned between the emission originating from the bulk hex-Ge and hex-Si_{0.2}Ge_{0.8} reference samples, thus providing experimental evidence for type-I band alignment. We note that for type-II band-alignment, one would expect emission below the energy of (strained) bulk hex-Ge [38]. The width of the QW emission peaks is larger than that of the reference samples and for some samples multiple peaks have been observed; this is probably due to fluctuations in QW thickness and, for the wider QWs, the presence of the second confined level. The intensity of the QW emission is exceeding that of the reference sample (See Extended data Fig. E7a), indicating that many carriers diffuse towards the QWs. In Fig 3b, the experimental results are compared with the calculated bandgaps (black dots) by ab initio calculations as already shown in Fig 1c for 4, 6, 8, 11 and 15 nm QWs. For properly comparing theory with experiment, we shift the calculated theoretical bandgap (\approx 0.30 eV) [17] with +60 meV to match the experimentally observed bandgap of bulk hex-Ge (\approx 0.36 eV) [16], which is within the er-
+FIG. 3. Quantum confinement in hex-Ge/Si_{0.2}Ge_{0.8} QWs. a) Ge/Si_{0.2}Ge_{0.8} PL spectrum for varying growth time at low temperature (\( T \approx 4\text{K} \)) and low excitation density (\( P \leq 65\text{W cm}^{-2} \)). b) The PL emission versus the QW thicknesses \( t_{QW} \) determined from TEM together with the confinement energy predicted from theory shifted up by 60 meV to account for the difference in the theoretical and experimental bandgap of the hex-Ge. The dashed line shows the confinement energies using a simple finite QW model. We also include the reference spectra of bulk-Ge and the bulk Si_{0.2}Ge_{0.8} barrier as horizontal lines with the FWHM of the spectra shown as horizontal grey bars.
+
+ror margin of the *ab initio* DFT calculations (\( \approx 0.1\text{eV} \) or 25\% [39]). Based on the predicted band offsets and effective masses [17, 40, 41], the emission energy versus thickness is also calculated using a conventional finite QW model (dashed line) [42] to generalize the prediction to any QW thickness. A qualitative agreement between theory and experiment is obtained, but the experimental emission energies are all higher than the predicted values. The deviation can have several possible explanations. The Ge QW thicknesses, measured from TEM images, are slightly overestimated (See Extended data Fig. E4). Moreover, likely a 1–2\% Si is incorporated in the Ge QWs, which elevates the bandgap of the well. Lastly, the potential at the QW-Barrier interface could be closer to triangular than to a step function due to interdiffusion of Si and Ge, which increases the confinement energies.
+
+The optoelectronic properties of the Ge/Si_{0.2}Ge_{0.8} QWs are investigated in more detail by power- and temperature-dependent photoluminescence spectroscopy. We focus here on two specific samples: (i) a relatively thin (10 ± 4) nm QW showing single peak emission with strong confinement and (ii) a thick (24 ± 7) nm QW with small confinement energy and a large separation between the confinement level in the QW and the barrier, as shown in Fig. 4a,b and c,d respectively. Besides the emission being between the hex-Ge reference and the Si_{0.2}Ge_{0.8} barrier, as mentioned before, we observe that the emission peak energy is nearly independent of the excitation density. At low excitation densities a slight <5 meV blue shift is observed, followed by a red shift at high excitation. These shifts are likely due to Burstein-Moss bandfilling and bandgap renormalization (BGR).
+
+Importantly, we do not observe the significant blue shift expected for a type-II QW. The absence of such a blue-shift provides additional evidence for a type-I band offset. Similar trends have been observed for the other QW samples[43]. Additionally we show higher light-in light-out curve slopes compared to the barrier reference sample for all quantum wells in Extended data Fig. E7b, indicating that the carriers are shielded from non-radiative loss mechanisms. The spectra of the thick (24 ± 7) nm QW sample are plotted in Fig. 4c,d as a function of temperature. Notably, room temperature emission from an ensemble of NWs with a *single* coaxial hex-Ge/Si_{0.2}Ge_{0.8} QW is demonstrated. The spectra contain two peaks for all temperatures, which can be attributed to the presence of distinct QW thicknesses e.g. at different facets of the nanowire shells or to a second confined level within the quantum well. In the range \( T = 2.4–100\text{K} \) the relative magnitude of the higher energy peak increases, which is likely due to the de-trapping of carriers from the potential landscape due to alloy fluctuations in the SiGe barrier, allowing more carriers to diffuse to the QW, while the lower energy QW level is already fully occupied. Above 250 K the low energy peak again becomes more dominant, which is likely due to a higher probability of carrier evaporation from the higher energy QW level, while also allowing the carriers to be even more mobile to find the lowest energy states. The temperature dependence of the integrated PL intensity is shown in Fig. 4d and shows a monotonous decay of the intensity with temperature, which is a strong indication for direct bandgap emission [16]. Moreover, the intensity of the QW emission outperforms the emission of the bulk hex-Si_{0.2}Ge_{0.8} reference
+FIG. 4. Type-I band alignment in hex-Ge/Si0.2Ge0.8 QWs. a) The (10 ± 4) nm (2.5 min) low temperature (\( T \approx 4\text{K} \)) QW photoluminescence spectrum as a function of excitation density showing a constant lineshape over two orders of magnitude with the peak position in between the bulk-Ge and Si0.2Ge0.8 barrier reference measurements, b) the emission peak energy of the 2.5 min QW showing a nearly constant magnitude through excitation density. Initially the peak blue-shifts due to band-filling of the QW and then red-shift around 100 W cm\(^{-2}\) likely due to Bandgap renormalization (BGR).\(^{320}\) c) The (24 ± 7) nm (9 min) QW sample as a function of temperature showing emission up to room temperature. d) The Arrhenius plot of the 9 min QW and SiGe barrier reference samples measured at an excitation density of 0.88 kW cm\(^{-2}\). It can be seen that the temperature behaviour of the QW exceeds the bulk hex-Si0.2Ge0.8 reference.
+
+sample at elevated temperatures (observed for all QW samples shown in Extended data Fig. E7c), which is an important advantage for devices e.g. a hex-Ge/Si0.2Ge0.8 QW laser.
+
+Having confirmed quantum confinement and wavelength tunability of emission from the hex-Ge/Si0.2Ge0.8 QWs, we subsequently like to demonstrate type-I confine-
+
+ment in hex-Si0.1Ge0.9/Si0.3Ge0.7 QWs that emit light at an even higher energy by making use of alloys with a larger bandgap [16]. These hex-Si0.1Ge0.9/Si0.3Ge0.7 QWs are realized as coaxial nanowire shells, similar to those presented in Fig. 2a-c. A cross-sectional view of the (5 ± 1) nm Si0.1Ge0.9/Si0.3Ge0.7 QW is presented in Fig. 5a, and an overview of all studied Si0.1Ge0.9/Si0.3Ge0.7 QWs is presented in Extended data Fig. E8. There are two main differences compared to the Ge/Si0.2Ge0.8 system studied. Additional radial contrast lines, that do not terminate at the NW corners, are recognizable in the TEM image. These lines correspond to dislocations, whose occurrence is correlated with the lattice mismatch between the wurtzite GaAs core and the hex-Si1−xGex shell. Secondly, there is a compositional gradient in the Si1−xGex barrier, where the Si concentration increases with increasing distance to the GaAs core (see Extended data Fig E9). Both effects arise from the lattice mismatch in this system, which is either relaxed through dislocations or mitigated by forming a self-assembled compositional gradient buffer layer.
+
+The photoluminescence emission from the Si0.1Ge0.9/Si0.3Ge0.7 QW is between the emission of the bulk Si0.1Ge0.9 well material, and the barrier material, as shown in (Fig. 5b), signifying a type-I band offset also for these compositions. We again fit the observed QW emission energies with the conventional finite QW model, showing qualitative agreement in Fig. 5c. This suggests that the band alignment of the broader family of the Si1−xGex/Si1−yGey QWs is of type-I nature.
+
+We finally measure the carrier recombination lifetime using a Time-Correlated Single Photon Counting (TC-SPC) system employing a Superconducting Nanowire Single Photon Detector (SNSPD) for the (5 ± 1) nm QW (Single nanowire spectrum shown in Extended data Fig. E10a). We measure the PL lifetime (at \( \approx 4\text{K} \)) for varying laser fluence (Fig. 5d). Importantly, we observe an initial carrier lifetime of \( \approx 1 \) ns for the lowest fluence (Full time decays are provided in Extended data Fig. E10b), confirming direct bandgap emission. Notably, this lifetime falls within the same range as that reported by Fadaly et. al. [16, 44] for bulk hex-SiGe nanowires.
+
+In conclusion, we have grown coaxial hex-Ge/Si0.2Ge0.8 and Si0.1Ge0.9/Si0.3Ge0.7 QWs showing direct bandgap light emission. We observe clear quantum confinement combined with type-I alignment. Our results are unlocking the hex-Si1−xGex/Si1−yGey system for different low-dimensional devices, such as quantum well lasers, optical amplifiers and single photon sources using Si1−xGex alloys. This has the potential to revolutionize the Si-Photonics platform with a CMOS compatible light source.
+
+Acknowledgements
+We thank R. van Veldhoven and M.G. van Dijstelbloem for the technical support of the MOVPE reactor. We thank Orson A.H. van der Molen for the GPA analysis. This project received funding from the European
+FIG. 5. Studies of hex-Si0.1Ge0.9/Si0.3Ge0.7 QWs. a) False coloured HAADF-STEM of a cross-sectional lamella, viewing the (5 ± 1) nm (5 min) Si0.1Ge0.9/Si0.3Ge0.7 QW in the [0001] zone-axis. b) Background corrected photoluminescence spectra for varying QW growth time at low temperature (≈4 K) and high excitation density <0.88 kW cm⁻². Reference spectra of bulk Si0.1Ge0.9 and Si0.3Ge0.7 are included. c) The PL emission versus the QW thicknesses t_QW determined from TEM. Spectra of the Si0.1Ge0.9 well and Si0.3Ge0.7 barrier alloys are included as horizontal lines with the FWHM of the spectra as horizontal grey bars. A simple finite QW model is calculated for this heterostructure which shows reasonable agreement with the experiment. d) Initial QW lifetime measured using TCSPC for the (5 ± 1) nm QW for varying laser fluence.
+
+Union’s Horizon 2020 research and innovation program under grant agreement number 964191 (Opto Silicon), the Dutch Organization for Scientific Research (NWO) in the Zwaartekracht Project (Grant No. 024.002.033), the Mat4Sus project (Grant No.739.017.002) and Solliance385 and the Dutch province of Noord-Brabant for funding the TEM facility.
+
+Author contributions
+W.H.J.P., M.M.J. carried out the growth of hex-Ge/SiGe360 quantum wells. W.H.J.P. analysed the data. V.T.L. and M.C.H. carried out the photoluminescence spectroscopy and analysed the optical data. R.F. and M.A.J.T. performed time-resolved spectroscopy on single quantum well nanowires. A.B. and F.B. performed the DFT365 calculations. W.H.J.P. performed the XRD measurements. W.H.J.P. and M.M.J. performed the FIB cuts and M.A.V. performed the TEM analysis. S.B., F.B., J.E.M.H. and E.P.A.M.B. supervised the project. F.B. contributed to the interpretation of data and W.H.J.P., V.T.L., F.B., J.E.M.H. and E.P.A.M.B. contributed to the writing of the manuscript. All authors discussed the results and commented on the manuscript
+
+Competing interests
+The authors declare no competing interests.
+
+Additional information
+Correspondence and requests for materials should be addressed to E.P.A.M.B. Reprints and permissions information is available at http://www.nature.com/reprints.
+
+Data availability
+All data underlying this study are available from (to be done)
+
+[1] R. Soref, The Past, Present, and Future of Silicon Photonics, IEEE Journal of Selected Topics in Quantum Electronics 12, 1678 (2006).
+[2] Y. Shi, L. C. Kreuzer, N. C. Gerhardt, M. Pantou-vaki, J. Van Campenhout, M. Baryshnikova, R. Langer, D. Van Thourhout, and B. Kunert, Time-resolved photoluminescence characterization of ingaas/gaas nano-ridges monolithically grown on 300 mm si substrates, Journal of Applied Physics 127 (2020).
+[3] A. W. Elshaari, W. Pernice, K. Srinivasan, O. Benson, and V. Zwiller, Hybrid integrated quantum photonic circuits, Nature Photonics 14, 285 (2020).
+[4] S. Wirths, R. Geiger, N. Von Den Driesch, G. Mussler, T. Stoica, S. Mantl, Z. Ikonic, M. Luysberg, S. Chiussi, J. M. Hartmann, H. Sigg, J. Faist, D. Buca, and D. Grützmacher, Lasing in direct-bandgap GeSn alloy grown on Si, Nature Photonics 9, 88 (2015).
+[5] A. Elbaz, D. Buca, N. von den Driesch, K. Pantzas, G. Patriarche, N. Zerounian, E. Herth, X. Checoury, S. Sauvage, I. Sagnes, A. Foti, R. Ossikovski, J.-M. Hartmann, F. Boeuf, Z. Ikonic, P. Boucaud, D. Grützmacher, and M. El Kurdi, Ultra-low-threshold continuous-wave and pulsed lasing in tensile-strained GeSn alloys, Nature Photonics 14, 375 (2020).
+[6] Y. Kim, S. Assali, H.-J. Joo, S. Koelling, M. Chen, L. Luo, X. Shi, D. Burt, Z. Ikonic, D. Nam, and O. Moutanabbir, Short-wave infrared cavity resonances in a single GeSn nanowire, Nature Communications 14, 4393 (2023).
+[7] F. T. Armand Pilon, A. Lyasota, Y.-M. Niquet, V. Re-
+boud, V. Calvo, N. Pauc, J. Widiez, C. Bonzon, J. M. Hartmann, A. Chelnokov, J. Faist, and H. Sigg, Lasing in strained germanium microbridges, Nature Communications 10, 2724 (2019).
+[8] F. T. Armand Pilon, Y. M. Niquet, J. Chretien, N. Pauc, V. Reboud, V. Calvo, J. Widiez, J. M. Hartmann, A. Chelnokov, J. Faist, and H. Sigg, Investigation of lasing in highly strained germanium at the crossover to direct band gap, Physical Review Research 4, 033050 (2022).
+[9] Y. H. Kuo, Y. K. Lee, Y. Ge, S. Ren, J. E. Roth, T. I. Kamins, D. A. Miller, and J. S. Harris, Strong quantum-confined Stark effect in germanium quantum-well structures on silicon, Nature 437, 1334 (2005).
+[10] E. Gatti, E. Grilli, M. Guzzi, D. Chrastina, G. Isella, and H. Von Känel, Room temperature photoluminescence of ge multiple quantum wells with ge-rich barriers, Applied physics letters 98 (2011).
+[11] F. Pezzoli, F. Isa, G. Isella, C. V. Falub, T. Kreiliger, M. Salvalaglio, R. Bergamaschini, E. Grilli, M. Guzzi, H. Von Känel, and L. Miglio, Ge Crystals on Si Show Their Light, Physical Review Applied 1, 1 (2014).
+[12] P. Chaisakul, D. Marris-Morini, J. Frigerio, D. Chrastina, M. S. Roufied, S. Cecchi, P. Crozat, G. Isella, and L. Vivien, Integrated germanium optical interconnects on silicon substrates, Nature Photonics 8, 482 (2014).
+[13] L. E. A. Stehouwer, A. Tosato, D. Degli Esposti, D. Costa, M. Veldhorst, A. Sammak, and G. Scappucci, Germanium wafers for strained quantum wells with low disorder, Applied Physics Letters 123, 92101 (2023).
+[14] T. A. Hutchins-Delgado, A. J. Miller, R. Scott, P. Lu, D. R. Luhman, and T.-M. Lu, Characterization of Shallow, Undoped Ge/SiGe Quantum Wells Commercially Grown on 8-in. (100) Si Wafers, ACS Applied Electronic Materials 4, 4482 (2022).
+[15] P. Chaisakul, V. Vakarin, J. Frigerio, D. Chrastina, G. Isella, L. Vivien, and D. Marris-Morini, Recent progress on Ge/SiGe quantum well optical modulators, detectors, and emitters for optical interconnects, Photonics 6, 24 (2019).
+[16] E. M. T. Fadaly, A. Dijkstra, J. R. Suckert, D. Ziss, M. A. van Tilburg, C. Mao, Y. Ren, V. T. van Lange, K. Korzun, S. Kölling, et al., Direct-bandgap emission from hexagonal ge and sige alloys, Nature 580, 205 (2020).
+[17] C. Rödl, J. Furthmüller, J. R. Suckert, V. Armuzza, F. Bechstedt, and S. Botti, Accurate electronic and optical properties of hexagonal germanium for optoelectronic applications, Phys.Rev.Materials 3, 034602 (2019).
+[18] P. Borlido, J. R. Suckert, J. Furthmüller, F. Bechstedt, S. Botti, and C. Rödl, From pseudo-direct hexagonal germanium to direct silicon-germanium alloys, Physical Review Materials 5, 114604 (2021).
+[19] P. Michler, A. Kiraz, C. Becher, W. Schoenfeld, P. Petroff, L. Zhang, E. Hu, and A. Imamoglu, A quantum dot single-photon turnstile device, science 290, 2282 (2000).
+[20] J. Wang, F. Sciarrino, A. Laing, and M. G. Thompson, Integrated photonic quantum technologies, Nature Photonics 14, 273 (2020).
+[21] C. L. Salter, R. M. Stevenson, I. Farrer, C. A. Nicoll, D. A. Ritchie, and A. J. Shields, An entangled-light-emitting diode, Nature 465, 594 (2010).
+[22] T. McNunkin, B. Harpt, Y. Feng, M. P. Losert, R. Rahman, J. P. Dodson, M. A. Wolfe, D. E. Savage, M. G. Lagally, S. N. Coppersmith, M. Friesen, R. Joynt, and M. A. Eriksson, SiGe quantum wells with oscillating Ge concentrations for quantum dot qubits, Nature Communications 13, 7777 (2022).
+[23] Y. Arakawa and H. Sakaki, Multidimensional quantum well laser and temperature dependence of its threshold current, Applied physics letters 40, 939 (1982).
+[24] P. Täschler, M. Bertrand, B. Schneider, M. Singleton, P. Jouy, F. Kapsalidis, M. Beck, and J. Faist, Femtosecond pulses from a mid-infrared quantum cascade laser, Nature Photonics 15, 919 (2021).
+[25] S. Coe, W.-K. Woo, M. Bawendi, and V. Bulović, Electroluminescence from single monolayers of nanocrystals in molecular organic devices, Nature 420, 800 (2002).
+[26] V. Wood and V. Bulović, Colloidal quantum dot light-emitting devices, Nano Reviews 1, 5202 (2010).
+[27] S. Pradhan, F. Di Stasio, Y. Bi, S. Gupta, S. Christodoulou, A. Stavrinadis, and G. Konstantatos, High-efficiency colloidal quantum dot infrared light-emitting diodes via engineering at the suprananocrystalline level, Nature Nanotechnology 14, 72 (2019).
+[28] N. W. Hendrickx, W. I. Lawrie, M. Russ, F. van Riggelen, S. L. de Snoo, R. N. Schouten, A. Sammak, G. Scappucci, and M. Veldhorst, A four-qubit germanium quantum processor, Nature 591, 580 (2021).
+[29] E. Pelucchi, G. Fagas, I. Aharonovich, D. Englund, E. Figueroa, Q. Gong, H. Hannes, J. Liu, C.-Y. Lu, N. Matsuda, J.-W. Pan, F. Schreck, F. Sciarrino, C. Silberhorn, J. Wang, and K. D. Jöns, The potential and global outlook of integrated photonics for quantum technologies, Nature Reviews Physics 4, 194 (2021).
+[30] A. Schleife, F. Fuchs, C. Rödl, J. Furthmüller, and F. Bechstedt, Branch-point energies and band discontinuities of iii-nitrides and iii-ii-oxides from quasiparticle band-structure calculations, Applied Physics Letters 94 (2009).
+[31] A. Belabbes, S. Botti, and F. Bechstedt, Band lineup at hexagonal SiₓGe₁₋ₓ/SiₓGe₁₋ₓ alloy interfaces, Physical Review B 106, 085303 (2022).
+[32] H. I. T. Hauge, S. Conesa-Boj, M. A. Verheijen, S. Koelling, and E. P. Bakkers, Single-crystalline hexagonal silicon–germanium, Nano Letters 17, 85 (2017).
+[33] D. Saxena, N. Jiang, X. Yuan, S. Mokkapati, Y. Guo, H. H. Tan, and C. Jagadish, Design and Room-Temperature Operation of GaAs/AlGaAs Multiple Quantum Well Nanowire Lasers, Nano Letters 16, 5080 (2016).
+[34] S. Skalsky, Y. Zhang, J. A. Alanis, H. A. Fonseka, A. M. Sanchez, H. Liu, and P. Parkinson, Heterostructure and Q-factor engineering for low-threshold and persistent nanowire lasing, Light: Science & Applications 9, 43 (2020).
+[35] M. M. Sonner, A. Sitek, L. Janker, D. Rudolph, D. Ruhstorfer, M. Döblinger, A. Manolescu, G. Abstreiter, J. J. Finley, A. Wixforth, et al., Breakdown of corner states and carrier localization by monolayer fluctuations in radial nanowire quantum wells, Nano Letters 19, 3336 (2019).
+[36] R. Bergamaschini, R. C. Plantenga, M. Albani, E. Scalise, Y. Ren, H. I. T. Hauge, S. Kölling, F. Montalenti, E. P. Bakkers, M. A. Verheijen, et al., Prismatic ge-rich inclusions in the hexagonal sige shell of gap–sige nanowires by controlled faceting, Nanoscale 13, 9436
+(2021).
+[37] E. M. Fadaly, A. Marzegalli, Y. Ren, L. Sun, A. Dijkstra, D. De Matteis, E. Scalise, A. Sarikov, M. De Luca, R. Rurali, et al., Unveiling planar defects in hexagonal group iv materials, Nano Letters 21, 3619 (2021).
+[38] Z. Zheng, X. Zu, Y. Zhang, and W. Zhou, Rational design of type-II nano-heterojunctions for nanoscale optoelectronics, Materials Today Physics 15, 100262 (2020).
+[39] W. Li, C. F. J. Walther, A. Kuc, and T. Heine, Density Functional Theory and Beyond for Band-Gap Screening Performance for Transition-Metal Oxides and Dichalcogenides, Journal of Chemical Theory and Computation 9, 2950 (2013).
+[40] A. De and C. E. Pryor, Electronic structure and optical properties of Si, Ge and diamond in the lonsdaleite phase, Journal of Physics: Condensed Matter 26, 045801 (2014).
+[41] J. Sink and C. Pryor, Empirical tight-binding parameters for wurtzite group III–V(non-nitride) and IV materials, AIP Advances 13, 025354 (2023).
+[42] M. Fox and R. Ispasoiu, Quantum Wells, Superlattices, and Band-Gap Engineering, Springer Handbooks , 1021 (2007).
+[43] M. Dinu, J. E. Cunningham, F. Quochi, and J. Shah, Optical properties of strained antimonide-based heterostructures, Journal of Applied Physics 94, 1506 (2003).
+[44] M. A. J. van Tilburg, W. H. J. Peeters, M. Vettori, V. T. van Lange, E. P. A. M. Bakkers, and J. E. M. Haverkort, Polarized emission from hexagonal-silicon–germanium nanowires, Journal of Applied Physics 133, 065702 (2023).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• HexGeSiGeQWsExtendedData.pdf

0c34a8e31faa7739408a5ccd9d2f5e997e03e8f8f88c468ddded81aa6cf416eb/peer_review/peer_review.md

@@ -0,0 +1,113 @@
+Peer Review File
+
+Diversity and evolution of the vertebrate chemoreceptor gene repertoire
+
+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/.
+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:
+I acknowledge the great effort that the authors have put through the revision of this manuscript. Unfortunately, one of main criticism (that is, the lack of functional validation of bitter or sweet genes) from the first round of review wasn't really taken into account.
+
+Reviewer #2:
+Remarks to the Author:
+The authors present a comprehensive study of chemosensory receptor gene numbers (OR, TAAR, V1R, V2R, T1R, T2R) in over thousand vertebrate genomes. The methods for identifying the receptor repertoires are solid. Several of the results stated in the abstract such as dynamic evolution, lineage-specific expansions and losses of chemosensory receptor gene families are not new and have in fact been reported in many previous publications. However, the clear advance of the present study is that now the conclusions are based on a much larger dataset. Furthermore, the enlarged data set allowed more fine-grained correlation with ecological parameters than previously possible. These correlations have now been included in a new Figure 4, which is a clear improvement above the original version of this manuscript. The correlation coefficients are often relatively small, but that may be expected in large scale species sampling. Overall the revisions have improved the manuscript considerably. It does contain a wealth of data points and as such constitutes an important addition to the knowledge about chemosensory receptors.
+
+The manuscript is generally not easy to read due to the listing of many numbers, genes and Latin species names in the main text. This has not changed from the original version. I would strongly suggest to put e.g. the association values (correlation coefficient, p values) in a dedicated table, this would increase readability of the corresponding segment considerably.
+
+The newly introduced text about correlations between ecology and gene number is a welcome improvement. However it could be better integrated. For example, in new text line 319-337 the authors check the correlations on three levels, BUSCO80 set of genomes, BUSCO90 set and set of chromosomal level assemblies. The same procedure should also be used for the original version text, e.g. line 349-360. Alternatively all such discussions could go to supplementary material, and the differences between the three levels only mentioned summarily. Such a summarily mention is also missing for the respective pGLS discussions (albeit it is mentioned in the rebuttal).
+
+Minor points:
+
+line 44
+Reference 10 does not mention lobe-finned fishes.
+
+Line 286
+omega ratio should be explained at first occurrence in the main text, not just in figure legend of SI 30.
+
+line 311
+'as not true' seems to be a typo.
+
+Lines 319-337
+Here the correlation coefficients should be given also, as was done below, line 352ff. Also all these values should go into a table, not be scattered in the text.
+line 448-450 Rephrase to make clear that you are talking about T1R loss in carnivore marine mammals. Sirenians are (sort of) marine mammals and herbivores, but here it is claimed that marine mammals are all carnivores.
+
+Abstract, „examination of 2,210 vertebrate genomes“
+This number is different from the BUSCO80 number (1423 genomes). Please clarify.
+
+Line 453
+The grammar of this sentence is broken.
+
+Figure 4 panel a
+The borders of the rectangles are partially obscuring the text.
+
+Reviewer #3:
+Remarks to the Author:
+The authors have satisfactory addressed my concerns, including further validation of their species sampling (e.g., chromosome-scale genomes) as well as an in-depth selection analysis. I find the manuscript substantially improved and have no further criticism.
+REVIEWERS' COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+I acknowledge the great effort that the authors have put through the revision of this manuscript. Unfortunately, one of main criticism (that is, the lack of functional validation of bitter or sweet genes) from the first round of review wasn't really taken into account.
+
+We thank this Reviewer very much for acknowledging our effort. We absolutely agree with this Reviewer that having functional validations for bitter and/or sweet receptor genes would be great. At the same time, we believe that such a massive endeavor is way beyond the focus of the current study.
+
+Reviewer #2 (Remarks to the Author):
+
+The authors present a comprehensive study of chemosensory receptor gene numbers (OR, TAAR, V1R, V2R, T1R, T2R) in over thousand vertebrate genomes. The methods for identifying the receptor repertoires are solid. Several of the results stated in the abstract such as dynamic evolution, lineage-specific expansions and losses of chemosensory receptor gene families are not new and have in fact been reported in many previous publications. However, the clear advance of the present study is that now the conclusions are based on a much larger dataset. Furthermore, the enlarged data set allowed more fine-grained correlation with ecological parameters than previously possible. These correlations have now been included in a new Figure 4, which is a clear improvement above the original version of this manuscript. The correlation coefficients are often relatively small, but that may be expected in large scale species sampling. Overall the revisions have improved the manuscript considerably. It does contain a wealth of data points and as such constitutes an important addition to the knowledge about chemosensory receptors.
+
+The manuscript is generally not easy to read due to the listing of many numbers, genes and Latin species names in the main text. This has not changed from the original version. I would strongly suggest to put e.g. the association values (correlation coefficient, p values) in a dedicated table, this would increase readability of the corresponding segment considerably.
+
+We thank the reviewer for this suggestion. We added a table that summarizes the associations that we found between chemoreceptors and ecological factors.
+
+The newly introduced text about correlations between ecology and gene number is a welcome improvement. However it could be better integrated. For example, in new text line 319-337 the authors check the correlations on three levels, BUSCO80 set of genomes, BUSCO90 set and set of chromosomal level assemblies. The same procedure should also be used for the original version text, e.g. line 349-360. Alternatively all such discussions could go to supplementary material, and the differences between the three levels only mentioned summarily. Such a summarily mention is also missing for the respective pGLS discussions (albeit it is mentioned in the rebuttal).
+
+We thank the reviewer for this suggestion. In the revised version, we added the associations between the number of chemoreceptors and the olfactory organ morphology, taking only chromosome-scale assemblies, to the Supplementary Figure 43 (now Supplementary Figure 53) and added this sentence to the manuscript: “These associations with olfactory organ morphologies hold true when considering the BUSCO90 or chromosome-scale assemblies datasets (Supplementary Fig. 53).”. We also added a panel (c) to the Supplementary Figure 6 (now Supplementary Figure 7) showing the correlations between the number of genes in each chemoreceptor family, when only considering chromosome-scale assemblies, which was previously only shown in the responses to reviewers.
+Minor points:
+
+line 44
+Reference 10 does not mention lobe-finned fishes.
+
+Thank you for detecting this error. We changed the sentence according to the reference: “*V1R* and *V2R* genes are expressed in the sensory epithelium of the vomeronasal organ in tetrapods (except in amphibian, where *V1R* and a subset of *V2R* genes are expressed in the main olfactory epithelium), while in cartilaginous and ray-finned fishes these genes – often referred to as ORA and OlfC in these clades – are expressed in the main olfactory epithelium”
+
+Line 286
+omega ratio should be explained at first occurrence in the main text, not just in figure legend of SI 30.
+
+We added “(that is, dN/dS)” after the first occurrence of ω ratio at line 286.
+
+line 311
+'as not true' seems to be a typo.
+
+Changed to “is not true”.
+
+Lines 319-337
+Here the correlation coefficients should be given also, as was done below, line 352ff. Also all these values should go into a table, not be scattered in the text.
+
+Thank you for this suggestion. We added a table to the manuscript for the ecological associations discussed.
+
+line 448-450 Rephrase to make clear that you are talking about T1R loss in carnivore marine mammals. Sirenians are (sort of) marine mammals and herbivores, but here it is claimed that marine mammals are all carnivores.
+
+Modified.
+
+Abstract, „examination of 2,210 vertebrate genomes“
+This number is different from the BUSCO80 number (1423 genomes). Please clarify.
+
+2,210 is the number of genomes that was initially investigated, before any filter of the quality of these genomes. We thus replaced 2,210 with the number of genomes in the smallest dataset (BUSCO80, N=1,527). 1,423 correspond to the number of genomes in the BUSCO80 dataset, for which we retrieved phylogenetic information, and for which we could perform pGLS analysis presented in the study.
+
+Line 453
+The grammar of this sentence is broken.
+
+Corrected.
+
+Figure 4 panel a
+The borders of the rectangles are partially obscuring the text.
+
+We modified figure 4a for the species label to be visible. Furthermore, we modified the species names so that they are now in italic, and we replaced the underscore between the genus name and the species
+specific name by a space.
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have satisfactory addressed my concerns, including further validation of their species sampling (e.g., chromosome-scale genomes) as well as an in-depth selection analysis. I find the manuscript substantially improved and have no further criticism.
+
+Thank you very much for your suggestions that helped improving the manuscript.

0c9c35887efce41d19e5461230453d6939a7b149f11e6e2333d3d3e031155982/peer_review/peer_review.md

@@ -0,0 +1,218 @@
+Peer Review File
+
+Calibration-free NGS Quantitation of Mutations below 0.01% VAF
+
+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/.
+Reviewers' Comments:
+
+Reviewer #1:
+Remarks to the Author:
+• Please expand on strategy to address unique copy number alterations at each locus to be interrogated ("N") particularly in a panel encountering a diverse range of cancer specimens with unique CNVs
+• The incorporation of more validation samples, especially FFPE specimens of different quality (age, decal, etc) will help to highlight the potential of this technology
+• A section devoted to the more practical aspects of QBDA in FFPE specimens would be helpful. For example, the impact of resolving low allelic events from low input material (10ng).
+• Line 339 – Typo – “illumine” should be “Illumina”
+
+Stephen Yip
+
+Reviewer #2:
+Remarks to the Author:
+Dai and Wu et al. present a method incorporating UMI barcoding and blocker displacement amplification to more accurately determine variant allele frequency. This is applied to the detection of minimal residual disease in AML and other cancers. If adapted the approach would allow for more accurate and inexpensive detection of these low frequency variants. I have two major points that I would like to see addressed:
+
+1) The authors present a method for estimating the total UMI family count for a locus (Mt). This is essential to determine an accurate VAF. I do not believe that the authors properly perform this normalization. First, estimation of w_input * c_genome is critical to the accurate estimation of total amplifiable DNA - the authors account for this properly only in the pan-cancer panel where they can directly estimate these counts based on internal positive controls. The estimated count is likely to be very sensitive to the DNA prep which may be highly variable across individuals performing the experiment. Second, I am unclear what X as barcoding conversion yield is meant to represent. Based on how this is calculated however, it should represent any bias derived from the efficiency of polymerase blocking, and GC-content bias in the PCR/NGS, and influence of the variant type on these factors. This makes me a bit skeptical of the authors claim (for which they should provide data) that X is consistent across NGS runs. I would anticipate this normalization value to be site-specific as type of variant (homopolymer indel, SNV, multi-base insertion, etc) would likely affect the binding of the blocker and I would absolutely expect the GC content of the region to drive amplification biases at individual regions.
+
+2) The authors also present a nice control of healthy samples in Figures 2 and 4. However, I am a bit perplexed by the authors claim that in a healthy PBMC they find 10/22 known AML-associated variants (and 15/20 of the pan-cancer panel) and say that this is possible due to clonal hematopoiesis. While these mutations fall below the authors self-defined level of detection threshold, I think that this represents some sort of false-positive artifact that needs to be addressed further. How are the exact mutations that are in the cancer panels all appearing in this healthy individual?
+Reviewer #1 (Remarks to the Author):
+
+• Please expand on strategy to address unique copy number alterations at each locus to be interrogated ("N") particularly in a panel encountering a diverse range of cancer specimens with unique CNVs
+
+We thank the reviewer for the comment and suggestion. The DNA from a diverse range of cancer specimens can be split for QBDA and CNV quantitation. The copy number of gene of interest can be measured using any existing method including ddPCR, WES or WGS. Here we performed additional experiments to demonstrate a strategy of addressing CNV using QBDA panel without blocker. Without variant enrichment, the molecule count difference at each locus is due to copy number change. We showed the BRAF copy number in FFPE12 is identified accurately.
+
+Target 15 is an amplicon in BRAF so that BRAF ploidy can be calculated as the ratio between molecule count of target 15 and the mean molecule count of target 1-14. The ratio of BRAF/Ref (2.48) calculated from QBDA in FFPE 12 is consistent with ddPCR result (2.39). In a normal genomic DNA (NA18537) the ratio of BRAF/ref is close to 1.
+
+In discussion section, we added: “The copy number of gene of interest can be measured by ddPCR, whole genome or whole exome sequencing. We further demonstrate QBDA panel without blocker can be used for CNV calculation (Supplementary Fig. S9) and the ploidy of BRAF in FFPE 12 is highly consistent with ddPCR result.”
+
+In SI, we added Figure S9:
+
+![Bar graphs comparing molecule counts for FFPE12 and NA18537 across targets, with BRAF/ref ratios calculated](page_364_682_1012_312.png)
+
+BRAF/ref = Target 15/mean(Target 1-14) = 2.48      BRAF/ref = Target 15/mean(Target 1-14) = 0.97
+
+Figure S9. BRAF ploidy calculation using QBDA Melanoma panel without blocker. Target 15 is an amplicon in BRAF. The BRAF ploidy is calculated as the ratio between molecule count of target 15 and the mean molecule count of target 1-14. The ratio of BRAF/Ref (2.48) calculated from QBDA in FFPE 12 is consistent with ddPCR result (2.39). In a normal genomic DNA (NA18537) the ratio of BRAF/ref is close to 1.
+
+• The incorporation of more validation samples, especially FFPE specimens of different quality (age, decal, etc) will help to highlight the potential of this technology
+
+We thank the reviewer’s comments. To validate the utility and potential of QBDA technology, we tested a total of 22 FFPE samples from breast, colorectal, lung cancer patients or melanoma patients, and also tested a variety of other clinical samples including bone marrow DNA, cell-free DNA, PBMC DNA and fresh/frozen tissue DNA.
+We would like to pursue deeper on the potential of QBDA technology in collaboration with clinicians in the next papers, and really appreciate the suggestion of including more FFPE specimens of different quality (age and decalcification).
+
+• A section devoted to the more practical aspects of QBDA in FFPE specimens would be helpful. For example, the impact of resolving low allelic events from low input material (10ng).
+
+We thank the reviewer’s comments. QBDA allows detection of mutations down to 0.1% variant allele frequency from FFPE specimens with as low as 6-20 ng input, which can potentially help to understand resistance mechanism and clonal evolution to guide treatment. As an exemplary case study (Melanoma FFPE5), uncommon co-existence of BRAF V600E and low frequency NRAS Q61K mutations was observed by QBDA.
+
+Although BRAF and NRAS mutations are usually mutually exclusive in melanoma patients, BRAF/NRAS dual mutation may derive from two subclonal populations. As the patient was treated with BRAF inhibitor, co-existence of low frequency NRAS indicated potential clonal evolution and resistance mechanism related to NRAS. Consistent with our observation, there were recent reports in which BRAF and NRAS co-mutations were observed in the same cell after treated with a BRAF inhibitor (doi: 10.18632/oncotarget.12848.)
+
+We added more details and discussion on the FFPE5 case study and QBDA’s potential benefit of identifying drug resistance mutations to guide treatment decisions in section “Low-depth sequencing with pan-cancer panel and melanoma panel”, Discussion section in manuscript, and Supplementary Section 5 in SI.
+
+• Line 339 – Typo – “illumine” should be “Illumina”
+We thank the reviewer’s comments and have corrected the typo.
+
+Stephen Yip
+
+Reviewer #2 (Remarks to the Author):
+
+Dai and Wu et al. present a method incorporating UMI barcoding and blocker displacement amplification to more accurately determine variant allele frequency. This is applied to the detection of minimal residual disease in AML and other cancers. If adapted the approach would allow for more accurate and inexpensive detection of these low frequency variants. I have two major points that I would like to see addressed:
+
+1) The authors present a method for estimating the total UMI family count for a locus (Mt). This is essential to determine an accurate VAF. I do not believe that the authors properly perform this normalization. First, estimation of w_input * c_genome is critical to the accurate estimation of total amplifiable DNA - the authors account for this properly only in the pan-cancer panel where they can directly estimate these counts based on internal positive controls. The estimated count is likely to be very sensitive to the DNA prep which may be highly variable across individuals performing the experiment.
+Second, I am unclear what X as barcoding conversion yield is meant to represent. Based on how this is calculated however, it should represent any bias derived from the efficiency of polymerase blocking, and GC-content bias in the PCR/NGS, and influence of the variant type on these factors. This makes me a bit skeptical of the authors claim (for which they should provide data) that X is consistent across NGS runs. I would anticipate this normalization value to be site-specific as type of variant (homopolymer indel, SNV, multi-base insertion, etc) would likely affect the binding of the blocker and I would absolutely expect the GC content of the region to drive amplification biases at individual regions.
+We thank the reviewer for the comments and suggestions. We compared \( w_{input} \) (the amount of input DNA in ng) measure by Qubit or qPCR with the input amount (in ng) calculated based on internal positive control (IPC) amplicons in pan-cancer panel for all the 16 DNA samples reported. As summarized in Table 1 below, the inferred input amounts based on IPC in both tubes of pan-cancer panel are mostly consistent with Qubit or qPCR.
+
+Table 1. Pan-cancer panel DNA input quantitation by different methods
+
+<table>
+  <tr>
+    <th>Sample ID</th>
+    <th>Library input (ng)</th>
+    <th>Quantitation Method</th>
+    <th>IPC report (ng), Tube 1</th>
+    <th>IPC report (ng), Tube 2</th>
+  </tr>
+  <tr>
+    <td>FF4146</td>
+    <td>20.0</td>
+    <td rowspan="7">Qubit</td>
+    <td>24.7</td>
+    <td>27.8</td>
+  </tr>
+  <tr>
+    <td>FF4934</td>
+    <td>20.0</td>
+    <td>26.4</td>
+    <td>34.4</td>
+  </tr>
+  <tr>
+    <td>FF3176</td>
+    <td>20.0</td>
+    <td>27.6</td>
+    <td>31.0</td>
+  </tr>
+  <tr>
+    <td>FF4850</td>
+    <td>20.0</td>
+    <td>31.2</td>
+    <td>36.1</td>
+  </tr>
+  <tr>
+    <td>FF4927</td>
+    <td>20.0</td>
+    <td>32.6</td>
+    <td>38.7</td>
+  </tr>
+  <tr>
+    <td>FFPE25</td>
+    <td>10.5</td>
+    <td>25.4</td>
+    <td>28.1</td>
+  </tr>
+  <tr>
+    <td>FFPE26</td>
+    <td>11.2</td>
+    <td>25.8</td>
+    <td>27.2</td>
+  </tr>
+  <tr>
+    <td>FFPE23</td>
+    <td>20.0</td>
+    <td rowspan="4">qPCR</td>
+    <td>28.5</td>
+    <td>32.0</td>
+  </tr>
+  <tr>
+    <td>FFPE20</td>
+    <td>20.0</td>
+    <td>28.0</td>
+    <td>31.3</td>
+  </tr>
+  <tr>
+    <td>FFPE6</td>
+    <td>20.0</td>
+    <td>26.1</td>
+    <td>27.9</td>
+  </tr>
+  <tr>
+    <td>FFPE5</td>
+    <td>20.0</td>
+    <td>29.8</td>
+    <td>34.3</td>
+  </tr>
+  <tr>
+    <td>DLS4 (cfDNA)</td>
+    <td>10.0</td>
+    <td rowspan="4">Qubit</td>
+    <td>9.4</td>
+    <td>10.7</td>
+  </tr>
+  <tr>
+    <td>cfDNA sample A</td>
+    <td>10.0</td>
+    <td>8.2</td>
+    <td>10.3</td>
+  </tr>
+  <tr>
+    <td>cfDNA sample C</td>
+    <td>10.0</td>
+    <td>9.6</td>
+    <td>11.9</td>
+  </tr>
+  <tr>
+    <td>cfDNA sample D</td>
+    <td>10.0</td>
+    <td>8.1</td>
+    <td>7.8</td>
+  </tr>
+  <tr>
+    <td>cfDNA sample E</td>
+    <td>10.0</td>
+    <td></td>
+    <td>13.1</td>
+    <td>13.6</td>
+  </tr>
+</table>
+
+Conversion yield \( \chi \) reflects the barcoding yield, i.e. the percentage of original molecules that can be attached with a UMI. The PCR efficiency in the first two cycles of PCR (UMI incorporation step) determines the value of \( \chi \). Bias after the UMI incorporation step will only impact reads distribution in the final library and thus influence the UMI family size (the number of reads bearing the same UMI); after UMI-based data processing, bias introduced in the later PCR steps will be corrected, and \( \chi \) values will not be impacted.
+Polymerase blocking by blocker does not impact conversion yield \( \chi \) as well because it is after UMI attachment step. Blocker will influence the PCR yield per cycle for WT and variant, and for different types of variant molecules during BDA step and thus influence the final reads distribution in the sequencing library. We added Figure S10 in the SI to demonstrate this.
+Figure S10. Influence on conversion yield and amplification bias of each step in QBDA.
+
+Intra-operator (Figure S11) and inter-operator (Figure S12) experiments were added in SI to demonstrate the consistency of conversion yield \( \chi \) between runs. High reproducibility was observed in triplicates of the same operator. Good inter-operator reproducibility is observed as well without optimization of operation difference, especially considering one operator has never performed QBDA previously. We think the inter-operator difference may derive from small difference in setting up the first PCR reaction from person to person and may be further reduced.
+
+Figure S11. Intra-operator conversion yield reproducibility of QBDA Melanoma panel in triplicate experiments.
+Figure S12. QBDA Melanoma panel conversion yield is generally consistent between two different operators.
+
+Conversion yield \( \chi \) is a parameter for amplicon so that it is amplicon-specific; it should not be sensitive to the type of variants that influences blocker binding. Accurate quantitation of different types of mutations in the same amplicon (and same enrichment region) is observed in both TB panel and SNP panel, where different SNPs and indels up to 3-bp are tested. Although it is possible that variant especially long indel might impact \( \chi \) due to significant change in sequence length and properties, the observation in QBDA quantitation accuracy indicates this would not introduce a quantitation error over 2-fold (Fig. 1c).
+
+2) The authors also present a nice control of healthy samples in Figures 2 and 4. However, I am a bit perplexed by the authors claim that in a healthy PBMC they find 10/22 known AML-associated variants (and 15/20 of the pan-cancer panel) and say that this is possible due to clonal hematopoiesis. While these mutations fall below the authors self-defined level of detection threshold, I think that this represents some sort of false-positive artifact that needs to be addressed further. How are the exact mutations that are in the cancer panels all appearing in this healthy individual?
+
+We thank the reviewer for the comments. We think the observed mutation with VAF lower than LoD in healthy PBMC DNA in the AML study are most likely due to clonal hematopoiesis or DNA damage for two reasons:
+
+(1) Although 10 mutations were reported out of the 22 targets, all the 10 none-zero mutations (Fig. 2b) were all C>T or G>A substitutions, which are the dominant mutation types in clonal hematopoiesis. In a recent study (Clonal haematopoiesis harbouring AML-associated mutations is ubiquitous in healthy adults, doi: 10.1038/ncomms12484) Young et al. observed clonal hematopoiesis, frequently harboring mutations in DNMT3A and TET2, in 95% of individuals studied using a sequencing method with 0.03% VAF LoD. C>T or G>A substitutions are the most dominant observed mutation in clonal haematopoiesis in this study. Because LoD for QBDA Leukemia panel is further improved by 10- to 100-fold comparing to previous study, it is likely the known AML-associated mutations in “healthy” PBMC DNA are existing in the sample as opposed to random error.
+
+(2) We prepared 5 replicated libraries from one healthy PBMC gDNA sample to analyze specificity of QBDA AML panel (Section “QBDA AML panel for MRD detection” in manuscript). If a mutation is observed in ≥4 out of the 5 libraries, we believe this is a true positive mutation existing in the DNA sample, not an artifact caused by polymerase misincorporation or sequencing error, because the probability of the same error appearing 4 times out of 5 experiments is extremely low. After filtering out the true positives, we observed only 1 false positive mutation call out of the 5 libraries. The specificity study with technical replicate experiments further reduced the likelihood of random error.
+The mutation might be originated from contamination, especially when synthetic templates were added as spike-in positive controls as the case in pan-cancer panel. Because the low frequency mutation in healthy sample due to clonal hematopoiesis, DNA damage, or contamination are all below LoD, the impact on MRD detection should be not significant.
+Reviewers' Comments:
+
+Reviewer #1:
+Remarks to the Author:
+I am satisfied with the edits and responses to my comments from my review. They did not address all the comments (particularly about testing FFPE samples of different quality/age for this paper but did mention they will pursue this in a future study). Stephen Yip
+
+Reviewer #2:
+Remarks to the Author:
+I am happy with the revisions and appreciate the detail that the authors took in their reponse.
+Reviewer #1 (Remarks to the Author):
+
+I am satisfied with the edits and responses to my comments from my review. They did not address all the comments (particularly about testing FFPE samples of different quality/age for this paper but did mention they will pursue this in a future study). Stephen Yip
+
+We thank the reviewer for all the comments.
+
+Reviewer #2 (Remarks to the Author):
+
+I am happy with the revisions and appreciate the detail that the authors took in their reponse.
+
+We thank the reviewer for all the comments.

0c9c35887efce41d19e5461230453d6939a7b149f11e6e2333d3d3e031155982/preprint/preprint.md

@@ -0,0 +1,291 @@
+Calibration-free NGS Quantitation of Mutations below 0.01% VAF
+
+Peng Dai
+Rice University    https://orcid.org/0000-0002-4581-3473
+Lucia Wu
+Rice University
+Sherry Chen
+Rice University
+Michael Wang
+Rice University    https://orcid.org/0000-0001-7009-6958
+Lauren Chen
+Rice University
+Jinny Zhang
+NuProbe USA Inc.
+Christina Hao
+NuProbe USA Inc.
+Weijie Yao
+NuProbe USA Inc.
+Jabra Zarka
+The University of Texas MD Anderson Cancer Center    https://orcid.org/0000-0001-8668-6741
+Ghayas Issa
+The University of Texas MD Anderson Cancer Center    https://orcid.org/0000-0002-4339-8683
+Lawrence Kwong
+UT MD Anderson Cancer Center
+David Zhang (dyz1@rice.edu)
+Rice University
+
+Article
+
+Keywords: somatic mutations, genetic sequencing, DNA
+
+Posted Date: June 10th, 2021
+
+DOI: https://doi.org/10.21203/rs.3.rs-579121/v1
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Version of Record: A version of this preprint was published at Nature Communications on October 21st, 2021. See the published version at https://doi.org/10.1038/s41467-021-26308-6.
+Calibration-free NGS Quantitation of Mutations below 0.01% VAF
+
+Peng Dai1,6,7, Lucia Ruojia Wu1,7, Sherry Xi Chen1,6, Michael Xiangjiang Wang1, Lauren Yuxuan Cheng1, Jinny Xuemeng Zhang3, Christina Pengying Hao3, Weijie Yao3, Jabra Zarka4, Ghayas C. Issa4, Lawrence Kwong5, David Yu Zhang1,2*
+
+1 Department of Bioengineering, Rice University, Houston, TX, USA.
+2 Systems, Synthetic, and Physical Biology, Rice University, Houston, TX, USA.
+3 NuProbe USA, Houston, TX, USA.
+4Department of Leukemia, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
+5Department of Translational Molecular Pathology, The University of Texas MD Anderson Cancer Center, Houston, TX, USA.
+6Present affiliation: NuProbe USA, Houston, TX, USA.
+7These authors contributed equally: Peng Dai, Lucia Ruojia Wu
+
+*e-mail: dyz1@rice.edu
+Abstract
+
+The quantitation of rare somatic mutations is essential for basic research and translational clinical applications including minimal residual disease (MRD) detection. Though unique molecular identifier (UMI) has suppressed sequencing error and allowed detection rare mutation, the sequencing depth requirement is high. The blocker displacement amplification (BDA) allele enrichment method allows detection of rare mutations using low sequencing depth, but requires calibration to accurately quantitate the VAF of novel mutations. Here, we present Quantitative Blocker Displacement Amplification (QBDA), a method that allows accurate detection and quantitation of mutations below 0.01% VAF at only 23,000X depth. QBDA integrates sequence-selective variant enrichment into UMI quantitation allowing confident detection of rare mutations and reduced sequencing depth. Using a panel of 20 genes recurrently altered in acute myeloid leukemia, we demonstrate quantitation of various mutations including single base substitutions and indels down to a VAF of 0.001% at a single locus with less than 4 million sequencing reads, allowing a sensitive minimal residual disease (MRD) detection in patients during complete remission. In a comprehensive pan-cancer panel covering 61 genes and a melanoma hotspot panel covering 8 genes, we detect mutations down to 0.1% VAF using only 1 million reads in a broad range of clinical samples including cell-free DNA and FFPE DNA, enabling tissue or liquid biopsy genetic tests with de-centralized sequencing instruments. QBDA thus provides a convenient and versatile method for sensitive mutation quantitation using low-depth sequencing.
+
+Introduction
+
+DNA variants with low allelic frequencies have important clinical and biological implications, as they often lead to resistance or recurrence in infection\(^{1,2}\) and cancer treatments\(^{3-}\)
+5. Sensitive genetic testing is highly desired in both minimal residual disease (MRD)\(^{6-8}\) detection and liquid biopsy\(^{9,10}\). Detection of MRD in acute myeloid leukemia (AML) has prognostic and therapeutic implications aimed at preventing morphologic relapse\(^8\). Sensitive detection of leukemia-specific mutation markers could improve prognostication by identifying submicroscopic disease during remission\(^6\). Compared to MRD detection by multicolor flow cytometry (MFC)\(^{11,12}\), NGS MRD assays have the potential for detection of “actionable” mutations to guide therapy selection. The cell-free DNA (cfDNA) in circulation plasma provides a ‘snapshot’ of dying cells around the body and thus is widely used in liquid biopsy for non-invasive genetic testing. It is frequently the most accessible clinical sample for applications such as therapy selection, post-treatment monitoring, and early cancer screening. Because the tumor-derived DNA is mixed with large amount of normal DNA\(^{13,14}\), VAF for cancer-related mutations is often low requiring high assay sensitivity.
+
+Polymerase error during amplification\(^{15,16}\) and sequencing error of NGS platforms\(^{17,18}\) made it difficult to robustly quantitate low frequency mutations < 1% VAF using conventional NGS technologies. Unique molecular identifiers (UMIs) have been developed to suppress the errors to detect mutations below 0.1% VAF\(^{19,20}\). Recent advances in DuplexSeq\(^{21}\), NanoSeq\(^{22}\) and SaferSeqS\(^{23}\) has further reduced errors by grouping both strands of a DNA molecule together into a duplex family to distinguish DNA damage with real mutation achieving confident variant calling at 0.01% VAF or lower. However, since all template molecules, regardless of wild type or variant molecules, are sequenced redundantly in current UMI-based methods, they require sequencing to extremely high depths proportional to input molecule amount. On the other hand, high input DNA amount is needed for successful sampling of rare variants. For a mutation with 0.005% VAF, a total of 75,000 diploid human genomic DNA (gDNA) is required to achieve an
+average of 3.75 mutant copies. This corresponds to approximately 500 ng gDNA. The combination of UMI and high input amount leads to sequencing depth unaffordable for many researchers, clinicians and patients. Blocker displacement amplification (BDA)\(^{24,25}\) enriches variant alleles by introducing rationally designed blocker oligonucleotides that competes with forward primer to suppress the amplification of wild type molecules. BDA allows detection of rare mutations using low sequencing depth, but loses VAF quantitation without calibration.
+
+To overcome these challenges, herein we have developed QBDA, a method that allows calibration-free accurate VAF quantitation with low-depth sequencing by integrating molecular barcoding with BDA technology for variant enrichment. Because the amplification of wild type (WT) molecules is suppressed, the number of WT UMI families does not represent actual number of WT molecules. Thus, VAF is calculated based on variant molecule count from QBDA and the input molecule count (i.e. number of input genome copies), which can be calculated from input DNA amount or by adding internal positive control amplicons that quantify a small portion of the input molecules at several different loci in house-keeping genes.
+
+Herein, we demonstrate that mutations within targeted regions are simultaneously enriched and accurately quantified, including single base substitutions and indels. We apply the QBDA technology to a 20-gene AML panel and demonstrate a robust quantitation of single base substitutions and indels down to 0.001% VAF at a single locus for MRD analysis. Finally, two QBDA cancer panels including a comprehensive pan-cancer panel and a specific melanoma panel are demonstrated on tumor tissue samples and cfDNA samples.
+
+Results
+
+Development of QBDA. A PCR-based UMI addition approach is performed to attach UMI to each individual DNA single strand in the original DNA templates, followed by BDA to enrich
+variant amplicons (Fig. 1a). In BDA, a rationally designed blocker DNA oligonucleotide that partially overlaps with the 3’ of the forward primer is introduced to suppress the amplification of wild type molecules. The nucleotide sequence unique to the blocker and not in the forward primer is the enrichment region; any nucleotide change in this region will prevent the hybridization of blocker to the template, thus allows extension of forward primer.
+
+VAF calculation in QBDA does not require counting wild type molecules. In standard UMI-based, non-allele-enrichment NGS methods, the VAF of a mutation call can be calculated as:
+
+\[
+VAF = M_v / M_t
+\]
+
+where \( M_v \) is the UMI family count of the mutation, and \( M_t \) is the total number of UMI family count for this locus.
+
+In QBDA, because the amplification of WT is suppressed, the number of WT reads is small and thus UMI count of WT does not represent actual number of WT molecules. Therefore, we calculate \( M_t \) as the following:
+
+\[
+M_t = 2 * w_{input} * c_{genome} * \chi * N
+\]
+
+Here \( w_{input} \) is the amount of input DNA in ng, \( c_{genome} \) is the number of haploid genomes per 1 ng DNA (for human gDNA, \( c_{genome} = 300\ ng^{-1} \)), \( \chi \) is the UMI barcoding conversion yield, and \( N \) is the copy number of this loci relative to the genome (N = 1 for normal loci, >1 for copy number amplification, < 1 for copy number loss). We assume N =1 if no CNV data is available. Because two different UMIs are attached to the two strands of one original DNA molecule in QBDA, the number is multiplied by 2.
+
+Based on our observations, the UMI barcoding conversion yield \( \chi \) for each amplicon remains consistent across different NGS runs. \( \chi \) was characterized using a library prepared with QBDA
+protocol but without the blockers (i.e. no enrichment). From this library, \( \chi \) for each amplicon was calculated as \( \chi = M_t / (2 * w_{input} * c_{genome}) \).
+
+The pan-cancer panel further incorporates internal positive control amplicons without blocker into the panel, which quantitates the molecule at several loci in house-keeping genes to estimate the DNA input amount. In pan-cancer panel, \( M_t \) is calculated from the UMI counts of internal positive control amplicons.
+
+**QBDA demonstration.** We first demonstrated the variant enrichment, error correction, and quantitation of QBDA using a single-plex QBDA (Supplementary Table S1 and Supplementary Section 1). Here 9 different mutations including single-base substitution, insertion and deletion within an 18 nt region (Supplementary Fig. S1) were enriched using the same BDA primer-blocker set; these mutations are from rpoB (Rv0667) gene of *M. tuberculosis*, and are relevant to tuberculosis (TB) drug resistance. We mixed H37Rv (WT) DNA with 9 synthetic DNA templates each bearing a different mutation to prepare a sample containing approximately 1% VAF for each of the 9 mutations.
+
+QBDA simultaneously enriches mutations and corrected errors. Using standard, PCR-based NGS, the majority of reads (87.6%) were WT, which do not contribute to variant sequencing depth. Using QBDA, sequencing reads became more focused on the mutations, and the WT reads were suppressed to only 2.4% (Fig. 1b). In BDA-based enrichment, the amplification efficiency is not the same for different mutations. Instead of performing calibration curve to obtain the variant enrichment efficiency for all the possible mutations, here we used UMI to improve mutation quantitation accuracy and suppress error (Fig. 1c). In standard NGS, 11.7% of the variant reads did not match the 9 expected spike-in mutations, thus were false positive variants. In QBDA after UMI-based error correction, all the false variants were
+removed (Fig. 1c, see Methods Section for bioinformatics and molecule count calculation). We calculated the counts of unique UMI families for each variant in QBDA, and compared them with expected variant molecule counts. Here the expected variant molecule counts were obtained from a UMI-based NGS library without BDA enrichment. All the observed molecule counts were within 2-fold of the expected values.
+
+Multiplexed QBDA quantitation. We validated QBDA quantitation capability on a 0.1% and 1% VAF sample prepared by mixing repository human cell line DNA sample NA18562 with NA18537. A 10-plex QBDA panel covering 10 SNP loci with different genotypes in the two cell line DNA samples was built (Supplementary Fig. S2 and Supplementary Table S2). The calculated VAFs for all the loci were within 2-fold of expected true value in 1% sample, and 7 out of 10 were within 2-fold in 0.1% sample, with the other 3 were still within 3-fold (Supplementary Fig. S2c). Stochasticity in sampling a small number of molecules contributed to quantitation error in 0.1% sample as only 30 ng gDNA is used, corresponding to only 9 haploid of variant at 0.1% VAF. Furthermore, variant enrichment does not lead to higher error rate comparing to no enrichment (Supplementary Fig. S2e).
+
+QBDA AML panel for MRD detection. To demonstrate quantitation of <0.01% VAF rare mutation for MRD analysis, we next built a 22-plex QBDA panel covering AML-related mutation hotspot regions in 20 different genes for MRD detection (Supplementary Table S3-S4). De novo mutation calling was performed for all 382 nucleotide positions in 22 enrichment regions; mutations with \( \geq 6 \) unique UMI families (corresponding to \( \geq 3 \) original DNA molecules in QBDA) and having VAFs above or equal to the LoD threshold were reported. The LoD threshold is below 0.01% VAF, but varies for different types of mutations (Fig. 2a, Supplementary Section 3.2 for LoD).
+Validation of the AML panel was performed using a positive sample containing 22 mutations, which was prepared by mixing PBMC DNA from a healthy donor, Horizon Myeloid DNA Reference Standard, and 3 synthetic DNA templates (Supplementary Section 3.1). The expected VAF was between 0.001% and 0.1%; 16 out of 22 mutations were around 0.01% (between 0.005% and 0.02%). There were 19 single-base substitutions, 2 insertions, and 1 deletion in this positive sample. Using 1 \( \mu \)g of DNA input, all 22 mutation were observed; 82% of the mutations were within 2-fold of expected VAF, and 100% were within 1 order of magnitude. Here the expected VAF was quantitated by UMI-based NGS without enrichment. The quantitation is less accurate for some lower VAF mutations, which is likely a result of stochasticity in sampling a small number of DNA molecules (Fig. 2b). The healthy PBMC DNA used in the positive sample was also assayed using the AML panel as a negative control. Using the same input amount (1 \( \mu \)g), none of the 22 mutations was above the LoD threshold in Fig. 2a. In this experiment, the none-zero mutations were all C>T or G>A substitutions, which are possibly results of clonal hematopoiesis\(^{26,27}\) (Fig. 2b).
+
+Technical sensitivity was analyzed by testing the abovementioned positive sample in triplicates (1 \( \mu \)g DNA input each). There was only 1 false negative out of the 3 libraries, corresponding to 1-1/(22*3) = 98.5% technical sensitivity. If we only consider the 16 mutations between 0.005% and 0.02% VAF, the technical sensitivity was 1-1/(16*3) = 97.9% (Fig. 2c).
+
+The specificity of AML panel was assessed using a “negative sample”. Because QBDA is highly sensitive to mutations below 0.01% VAF, and even healthy blood donors have low-level mutations in their PBMC DNA as a result of DNA damage or clonal hematopoiesis, such as C>T or G>A substitutions\(^{26,27}\), there is no perfect “negative sample” for MRD detection (Supplementary Fig. S3). We prepared 5 replicated libraries from the same healthy PBMC
+gDNA sample to analyze specificity of QBDA AML panel; each library had 1 \( \mu \)g of gDNA input. If a mutation is observed in \( \geq 4 \) out of the 5 libraries, we believe this is a true positive mutation existing in the DNA sample, not an artifact caused by polymerase misincorporation or sequencing error, because the probability of the same error appearing 4 times out of 5 experiments is extremely low. After filtering out the true positives, we observed only 1 false positive mutation call out of the 5 libraries. Therefore, the technical specificity of AML panel can be calculated as 1-1/(382*5) = 99.95% at the current LoD threshold, where 382 is the number of enriched nucleotide positions in the panel.
+
+We next prepared samples with 3-fold or 5-fold of the VAF in the abovementioned positive sample. For each of the 22 mutations, higher VAF input always generates higher observed VAF; therefore, we can confidently differentiate samples with 0.02% VAF difference (\( p = 3 \times 10^{-6} \) by paired Wilcoxon signed rank test, Fig. 2c). Sequencing depth down to 45,000X does not affect sensitivity in the 1X VAF (\( \approx 0.01\% \)) sample using *in silico* random down-sampling analysis (Fig. 2d). 23,000X depth is still acceptable for detection of 0.01% VAF, but we recommend 45,000X depth for more accurate quantitation (Supplementary Fig. S4).
+
+**Detection of ultra-low VAF mutations during AML complete remission.** QBDA AML panel was applied to clinical samples, and was compared with other MRD detection methods including MFC\(^{12}\) and conventional NGS\(^{28}\). 10 paired bone marrow aspirates from 5 AML patients sampled at diagnosis and during complete remission were tested by QBDA panel. All patients chosen were *NPM1* mutated at diagnosis given that mutations in *NPM1* are considered founder mutations in the pathogenesis of AML\(^{29}\) and *NPM1* is a validated MRD marker\(^{6}\).
+
+Mutation VAF and the percentage of blasts in bone marrow at diagnosis and during remission for each of the five patients were plotted (Fig. 3a-3e). The allele frequencies for
+mutations detected in five patients during remission were summarized (Fig. 3f). Full list of mutations and patient information were summarized in Supplementary Table S5-S6. Persistent mutations were detected in 3 out of the 5 patients. Preleukemic mutations in the epigenetic regulators \( DTA \) (i.e., \( DNMT3A \), \( TET2 \), and \( ASXL1 \)) were most common and were observed in all 3 patients with mutations detected during remission. This is consistent with previous observations that they are often present in persons with age-related clonal hematopoiesis, and are not significantly associated with increased relapse risk\(^{7,30-35}\). Other mutations observed during remission include \( NPM1 \), \( KIT \), \( NRAS \) and \( TP53 \).
+
+A swimmer plot of clinical course and molecular findings of each patient is summarized (Fig. 3g). QBDA identified NPM1 mutation in only one patient (patient #1) during remission at a VAF of 0.0052%. In spite of the low allele frequency detected, the duration of remission is only 7.0 months for this patient. However, flow cytometry reported MRD negative and conventional NGS failed to detect \( NPM1 \) mutation at the same time point for this patient. This NPM1 mutation was confirmed by conventional NGS at relapse, indicating QBDA’s accuracy of rare mutation detection and potential of early detection.
+
+QBDA reported no \( NPM1 \) mutation during remission in the other four patients which is in concordance with conventional NGS. Three of them were MRD negative by flow cytometry, with over 100 months of remission (patient #3~5). In one case, however, MRD positive is reported by flow cytometry and the duration of remission is 8.1 months (patient #2). \( NPM1 \) mutation was not observed in the two subsequent time points even after relapse using conventional NGS. Instead, *de novo* mutations in \( KDM6A \) and \( PHF6 \) were identified. We thus believe that QBDA is accurate in reporting no \( NPM1 \) mutation during remission but clonal evolution occurred as alternative cause of relapse\(^{29,36}\). QBDA assay allows sensitive detection of
+rare mutations in genes of interest, which we envision to be significant for relapse risk assessment.
+
+QBDA pan-cancer panel for MRD detection. Next, we demonstrated highly-multiplexed QBDA to simultaneously detect variants in 180 amplicons per tube. VarMap™ Pan-Cancer NGS Panel from NuProbe Inc. was developed based on QBDA technology, which covers 61 genes and 360 hotspot regions in two tubes (Supplementary Fig. S5). It is compatible with MRD detection at 0.01% VAF using 1 μg DNA input and 25 M reads per tube. Validation was performed similarly as AML panel using a positive sample containing 20 mutations, which was prepared by mixing PBMC DNA from a healthy donor and 20 synthetic DNA templates (Supplementary Table S7). All 20 mutations were observed; 60% of the mutations were within 2-fold of expected VAF, and 100% were within 1 order of magnitude (Fig. 4a). One of the spike-in mutation is observed in the healthy DNA in all five technical replicates at about 0.016% VAF, which we consider a true positive mutation existing in the healthy DNA sample. This background is subtracted from the reported VAF in positive samples. Technical sensitivity was analyzed by testing the abovementioned positive sample in duplicates. Setting LoD threshold at 0.006% VAF, there were two false negative out of the 2 libraries, corresponding to 1-2/(20*2) = 95% technical sensitivity. The healthy PBMC DNA used in the positive sample was also assayed as a negative control. Calculated similarly as AML panel, the specificity of pan-cancer panel is 99.997% with only 1 false positive mutation with > 0.006% VAF detected in five replicate libraries. We next prepared samples with 3-fold of the VAF in the abovementioned positive sample. For each of the 20 mutations, higher VAF input always generate higher observed VAF (Fig. 4b).
+Low-depth sequencing with pan-cancer panel and melanoma panel. In liquid biopsy samples, the available DNA amount is in ng range and thus too low for detecting 0.01% VAF mutations. In tumor tissue samples, the background mutation derived from Formalin-Fixed Paraffin-Embedded (FFPE) DNA damage is often higher than 0.1% VAF. Therefore, for tissue DNA or liquid biopsy analysis, an assay with an LoD of 0.1% VAF using low input DNA and low sequencing depth is more desired than an extremely sensitive assay requiring high input and high sequencing depth.
+
+Here we used QBDA panels for quantitating mutations above 0.1% VAF with 10 ng DNA input and 0.5 M reads per tube. We applied VarMap™ Pan-Cancer Panel to 16 samples, including 6 FFPE DNA samples from breast, colorectal or lung cancer patients, 5 fresh frozen (FF) DNA from hepatocellular carcinoma patients, 1 plasma cfDNA from breast cancer patients and 4 cfDNA from healthy people (Fig. 4c and Supplementary Table S8). On average, 1.9 somatic mutations at non-SNP loci were detected per sample. Because QBDA allowes low-depth detection of low frequency mutations, all the 16 samples can be sequenced in one Miniseq or MiSeq run, enabling tissue or liquid biopsy pan-cancer genetic tests with de-centralized sequencing instruments.
+
+A QBDA melanoma panel (Supplementary Section 5) was applied to 16 FFPE and 7 FF clinical tissue samples (Fig. 4d, Supplementary Table S11), and we found co-existence of \( BRAF \) V600E and low frequency \( NRAS \) Q61K mutations in one FFPE tissue. As the patient was treated with \( BRAF \) inhibitor, co-existence of low frequency \( NRAS \) indicated potential clonal evolution and resistance mechanism related to \( NRAS^{37} \).
+
+Discussion
+Considering the molecular heterogeneity of AML, MRD analysis based on mutation biomarkers in bone marrow DNA could provide actionability to guide treatment decision as a complementary method for MFC and morphology-based assessment of remission. The sensitivity and cost for NGS MRD analysis is dependent on the assay’s analytical LoD and sequencing depth respectively. QBDA combines variant enrichment with molecular barcoding in NGS to allow detection of mutation down to 0.001% VAF with about 23,000X sequencing depth. When applied to clinical samples, QBDA identified residual *NPM1* mutation at 0.005% VAF in one patient during remission while both flow cytometry and conventional NGS failed to detect MRD at the same time point for this patient. The accuracy of QBDA mutation call was supported by clinical outcome of short duration of remission as well as confirmation of such mutation at relapse by conventional NGS, indicating QBDA’s potential of early detection.
+
+QBDA quantitation is accurate. We extensively validated quantitation accuracy by comparing QBDA VAF with spike-in ratio of cell line DNA or synthetic template, with expected allele frequencies in commercial myeloid DNA Reference Standard, with VAF from digital droplet PCR (ddPCR), and with conventional NGS. QBDA reduced both false positive and false negative variant calls comparing to conventional NGS in the 23 melanoma clinical samples (Fig. 4e). Validation against ddPCR was performed in clinical DNA samples with BRAF/NRAS mutations (Supplementary Fig. S7, Supplementary Table S12). To validate no false negative call were made, one healthy donor PBMC gDNA sample and three FFPE samples without *BRAF*/NRAS* mutation by QBDA were also tested by ddPCR and were confirmed with no mutation (Supplementary Fig. S8, Supplementary Table S12). Remaining errors in quantitation may be due to Poisson distribution in sampling or DNA damage. We introduced different UMI sequences to each strand of DNA molecule by PCR and thus duplex family information is lost
+during denaturation of UMI attachment PCR. We expect that error from DNA damage may be further suppressed in QBDA by using ligation-based UMI attachment, so that in downstream bioinformatics analysis both strands of a DNA molecule can be grouped into a duplex family similar to DuplexSeq\(^{21}\), NanoSeq\(^{22}\) and SaferSeqS\(^{23}\) while still reducing sequencing depth by BDA variant enrichment.
+
+The gene ploidy impacts VAF in QBDA, but QBDA is able to accurately quantitate VAF in case CNV and mutation are simultaneously present in the gene of interest as long as copy number for the gene is normalized. As demonstrated in the formula of calculating total number of UMI family count for each locus (\(M_t\)), \(M_t\) needs to be adjusted by the copy number in genome if CNV occurs. As an example of copy number normalization, *BRAF* gene in melanoma FFPE12 sample underwent both copy number variation (CNV) and mutation; VAF for *BRAF* V600K mutation was consistent with ddPCR after normalizing the copy number of *BRAF* gene (Supplementary Fig. S7).
+
+Broad coverage, mutation sensitivity, and low sequencing cost are simultaneously explored by the 61-gene pan-cancer QBDA panel that detects mutations down to 0.1% VAF requiring only 1 M reads per sample, or detects MRD at 0.01% VAF using 1 \( \mu \)g DNA input and 50 M reads per sample. We envision MRD based on large Pan-Cancer panel can pick up *de novo* drug resistance mutations to guide treatment decisions based on its high coverage.
+Methods
+
+QBDA protocol. QBDA Library preparation consisted of three PCR reactions (Fig. 1a): UMI addition and pre-amplification, BDA for variant enrichment, and index PCR, all performed on a T100 Thermal Cycler (Bio-Rad). Firstly, DNA sample was mixed with the specific forward primer (SfP), Specific reverse primer (SrP) and amplified using high fidelity Phusion polymerase. The final concentration for each SfP and SrP was 15 nM unless otherwise noted. 2 cycles of long-extension PCR were performed for the addition of UMI on all target loci, followed by a universal amplification. In order to amplify the molecules to avoid sample loss during purification while preventing addition of multiple UMIs onto the same original molecule, the annealing temperature was raised with short annealing time (30 s) with Universal forward primer (UfP) and Universal reverse primer (UrP). Addition of UfP and UrP into the reaction was an open-tube step on the thermocycler to prevent temperature drop and primer dimer formation. Thermal cycling condition was: 98°C:30s - (98°C:10s - 63°C:30min - 72°C:60s)x2 - (98°C:10s - 63°C:20s - 72°C:60s)x2 - (98°C:10s - 71°C:20s -72°C:60s)x5 - (72°C:5min) - 4C:hold. During the last 5 min of the second 30 min at 63°C, 1.5 μM of each universal primer was added while keeping the reactions inside the thermal cycler. If the DNA input is less than 500 ng, the reaction mixture was purified using AMPure XP beads (1.6X ratio) twice to remove single-stranded primers. If the DNA input is over 500 ng, double-side size selection (0.3X, 1.6X ratio) was performed to remove long input gDNA, followed by another 1.6X AMPure XP beads purification.
+
+Secondly, BDA amplification was performed. BDA forward primer, BDA blocker, Phusion polymerase, dNTPs, and PCR buffer were mixed with the purified PCR product for BDA amplification. Thermal cycling condition was: 98 °C:30 s - (98 °C:10s - 63°C:5min -
+72°C:60s)x23 - 4°C:hold. The reaction mixture was purified using AMPure XP beads (1.8X ratio).
+
+Next, Adapter is added. BDA adaptor primer (Adp_fP, comprising illumine adapter sequence and BDA forward primer sequence) and UrP are mixed with the purified PCR reaction mixture and amplified. Thermal cycling condition was: 98 °C:30s - (98°C:10s - 63°C:5min - 72°C:1min)x2 - 4C:hold. The reaction mixture was purified using AMPure XP beads (1.6X ratio). Lastly, standard NGS index PCR is performed. Libraries are normalized and loaded onto an Illumina sequencer.
+
+Samples. Fresh frozen tissue samples were purchased from OriGene Technologies, Inc. in de-identified format. Sixteen formalin fixed paraffin embedded (FFPE) samples of patients with metastatic stage IV melanoma and ten bone marrow aspirates samples of patients with acute myeloid leukemia in de-identified format were collected from MD Anderson Cancer Center. All procedures performed in studies involving human participants were approved by Institutional Review Board at MD Anderson), and were in accordance with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. FFPE samples from breast, colorectal and lung cancer patients were purchased from OriGene Technologies, Inc. in de-identified format. Plasma from healthy people were purchased from Zen-Bio Inc. Plasma from breast cancer patients were purchased from Discovery Life Science.
+
+NA18537 and NA18562 DNA were purchased from Coriell Institute for Medical Research. Myeloid DNA Reference Standard was purchased from Horizon Discovery. DNA input was quantified by qubit for genomic DNA, by qPCR for fragmented DNA (FFPE DNA and cfDNA) to identify the amplifiable portion. Oligonucleotides and synthetic DNA templates (gBlock) were ordered from Integrated DNA Technologies.
+NGS data preprocessing. The QBDA libraries were analyzed using 130 nt + 21 nt paired-end sequencing on Illumina sequencers. Adapter sequences were removed from read 1 (130 nt), and UMI sequences were extracted from read 2 (21 nt). The processed read 1 sequences were then aligned to designed BDA amplicons using the Bowtie2 software\(^{38}\).
+
+UMI-based mutation calling. Next, reads aligned to each BDA amplicon were grouped by UMI. Reads carrying the same UMI sequence are amplified presumably from the same original DNA template, thus belong to the same UMI family. If the UMI sequence contained unexpected bases that do not match the expected format (H\(_{15}\)), the UMI family was removed.
+
+Because small UMI family size (i.e. number of reads in the UMI family) might be a result of amplification or sequencing error in the UMI region, UMI families with small family size are removed. To adjust for the difference derived from sequencing depth, we use a “dynamic cutoff” to remove small UMI families. If the family size was \( \leq 3 \) or smaller than 5% of the mean of top 3 family size in the same amplicon, the UMI family was removed.
+
+We next performed *do novo* variant call for each BDA enrichment region. In an effective NGS read, the forward primer and the 10 nt after the enrichment region need to match the corresponding regions in the BDA amplicon. The consensus sequence of each UMI family was the enrichment region sequence appearing most often in the UMI family. If two sequences had the same frequency and were the most common, consensus sequence was arbitrarily selected from these two. The consensus sequences were then compared to the wild type enrichment region, and variants were recorded.
+
+Mutation filtering by UMI count. Polymerase error may occur during the PCR cycle of UMI attachment. In order to minimize false positives, we applied UMI count filter and VAF filter to
+remove mutation calls that are less likely clinically relevant. The UMI count filter removes mutation calls with <6 UMI family count; and the VAF filter removes mutation calls with lower than defined LoD threshold. The count filter and VAF filter aim to address potential polymerase misincorporation errors, sequencing errors, potential DNA damage and clonal hematopoiesis.
+
+Digital droplet PCR. Digital PCR was performed using Bio-Rad QX200 Droplet Digital PCR System. Mutation VAF was confirmed using BioRad ddPCR NRAS Q61K Kit (BioRad Assay ID: dHsaMDV2010067) and BioRad ddPCR BRAF V600 Screening Kit (Catalogue # 12001037). Copy number of BRAF was confirmed with BRAF CNV FAM assay (BioRad Assay ID: dHsaCP2500366) and EIF2C1 (Ref) HEX assay (BioRad Assay ID: dHsaCP2500349).
+
+Conventional NGS for AML clinical samples: DNA was extracted from bone marrow samples and NGS was performed on clinical-grade, Clinical Laboratory Improvement Amendments-compliant platforms using an Illumina MiSeq system (Illumina, Inc., San Diego, CA, USA). The NGS panels included genes frequently affected in hematologic malignancies (panels of 28, 53, or 81 genes developed at MD Anderson\(^{28}\); see Supplementary Table S12 for the full list of genes). A minimum sequencing coverage of \( \times 250 \) (bidirectional true paired-end sequencing) was required. The analytical sensitivity was established at 5% mutant reads on a background of wild-type (WT) reads.
+
+Code availability: NGS data analysis pipeline for QBDA variant calling is available from Github (https://github.com/wrj915/QBDA).
+
+Acknowledgements: This work was supported by NIH awards U01CA233364 and R01CA203964 to D.Y.Z., and CPRIT award RP180147 to D.Y.Z.
+Author contributions: P.D., L.R.W. and D.Y.Z. conceived the project. P.D. and L.R.W. designed and conducted the experiments, and analyzed the data. S.X.C. and M.X.W. analyzed the data. L.Y.C designed melanoma panel. J.X.Z. and C.P.H. performed pan-cancer panel experiments. W.Y. performed NGS experiments. J.Z. and G.C.I. provided clinical AML samples and analyzed the data. L.K. provided melanoma clinical samples and analyzed the data. P.D. L.R.W. and D.Y.Z. wrote the paper with input from all authors.
+
+Competing interests: There are patents pending on the QBDA method. P.D., L.R.W., S.X.C., M.X.W., and L.Y.C. declares a competing interest in the form of consulting for Nuprobe USA. D.Y.Z. declares a competing interest in the form of consulting for and significant equity ownership in Nuprobe USA, Torus Biosystems, and Pana Bio.
+Figure 1. Quantitative Blocker Displacement Amplification (QBDA) technology.
+
+(a) QBDA library preparation workflow. UMIs are attached to DNA templates by 2 cycles of PCR, followed by pre-amplification using universal primers. Next, a nested BDA was performed to enrich variant sequence. The forward primer is closer to the variant position than the primer in the UMI additions step, in order to suppress primer dimer and nonspecific amplification; an overlapping Blocker suppresses the amplification of wild type (WT) templates, and allows enrichment of variant templates over many PCR cycles. The NGS adapter is added to the enrichment product, followed by index PCR and sequencing.
+
+(b) Reducing WT reads by QBDA enrichment. WT DNA was mixed with 9 synthetic DNA gBlocks, each containing a different single-base substitution or indel in a 16 nt region, resulting in about 1% variant allele frequency (VAF) for each mutation. Using standard amplicon-based sequencing without enrichment, 88% reads were used for unnecessary repeated sequencing of
+WT. Using QBDA, all 9 mutations were enriched using a single set of primer and Blocker, and the WT reads are suppressed to 2%.
+
+(c) Suppressing error and improving quantitation by UMI in QBDA. 675 types of false positive (non-expected) variants were observed in standard NGS in (b), occupying 12% of all variant reads, or 1.5% of total reads. All the false variants were removed using UMI-based error correction in QBDA. The observed molecule count for all spike-in variants were within 2-fold of the expected values.
+Figure 2. Characterization of QBDA AML panel for minimal residual disease (MRD) detection.
+
+(a) Limit of detection (LoD) threshold for different types of mutations.
+
+(b) Observed mutation VAF in a spike-in positive sample and a healthy PBMC sample.
+
+The positive sample was prepared by mixing Horizon Myeloid DNA Reference Standard, 3 synthetic gBlocks, and a gDNA sample extracted from healthy PBMC, resulting in VAF between 0.001% and 0.1% for 22 different mutations. 16 out of 22 mutations were around 0.01% VAF (between 0.005% and 0.02%). The “expected” VAF was quantitated by UMI-based NGS without mutation enrichment. All 22 mutations covered by the AML panel were observed in the positive sample (orange line); 82% of the mutations were within 2-fold of expected VAF. The same healthy PBMC sample was also analyzed alone as the paired negative sample using the AML panel (grey line). In healthy sample, some mutations (C>T or G>A) are observed at below-LoD level, possibly due to clonal hematopoiesis.
+Here 1 µg of gDNA was used for each library.
+
+(c) Quantitation accuracy. The positive sample in (b) was sequenced in triplicate NGS libraries; 2 additional positive samples with 3-fold or 5-fold VAF of the above-mentioned sample were also analyzed. For each of the 22 mutations, the observed VAF was in correct order for the 1X, 3X, and 5X VAF samples. In the triplicate experiment of the 1X VAF (≈0.01%) sample, 1 mutation was not observed in one of the replicates, thus the sensitivity is approximately 1-1/(22*3) = 98.5%. 1 µg of gDNA was used for each library.
+
+(d) Sequencing depth down to 45,000X does not affect sensitivity in 1X VAF (≈0.01%) sample. The 1X VAF positive sample (500 ng input) was sequenced with 350,000X depth (7.7 M reads). Even after down-sampling to 45,000X depth sequencing by random sampling 1.0 M reads from the original library, all mutations are observed. The median observed UMI counts from 20 independent simulations were plotted against observed UMI counts in the original library.
+Figure 3. QBDA for mutation detection during AML complete remission.
+
+(a-e) Changes of mutation VAF and the percentage of blasts in bone marrow from diagnosis to complete remission for each of the five patients. The mutations in \( NPM1 \) were highlighted in red and mutations in \( DTA \) (i.e., \( DNMT3A \), \( TET2 \), and \( ASXL1 \)) were highlighted in blue. Other mutations were shown in grey.
+
+(f) Summary of mutations detected from 5 patients during remission using the QBDA AML panel covering 22 hot spot regions in 20 genes.
+
+(g) Swimmer plot of clinical course and molecular findings of patients. QBDA identified NPM1 mutation in patient 1 during remission while flow cytometry reported MRD negative and conventional NGS failed to detect NPM1 mutation at the same time point. This NPM1 mutation was observed by conventional NGS during relapse. QBDA did not observe NPM1 mutation in
+patient 2 while MRD positive is reported by flow cytometry. In the two subsequent time points even after relapse NMP1 mutation was still not observed. Instead, de novo mutations in \( KDM6A \) and \( PHF6 \) were identified indicating clonal evolution occurred as alternative cause of relapse.
+Figure 4. Application of QBDA technology to pan-cancer hotspot large panel and Melanoma panel.
+
+(a) Compatibility of pan-cancer panel with ultralow frequency mutation analysis. Observed mutation VAF in a spike-in positive sample (VAF \( \approx 0.01\% \)) and the negative sample without spike-in are plotted. With mutation calling threshold setting at 0.006% VAF, the technical sensitivity was 95% based on duplicate test of spike-in positive sample.
+(b) Quantitation accuracy. The positive sample in (a) was sequenced in duplicate NGS libraries; 1 additional positive sample with 3-fold VAF of the above-mentioned sample were also analyzed. For each of the 20 spike-in mutations, the observed VAF was in correct order for the 1X and 3X samples.
+
+(c) Pan-cancer panel for quantitation of mutations down to 0.1% VAF using 1 M reads (mean sequencing depth 2,800X) in 16 clinical samples including FFPE, Fresh Frozen (FF) and cfDNA. 360 amplicons in hot spot regions of 61 genes are tested and only detected mutations are plotted here.
+
+(d) Melanoma panel for detection of mutations down to 0.1% in clinical samples. VAF of observed mutations in 23 FFPE or FF clinical samples from Melanoma patients are summarized. Co-existence of *BRAF* V600E and low frequency *NRAS* Q61K mutations in FFPE5 sample was observed.
+
+(e) QBDA quantitation exhibits less false negative and false positive variant calls than normal NGS without UMI. All observed variants in the 23 melanoma clinical samples are plotted.
+Reference:
+
+1. Ford, C. B. et al. Mycobacterium tuberculosis mutation rate estimates from different lineages predict substantial differences in the emergence of drug-resistant tuberculosis. Nat. Genet. **45**, 784–790 (2013).
+
+2. Lewis, K. Persister cells, dormancy and infectious disease. Nat. Rev. Microbiol. **5**, 48–56 (2007).
+
+3. Dagogo-Jack, I. & Shaw, A. T. Tumour heterogeneity and resistance to cancer therapies. Nat. Rev. Clin. Oncol. **15**, 81 (2018).
+
+4. Mansoori, B., Mohammadi, A., Davudian, S., Shirjang, S. & Baradaran, B. The different mechanisms of cancer drug resistance: a brief review. Adv. Pharm. Bull. **7**, 339 (2017).
+
+5. Kobayashi, S. et al. EGFR mutation and resistance of non–small-cell lung cancer to gefitinib. N. Engl. J. Med. **352**, 786–792 (2005).
+
+6. Ivey, A. et al. Assessment of Minimal Residual Disease in Standard-Risk AML. N. Engl. J. Med. **374**, 422–433 (2016).
+
+7. Jongen-Lavrencic, M. et al. Molecular Minimal Residual Disease in Acute Myeloid Leukemia. N. Engl. J. Med. **378**, 1189–1199 (2018).
+
+8. Short, N. J. et al. Association of measurable residual disease with survival outcomes in patients with acute myeloid leukemia: a systematic review and meta-analysis. JAMA Oncol. **6**, 1890–1899 (2020).
+
+9. Wan, J. C. M. et al. Liquid biopsies come of age: Towards implementation of circulating tumour DNA. Nat. Rev. Cancer **17**, 223–238 (2017).
+
+10. Schwarzenbach, H., Hoon, D. S. B. & Pantel, K. Cell-free nucleic acids as biomarkers in cancer patients. Nat. Rev. Cancer **11**, 426–437 (2011).
+11. Jaso, J. M., Wang, S. A., Jorgensen, J. L. & Lin, P. Multi-color flow cytometric immunophenotyping for detection of minimal residual disease in AML: past, present and future. Bone Marrow Transplant. **49**, 1129–1138 (2014).
+
+12. Xu, J., Jorgensen, J. L. & Wang, S. A. How do we use multicolor flow cytometry to detect minimal residual disease in acute myeloid leukemia. Clin Lab Med **37**, 787–802 (2017).
+
+13. Moss, J. *et al.* Comprehensive human cell-type methylation atlas reveals origins of circulating cell-free DNA in health and disease. *Nat. Commun.* **9**, (2018).
+
+14. Snyder, M. W., Kircher, M., Hill, A. J., Daza, R. M. & Shendure, J. Cell-free DNA Comprises an in Vivo Nucleosome Footprint that Informs Its Tissues-Of-Origin. *Cell* **164**, 57–68 (2016).
+
+15. Eckert, K. A. & Kunkel, T. A. DNA polymerase fidelity and the polymerase chain reaction. *Genome Res.* **1**, 17–24 (1991).
+
+16. Potapov, V. & Ong, J. L. Examining sources of error in PCR by single-molecule sequencing. *PLoS One* **12**, 1–19 (2017).
+
+17. Schirmer, M., D’Amore, R., Ijaz, U. Z., Hall, N. & Quince, C. Illumina error profiles: resolving fine-scale variation in metagenomic sequencing data. *BMC Bioinformatics* **17**, 1–15 (2016).
+
+18. Schirmer, M. *et al.* Insight into biases and sequencing errors for amplicon sequencing with the Illumina MiSeq platform. *Nucleic Acids Res.* **43**, e37–e37 (2015).
+
+19. Kinde, I., Wu, J., Papadopoulos, N., Kinzler, K. W. & Vogelstein, B. Detection and quantification of rare mutations with massively parallel sequencing. *Proc. Natl. Acad. Sci.* **108**, 9530–9535 (2011).
+
+20. Newman, A. M. *et al.* Integrated digital error suppression for improved detection of
+circulating tumor DNA. Nat. Biotechnol. **34**, 547–555 (2016).
+
+21. Schmitt, M. W. *et al.* Detection of ultra-rare mutations by next-generation sequencing. **2012**, 1–6 (2012).
+
+22. Abascal, F. *et al.* Somatic mutation landscapes at single-molecule resolution. *Nature* (2021) doi:10.1038/s41586-021-03477-4.
+
+23. Cohen, J. D. *et al.* Detection of low-frequency DNA variants by targeted sequencing of the Watson and Crick strands. *Nat. Biotechnol.* doi:10.1038/s41587-021-00900-z.
+
+24. Wu, L. R., Chen, S. X., Wu, Y., Patel, A. A. & Zhang, D. Y. Multiplexed enrichment of rare DNA variants via sequence-selective and temperature-robust amplification. *Nat. Biomed. Eng.* **1**, 714–723 (2017).
+
+25. Song, P. *et al.* Selective multiplexed enrichment for the detection and quantitation of low-fraction DNA variants via low-depth sequencing. *Nat. Biomed. Eng.* doi:10.1038/s41551-021-00713-0.
+
+26. Young, A. L., Challen, G. A., Birmann, B. M. & Druley, T. E. Clonal hematopoiesis harbouring AML-associated mutations is ubiquitous in healthy adults. *Nat. Commun.* **7**, (2016).
+
+27. Midic, D. *et al.* Prevalence and dynamics of clonal hematopoiesis caused by leukemia-associated mutations in elderly individuals without hematologic disorders. *Leukemia* **34**, 2198–2205 (2020).
+
+28. Luthra, R. *et al.* Next-generation sequencing-based multigene mutational screening for acute myeloid leukemia using MiSeq: applicability for diagnostics and disease monitoring. *Haematologica* **99**, 465 (2014).
+
+29. Cocciardi, S. *et al.* Clonal evolution patterns in acute myeloid leukemia with NPM1
+mutation. Nat. Commun. **10**, 1–11 (2019).
+
+30. Genovese, G. *et al.* Clonal hematopoiesis and blood-cancer risk inferred from blood DNA sequence. *N. Engl. J. Med.* **371**, 2477–2487 (2014).
+
+31. Buscarlet, M. *et al.* DNMT3A and TET2 dominate clonal hematopoiesis and demonstrate benign phenotypes and different genetic predispositions. *Blood, J. Am. Soc. Hematol.* **130**, 753–762 (2017).
+
+32. Jaiswal, S. *et al.* Age-related clonal hematopoiesis associated with adverse outcomes. *N. Engl. J. Med.* **371**, 2488–2498 (2014).
+
+33. Zink, F. *et al.* Clonal hematopoiesis, with and without candidate driver mutations, is common in the elderly. *Blood* **130**, 742–752 (2017).
+
+34. Shlush, L. I. Age-related clonal hematopoiesis. *Blood* **131**, 496–504 (2018).
+
+35. Morita, K. *et al.* Clearance of somatic mutations at remission and the risk of relapse in acute myeloid leukemia. *J. Clin. Oncol.* **36**, 1788 (2018).
+
+36. Morita, K. *et al.* Clonal evolution of acute myeloid leukemia revealed by high-throughput single-cell genomics. *Nat. Commun.* **11**, 1–17 (2020).
+
+37. Raaijmakers, M. I. G. *et al.* Co-existence of BRAF and NRAS driver mutations in the same melanoma cells results in heterogeneity of targeted therapy resistance. *Oncotarget* **7**, 77163–77174 (2016).
+
+38. Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2. *Nat. Methods* **9**, 357 (2012).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• QBDASlexcelfilev2.xlsx
+• QBDASI2021.docx

0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc/peer_review/peer_review.md

@@ -0,0 +1,331 @@
+Peer Review File
+
+A cleaner snow future mitigates Northern Hemisphere snowpack loss from warming
+
+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/.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The manuscript entitled “A cleaner snow future mitigates Northern Hemisphere snowpack loss from warming” by Dalei Hao and co-authors investigated the effects of future change in LAP (BC & Dust) on aerosol-induced snow darkening, radiative forcing and change in SWE. Using a climate model coupled with a sophisticated land surface scheme, they have shown that cleaner snow will alleviate the loss of snowpack in the future due to a warmer climate. However, the results presented in this study are well-known. The BC emission over the developing countries will decrease in the future (or that is already in the decreasing phase) and hence the amount of BC aerosols deposited on the snow surface will decrease and we expect cleaner snow in the future. Compared to historic conditions, cleaner snow in the future will absorb less radiation and a positive change in SWE is expected. So the novelty of this study is not very clear.
+
+If a reduction in BC could reduce 50% of the future snowpack loss over the Tibetan Plateau (Line no 18-20), what is the present-day contribution of BC to the observed snow cover loss? There are studies that reported comparable contributions of BC and dust on snowpack loss (Line no. 40). This gives the false impression that LAP is the sole contributor to the snow cover loss, irrespective of the warming due to greenhouse gases and LULC changes.
+
+Most of the Himalayan glaciers are debris covered. How does BC deposition on snow influence glacier melt? Most of the references cited in this manuscript investigated the effects of aerosol deposition on snow cover, not on glaciers. Please modify the manuscript accordingly. Do you have clear evidence for the effect of aerosols on glaciers instead of seasonal snow?
+
+Please provide the inter-comparison of simulated BC and observations over the distinct snow-covered regions, especially the Tibetan Plateau. Considering the uncertainties listed in line no. 247-270, what is the change in the contribution of BC-induced snow cover reduction? The flux of BC deposited on the snow surface is exorbitantly high. For example, to get 2 ng m-2 s-1 deposition flux at the surface, atmospheric BC concentration should be nearly 20 microgram m-3. Such a high concentration of BC does not exist over the Tibetan region.
+
+The BC RF at 2100 for SSP126 and SSP585 (Figure S4, solid lines: with future change of BC) are comparable. It shows that irrespective of the emission pathways (low or high), the BC forcing over the Tibetan Plateau remains the same in the future. Please explain this contradiction.
+The authors did not explain the reason for the increase in SAE over the Karakoram region. Is this indicating a decrease in temperature or an increase in precipitation for SSP126? If so, why is this pattern missing in SSP585?
+
+Most of the discussions in this manuscript were the seasonal mean of different variables from December to May. But for SWE, instead of considering the entire snow season, they have considered only April. Please explain. It looks like the change in SWE is not very significant for the entire snow season.
+
+The authors highlighted in the abstract that “The reduced black carbon contamination in snow over the Tibetan Plateau will alleviate future snowpack loss due to climate change by 52.1±8.0% and 8.0±1.1% for the two scenarios.” The authors later stated in line no. 200-204 that this change is due to LAP (BC+Dust). Please clarify.
+
+Minor comments:
+
+Line no. 219: The cited reference does not report “the melting of snow and the retreat of glaciers due to LAP”.
+
+Line no. 219-222: This is quite obvious. Anthropogenic emissions of LAP will decrease in the future, so the decrease in LAP deposited on the snow surface is of no surprise.
+
+Figure S1: unit is not given.
+
+Figure S2: Values reported in ng m-2 unit, but most of the in-situ measurements are in the unit of ng/g. Convert the unit used in all figures to units used for measurements.
+
+Reviewer #2 (Remarks to the Author):
+
+The authors estimate that the decreasing trend of the snow water equivalent caused by climate change will slow down in the future especially at the Tibetan Plateau. This is mainly due to a decrease in Black Carbon emissions. The Earth System Model ELM, is used for the calculation.
+
+I find the authors have used appropriate techniques and made a thorough analysis and interpretation. They have presented the results in sufficient details and careful discuss the method they have used.
+
+I find the analysis is valid and the evidence for the conclusions are sufficiently strong. However, I find the study is rather narrow and would have liked to see a more comprehensive discussion on global climate aspects and maybe also a discussion on feedback mechanisms. The study has some significance, suggesting the findings could apply to other areas as well but also the presentation of the method will have some impact.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+The manuscript entitled “A cleaner snow future mitigates Northern Hemisphere snowpack loss from warming” by Dalei Hao and co-authors investigated the effects of future change in LAP (BC & Dust) on aerosol-induced snow darkening, radiative forcing and change in SWE. Using a climate model coupled with a sophisticated land surface scheme, they have shown that cleaner snow will alleviate the loss of snowpack in the future due to a warmer climate. However, the results presented in this study are well-known. The BC emission over the developing countries will decrease in the future (or that is already in the decreasing phase) and hence the amount of BC aerosols deposited on the snow surface will decrease and we expect cleaner snow in the future. Compared to historic conditions, cleaner snow in the future will absorb less radiation and a positive change in SWE is expected. So the novelty of this study is not very clear.
+
+Thanks for your constructive comments and suggestions. We have revised the manuscript carefully.
+
+Sorry for the unclear statements on some research gaps filled by this study. We have reorganized and highlighted them as below:
+
+1. Quantify and understand the future evolution of LAP deposition and RF in snow. Although the historical and present-day LAP impacts on snowpack have been widely recognized, how the LAP emission and deposition over snow-covered regions will change in the future and further impact snowpack in a warmer climate remains unknown. A clean or dirty snow future could have very different impacts on snowmelt and snow water resources with large societal implications.
+
+2. Quantify separately the relative contribution of climate change due to greenhouse gases and LAP evolution to future snowpack change. Future global warming is projected to reduce the snow water resources globally. Simultaneously, the future change of LAP darkening effects due to changes in LAP emissions will also impact snow water availability. Understanding the relative contribution of climate change and LAP to future snowpack change under different SSP scenarios is critical for constraining projections of downstream freshwater availability from snowmelts.
+3. Quantify the uncertainty in snowpack projection from the model configurations using the state-of-art E3SM Land Model (ELM). Generally, climate models (e.g., CMIP6 models) predict a significant decrease of future global snowpack due to global warming. However, such snowpack projection can be biased because most climate models neglect or oversimplify the effects of snow darkening due to BC and dust deposition. The ELM can prognostically simulate the change of LAP concentrations in different snow layers after deposition and the impacts of these changes on snow albedo (Hao et al., 2022). By carrying out historical and future ELM simulations with different model configurations of topography, land use and land cover change, LAP-snow mix state, snow grain shape and BC meltwater scavenging efficiency, we quantified the uncertainties in snowpack projection associated with the model configurations.
+
+We have clarified and strengthened these key points throughout the revised manuscript (especially in the abstract and introduction).
+
+1. If a reduction in BC could reduce 50% of the future snowpack loss over the Tibetan Plateau (Line no 18-20), what is the present-day contribution of BC to the observed snow cover loss? There are studies that reported comparable contributions of BC and dust on snowpack loss (Line no. 40). This gives the false impression that LAP is the sole contributor to the snow cover loss, irrespective of the warming due to greenhouse gases and LULC changes.
+
+Sorry for the confusion. Our results show that future LAP change in snow over the Tibetan Plateau will alleviate future snowpack loss due to climate change by 52.1±8.0% and 8.0±1.1% for the two scenarios, mainly due to reduced black carbon contamination. We stress that future BC deposition reduction will alleviate the snowpack change caused by global warming. To avoid the confusion, we have revised the abstract to make the statement clear.
+
+We have also analyzed the historical ELM simulations to quantify the historical contribution of BC to snowpack by comparing the differences between the historical simulations with and without BC deposition (Figure R1). The results show that the deposited BC in snow reduces the historical Dec-May average SWE by up to 25-30 mm over the TP. Our findings are consistent with Ji (2016) that reported a decrease in average SWE in the non-monsoon season by up to more than 25 mm over the TP. Sarangi et al. (2020) revealed that the influence of dust
+on snow darkening could be greater than that of BC over the high-latitude high-mountain Asia regions, and our results show a similar pattern (Figure R2). We have now added these results in Line 127-128 and 251-253 of the revised manuscript.
+
+![BC-induced reduction of historical (1995-2014) average SWE from December to May. The SWE reduction is calculated based on ELM historical simulations with and without BC deposition. The grids where the average SWE during December to May is smaller than 5 mm are masked.](page_246_370_1017_370.png)
+
+Figure R1| BC-induced reduction of historical (1995-2014) average SWE from December to May. The SWE reduction is calculated based on ELM historical simulations with and without BC deposition. The grids where the average SWE during December to May is smaller than 5 mm are masked.
+
+![Ratio of historical (1995-2015) average dust-induced albedo reduction to total albedo reduction over the TP. The snow albedo reduction is calculated based on ELM outputs from December to May. The grids where the average SWE during December to May is smaller than 5 mm are masked.](page_246_770_1017_370.png)
+
+Figure R2| Ratio of historical (1995-2015) average dust-induced albedo reduction to total albedo reduction over the TP. The snow albedo reduction is calculated based on ELM outputs from December to May. The grids where the average SWE during December to May is smaller than 5 mm are masked.
+
+2. Most of the Himalayan glaciers are debris covered. How does BC deposition on snow influence glacier melt? Most of the references cited in this manuscript investigated the effects of aerosol deposition on snow cover, not on glaciers. Please
+modify the manuscript accordingly. Do you have clear evidence for the effect of aerosols on glaciers instead of seasonal snow?
+
+Thanks for the insightful comment. Considering that our study focuses on the LAP darkening effects on seasonal snowpack, we have deleted the irrelevant glacier-related citations and descriptions in the revised manuscript.
+
+3. Please provide the inter-comparison of simulated BC and observations over the distinct snow-covered regions, especially the Tibetan Plateau. Considering the uncertainties listed in line no. 247-270, what is the change in the contribution of BC-induced snow cover reduction? The flux of BC deposited on the snow surface is exorbitantly high. For example, to get 2 ng m-2 s-1 deposition flux at the surface, atmospheric BC concentration should be nearly 20 microgram m-3. Such a high concentration of BC does not exist over the Tibetan region.
+
+Thank you for the good suggestion! We have added a comparison of simulated and observed BC concentration across the NH in Line 409-435 of the revised manuscript. Specifically, considering the temporal coverage of our historical simulations (1995-2014) and the availability of field observations, we have collected some field measurements of BC concentration in snow during 2000-2014 over the Arctic, Tibetan Plateau, North China, and North America from Doherty et al. (2010, 2014), He et al. (2018), Kang et al. (2022), Zhang et al. (2018) and Wang et al. (2013) (Figure R3a; Table S4 of the revised manuscript). The snow samples affected by drifting snow and with poor spatial representativeness reported in the relevant studies were excluded. For the estimates from the Integrating Sphere integrating SandWich (ISSW) spectrophotometer, we scaled the measured BC concentration in snow (Table S4 of the revised manuscript) to match the BC mass absorption efficiency of 7.5 m^2 g^{-1} used in ELM, following Qian et al. (2014). We collected information of both the BC concentration in the top snow layer and snow column. For the snow samples with unreported BC concentration in snow column, we averaged the surface and sub-surface BC concentration to get the approximate values. For the model evaluation, the snow samples within the same model grid cell for the same month and year were aggregated, and observations and simulations for the same geo-location, month, and year are compared.
+
+Overall, the ELM simulations are statistically in good agreements with the observations for both the BC concentration in the top snow layer and snow column (Figure R3b,c). The correlation coefficients between the simulation and observation are 0.66 and 0.73, respectively, for BC concentration in the top snow layer and snow column. Overall, 78.4% of the simulated BC concentration in the
+top snow layer is within a factor of four of the observed concentrations, and the number increases to 84.9% for BC concentration in the snow column. The ELM simulations show an overestimation of BC concentration in the top snow layer, while the positive biases in ELM simulations are smaller for the BC concentration in the snow column, especially for the Arctic and North America sites. As suggested by Doherty et al. (2014), BC concentration in snow column could be a better metric for model-observation comparisons because it smooths out the effects of new snowfall events, variations in BC deposition rates, and the melting, percolation, and refreezing processes after deposition in snow. Note that the partial inconsistencies between observations and simulations are also possibly caused by the spatial and temporal mismatch between simulations and field measurements, and uncertainties in field sampling and lab measurements of BC in meltwater. These results demonstrate that ELM can accurately estimate the LAP darkening effects on snow. We have added this evaluation in the Methods section of the revised manuscript.
+Figure R3| Comparison of ELM-simulated and observed BC concentration in snow across the NH. a. Spatial distribution of field snow samples. b,c Scatter plots between observed and simulated BC concentration in the top snow layer and snow column. In (b,c), the dotted, dashed and solid lines are 1:1, 1:4 (or 4:1) and 1:10 (or 10:1) ratio lines, and the correlation coefficient and p value are labeled.
+
+Regarding the impact of model uncertainty on the estimates of BC-induced snow cover reduction, the above-mentioned evaluation of the ELM simulations against the field observations confirms the reliability of ELM in simulating BC concentration. Our sensitivity analysis also demonstrated that the model configurations related to topography, land use and land cover change, LAP-snow mixing state, snow grain shape, and BC melt-water scavenging efficiency have small impacts on our conclusions. Besides, our analyses are primarily based on relative differences rather than absolute values, mitigating the impact of model uncertainties. We presented these in Line 273-317 of the revised manuscript.
+
+Regarding the magnitude of BC deposition rate, we believe that the reviewer’s estimation of atmospheric BC concentration of 20 microgram m^{-3} to get 2 ng m^{-2} s^{-
+1 BC deposition flux assumed only dry deposition of BC. However, both wet and dry depositions contribute to BC deposition. Dry deposition is mainly determined by the underlying surface characteristics and micrometeorological conditions and its velocity is generally within the range of 0.01–0.07 cm s\(^{-1}\) (Zhou et al., 2018). In contrast, BC wet deposition can be more efficient, depending on local precipitation (Figure R4). Wet deposition over the snow-covered regions account for over 80% of the total deposition (Figure R5). These ELM results are consistent with Barrett et al. (2019), He et al. (2014) and Textor et al. (2006). Besides, the field measurements at three stations in the Himalayas and southern TP show that the annual BC deposition rate can be 58.9 mg·m\(^{-2}\)·year\(^{-1}\), which is equal to 1.9 ng m\(^{-2}\) s\(^{-1}\) (Table 2 in Yan et al., 2019). He et al. (2014) also showed that annual BC deposition can be larger than \(10^{-5}\) kg m\(^{-2}\) month\(^{-1}\), which is equal to 3.8 ng m\(^{-2}\) s\(^{-1}\). We have added Figure R4 in the revised manuscript.
+
+![Spatial patterns of historical precipitation over the Northern Hemisphere (NH). Here, the precipitation rate is calculated based on the ensemble mean of seven CMIP6 model outputs from December to May.](page_256_646_1024_482.png)
+
+Figure R4| Spatial patterns of historical precipitation over the Northern Hemisphere (NH). Here, the precipitation rate is calculated based on the ensemble mean of seven CMIP6 model outputs from December to May.
+Figure R5| Spatial patterns of historical (1995-2014) BC deposition rate over the Northern Hemisphere (NH): a total, b wet, c dry deposition, and d the ratio of wet to total deposition. Here, the BC deposition rates are calculated based on the ensemble mean of seven CMIP6 model outputs from December to May. In each panel, grids with an average SWE from December to May smaller than 5 mm are masked.
+
+4. The BC RF at 2100 for SSP126 and SSP585 (Figure S4, solid lines: with future change of BC) are comparable. It shows that irrespective of the emission pathways (low or high), the BC forcing over the Tibetan Plateau remains the same in the future. Please explain this contradiction.
+
+Although SSP126 and SSP585 show different decreasing trends during 2015-2100, the BC RFs at the end of this century (2081-2100) are comparable between
+SSP126 and SSP585 (0.57 and 0.61 W/m\(^{-2}\), respectively). Despite the significant difference in the CO2 emissions between SSP126 and SSP585, both scenarios present a significant BC emission reduction around the TP (Figure R6), which in turn leads to a significant reduction in the deposition of BC over the TP at the end of this century (Figure R7), This is because the main source of CO2 and BC emission are different. For example, the energy sector tends to dominate the behavior of CO2 emissions, while the residential commercial sectors (e.g., the biomass usage and mobile sources) overall dominate BC emissions across various future scenarios (Gidden et al., 2019; Xu et al., 2021).
+
+The reduction of BC emission and deposition in the future is because of the reduction of BC emissions from the residential and commercial sector (e.g., biomass burning and motor vehicle diesel), which account for nearly 40 % of all BC emissions in the historical time period. However, at the end of the century, the contribution of the emissions from the residential and commercial sector is projected to decrease to a low level under both SSP126 and SSP585 due to a transition away from traditional biomass usage with the increased economic development and population stabilization and emissions controls on mobile sources (Gidden et al., 2019). For instance, BC emissions around the TP are the highest in China and India primarily due to traditional biomass usage in the residential sector and secondarily due to transport-related activity. In scenarios of high socioeconomic development and technological progress, such as SSP126 and SSP585, BC emissions across most countries decline dramatically by the end of the century. For instance, the total BC emissions in China are equal to those of the USA today. Note that even in SSP585, air pollutant emissions including sulfate and carbonaceous aerosols (i.e., organic and black carbon) are tightly controlled for environmental and health reasons (Kriegler et al., 2017). We added these explanations in Line 87-105 and 145-148 of the revised manuscript.
+
+![Spatial patterns of historical (1995-2100) and future (2081-2100) BC emission rates over the globe. Here, the BC emission rates are calculated based on the ensemble mean of seven CMIP6 model outputs from December to May.](page_184_1342_1207_246.png)
+
+Figure R6 Spatial patterns of historical (1995-2100) and future (2081-2100) BC emission rates over the globe. Here, the BC emission rates are calculated based on the ensemble mean of seven CMIP6 model outputs from December to May.
+Figure R7| Time series of average BC deposition from December to May over snow-covered Tibetan Plateau (TP) regions (where the average SWE exceeds 5 mm in the historical period of 1995-2014) under SSP126 and SSP585. The solid line and background shading represent the mean and standard deviation of BC deposition rates, respectively, based on the seven CMIP6 models.
+
+![Time series of average BC deposition from December to May over snow-covered Tibetan Plateau (TP) regions under SSP126 and SSP585](page_184_312_1012_496.png)
+5. The authors did not explain the reason for the increase in SAE over the Karakoram region. Is this indicating a decrease in temperature or an increase in precipitation for SSP126? If so, why is this pattern missing in SSP585?
+
+Our results show that most TP regions will have increased precipitation and air temperature (Figure R8), as indicated by Yao et al. (2022). Different from other TP regions, the Karakoram region will have increased snowfall, which is consistent with Kapnick et al. (2014) and de Kok et al. (2018). This anomalous SWE phenomenon in the Karakoram could be linked to its unique seasonal cycle and winter precipitation, making the snow less sensitive to warming (Kapnick et al., 2014) and an increase in snowfall (de Kok et al., 2018; Yao et al., 2022). Consequently, the future SWE in Karakoram increases under SSP126 but stays relatively stable under SSP585. Compared to SS126, SSP585 projects significantly warmer air temperature (Figure R8h,i), leading to faster snowmelt than SSP126, which possibly explains the difference between these two scenarios. We have added these explanations in Line 203-207 of the revised manuscript.
+
+![Spatial patterns of historical and future precipitation, snowfall and air temperature over the TP.](page_186_1012_1207_496.png)
+
+Figure R8| Spatial patterns of historical and future precipitation, snowfall and air temperature over the TP. a,d,g Historical (1995-2014) spatial patterns of climate conditions. b,c,e,f,h,i The difference between future (2081-2100) and historical climate conditions under SSP126 and SSP585. Historical and future climate conditions are calculated based on CESM outputs from December to May.
+6. Most of the discussions in this manuscript were the seasonal mean of different variables from December to May. But for SWE, instead of considering the entire snow season, they have considered only April. Please explain. It looks like the change in SWE is not very significant for the entire snow season.
+
+Seasonal snowpack generally peaks around April in many NH mid-latitude snow-dominated regions, e.g., the Tibetan Plateau (Liu et al., 2021), Western US. The peak SWE is one of the metrics used to assess potential water supply outcomes (Kraaijenbrink et al., 2021). In this consideration, April SWE can provide useful insight into the expected spring runoff and inform reservoir operation and seasonal water supply forecasts that critically support agricultural and resource management decisions. Therefore, we selected April SWE in the analysis. We have added these explanations in Line 193-196 of the revised manuscript.
+
+Additionally, we have computed the December-May average SWE, which shows the same patterns as those for April SWE (Figure R10 and R11), but with a relatively smaller magnitude due to the averaging effects. Specifically, for April SWE, the future LAP change in snow over the TP will alleviate future snowpack loss due to climate change by 52.1±8.0% and 8.0±1.1% for SSP126 and SSP585, respectively. For the December-May average SWE, those values are 33.3±4.5% and 5.6±0.6%, respectively (Figure R10). We have added the corresponding figures on the December-May average SWE in the revised manuscript.
+Figure R9| Spatial distribution of peak SWE timing (represented by day of water year (DOWY)). The inset figure is the histogram of peak SWE DOWY. The three dates labeled in the color bar (DOWY 133, DOWY 169 and DOWY 217) correspond to the 10th, 50th and 90th percentile in the DOWY distribution and are marked with vertical dashed lines in the inset histogram. This figure is cited from Liu et al. (2021).
+
+Figure R10| Historical and future average SWE from December to May over the snow-covered TP regions (where the average SWE exceeds 5 mm in the historical period of 1995-2014) under different model configurations. For each panel, Climate_hist+LAP_hist represents the historical (1995-2014) simulations with historical LAP depositions, while Climate_future+LAP_future and Climate_future+LAP_hist represent future (2081-2100) simulations with and without a future change of LAP depositions, respectively. The Climate_future+LAP_hist simulations used the historical average LAP depositions from 1995-2014. The horizontal axis labels represent different model configurations (see Methods), where Control has the ELM default settings and the others represent major adjustments made from the Control case. Specifically, PP assumes that the terrain is flat and neglects topographic effects on solar radiation; Koch assumes a non-spherical snow grain shape (Koch snowflake); extBC assumes external mixing between hydrophilic BC and snow grains; intDust assumes internal mixing between dust and snow grains; noLULCC has no land use and land cover change; MSE_high assumes high melt-water scavenging efficiency
+(MSE = 2, much higher than the default value of 0.2) of hydrophilic BC; and MSE_low assumes a low MSE (0.02) of hydrophilic BC. In (a,b), the contribution (\( \delta_{LAP} \)) of future LAP change that mitigates snowpack loss under each ELM configuration is noted as a percentage and is calculated as the ratio of the SWE difference (\( \Delta SWE_{LAP} \)) between Climate_future+LAP_future and Climate_future+LAP_hist to the SWE difference (\( \Delta SWE_{climate} \)) between Climate_hist+LAP_hist and Climate_future+LAP_future.
+
+![Future average SWE from December to May and contributions of LAP change to future SWE. a Historical (1995-2014) and b,c future (2081-2100) spatial patterns of average SWE under SSP126 and SSP585. d,e The differences (ASWE_LAP) of future (2081-2100) SWEs with and without LAP change under SSP126 and SSP585. f,g Time series of relative \( \Delta SWE_{LAP} \) (calculated as the ratio of \( \Delta SWE_{LAP} \) to projected SWE without LAP change) over snow-covered regions (where the average SWE exceeds 5 mm in the historical period) in the TP. In (a-e), grids with an average SWE smaller than 5 mm in the historical period are masked. In (f,g), BC_future+Dust_future, BC_future+Dust_hist, and BC_hist+Dust_future represent different combinations of BC and dust depositions, where the subscripts of future and hist represent future and historical average depositions, respectively.](page_172_370_1207_1017.png)
+
+Figure R11| Future average SWE from December to May and contributions of LAP change to future SWE. a Historical (1995-2014) and b,c future (2081-2100) spatial patterns of average SWE under SSP126 and SSP585. d,e The differences (ASWE_LAP) of future (2081-2100) SWEs with and without LAP change under SSP126 and SSP585. f,g Time series of relative \( \Delta SWE_{LAP} \) (calculated as the ratio of \( \Delta SWE_{LAP} \) to projected SWE without LAP change) over snow-covered regions (where the average SWE exceeds 5 mm in the historical period) in the TP. In (a-e), grids with an average SWE smaller than 5 mm in the historical period are masked. In (f,g), BC_future+Dust_future, BC_future+Dust_hist, and BC_hist+Dust_future represent different combinations of BC and dust depositions, where the subscripts of future and hist represent future and historical average depositions, respectively.
+7. The authors highlighted in the abstract that “The reduced black carbon contamination in snow over the Tibetan Plateau will alleviate future snowpack loss due to climate change by 52.1±8.0% and 8.0±1.1% for the two scenarios.��� The authors later stated in line no. 200-204 that this change is due to LAP (BC+Dust). Please clarify.
+
+Thank you for pointing out this inconsistency. Here the snowpack change is indeed caused by LAP (sum of BC and Dust) change. However, our results also show that the future reduction of BC concentration dominates the snowpack change in Figure 4f,g of the revised manuscript. Therefore, we have revised Line 18-21 in the abstract to “the projected LAP changes in snow over the Tibetan Plateau will alleviate future snowpack loss due to climate change by 52.1±8.0% and 8.0±1.1% at the end of the century for the two scenarios, mainly due to reduced black carbon contamination.”.
+
+Minor comments:
+1. Line no. 219: The cited reference does not report “the melting of snow and the retreat of glaciers due to LAP”.
+
+Sorry for the wrong citation. As suggested in major comment #2 (Reviewer #1), we have deleted the less relevant, glacier-related citations in the revised manuscript, considering that our study focuses on the LAP darkening effects on seasonal snowpack.
+
+2. Line no. 219-222: This is quite obvious. Anthropogenic emissions of LAP will decrease in the future, so the decrease in LAP deposited on the snow surface is of no surprise.
+
+Indeed, the historical and present-day LAP impacts on snowpack have been widely recognized. However, the future evolution of LAP deposition and RF in snow are not well quantified. The relative contribution of climate change and LAP to future snowpack change remains unknown. We have clarified and strengthened these key points throughout the revised manuscript. Please also see our response to the overall comment from Reviewer #1 for details.
+
+3. Figure S1: unit is not given.
+
+We have added the units accordingly. This figure is for the relative trend of BC and dust deposition. We have also added a figure showing the absolute trend of BC and dust deposition in the revised manuscript.
+4. Figure S2: Values reported in ng m-2 unit, but most of the in-situ measurements are in the unit of ng/g. Convert the unit used in all figures to units used for measurements.
+
+Unfortunately, we did not output LAP concentration in the top snow layer with the units of (ng/g) in the ELM simulations. Thus, we have replaced it with the LAP concentration in snow column, which has the units of ng/g, in the revised manuscript. We have modified other figures accordingly in the revised manuscript.
+
+References
+Barrett, T.E., Ponette-González, A.G., Rindy, J.E. and Weathers, K.C., 2019. Wet deposition of black carbon: A synthesis. Atmospheric environment, 213, pp.558-567.
+de Kok, R.J., Tuinenburg, O.A., Bonekamp, P.N. and Immerzeel, W.W., 2018. Irrigation as a potential driver for anomalous glacier behavior in High Mountain Asia. Geophysical research letters, 45(4), pp.2047-2054.
+Doherty, S.J., Warren, S.G., Grenfell, T.C., Clarke, A.D. and Brandt, R.E., 2010. Light-absorbing impurities in Arctic snow. Atmospheric Chemistry and Physics, 10(23), pp.11647-11680.
+Gidden, M.J., Riahi, K., Smith, S.J., Fujimori, S., Luderer, G., Kriegler, E., Van Vuuren, D.P., Van Den Berg, M., Feng, L., Klein, D. and Calvin, K., 2019. Global emissions pathways under different socioeconomic scenarios for use in CMIP6: a dataset of harmonized emissions trajectories through the end of the century. Geoscientific model development, 12(4), pp.1443-1475.
+Hao, D., Bisht, G., He, C., Bair, E., Huang, H., Dang, C., Rittger, K., Gu, Y., Wang, H., Qian, Y. and Leung, L.R., 2022. Improving snow albedo modeling in E3SM land model (version 2.0) and assessing its impacts on snow and surface fluxes over the Tibetan Plateau. Geoscientific Model Development Discussions, pp.1-31.
+He, C., Flanner, M.G., Chen, F., Barlage, M., Liou, K.N., Kang, S., Ming, J. and Qian, Y., 2018. Black carbon-induced snow albedo reduction over the Tibetan Plateau: uncertainties from snow grain shape and aerosol–snow mixing state based on an updated SNICAR model. Atmospheric Chemistry and Physics, 18(15), pp.11507-11527.
+He, C., Li, Q.B., Liou, K.N., Zhang, J., Qi, L., Mao, Y., Gao, M., Lu, Z., Streets, D.G., Zhang, Q. and Sarin, M.M., 2014. A global 3-D CTM evaluation of black carbon in the Tibetan Plateau. Atmospheric Chemistry and Physics, 14(13), pp.7091-7112.
+Ji, Z.M., 2016. Modeling black carbon and its potential radiative effects over the Tibetan Plateau. Advances in Climate Change Research, 7(3), pp.139-144.
+Kang, S., Zhang, Y., Chen, P., Guo, J., Zhang, Q., Cong, Z., Kaspari, S., Tripathee, L., Gao, T., Niu, H. and Zhong, X., 2022. Black carbon and organic carbon dataset over the Third Pole. Earth System Science Data, 14(2), pp.683-707.
+Kraaijenbrink, P.D., Stigter, E.E., Yao, T. and Immerzeel, W.W., 2021. Climate change decisive for Asia’s snow meltwater supply. Nature Climate Change, 11(7), pp.591-597.
+Kriegler, E., Bauer, N., Popp, A., Humpenöder, F., Leimbach, M., Streifler, J., Baumstark, L., Bodirsky, B.L., Hilaire, J., Klein, D. and Mouratiadou, I., 2017. Fossil-fueled development (SSP5): An energy and resource intensive scenario for the 21st century. Global environmental change, 42, pp.297-315.
+Liu, Y., Fang, Y. and Margulis, S.A., 2021. Spatiotemporal distribution of seasonal snow water equivalent in High Mountain Asia from an 18-year Landsat–MODIS era snow reanalysis dataset. The Cryosphere, 15(11), pp.5261-5280.
+Qian, Y., Wang, H., Zhang, R., Flanner, M.G. and Rasch, P.J., 2014. A sensitivity study on modeling black carbon in snow and its radiative forcing over the Arctic and Northern China. Environmental Research Letters, 9(6), p.064001.
+Sarangi, C., Qian, Y., Rittger, K., Ruby Leung, L., Chand, D., Bormann, K.J. and Painter, T.H., 2020. Dust dominates high-altitude snow darkening and melt over high-mountain Asia. Nature Climate Change, 10(11), pp.1045-1051.
+Textor, C., Schulz, M., Guibert, S., Kinne, S., Balkanski, Y., Bauer, S., Berntsen, T., Berglen, T., Boucher, O., Chin, M. and Dentener, F., 2006. Analysis and quantification of the diversities of aerosol life cycles within AeroCom. Atmospheric Chemistry and Physics, 6(7), pp.1777-1813.
+Wang, X., Doherty, S.J. and Huang, J., 2013. Black carbon and other light-absorbing impurities in snow across Northern China. Journal of Geophysical Research: Atmospheres, 118(3), pp.1471-1492.
+Xu, H., Ren, Y.A., Zhang, W., Meng, W., Yun, X., Yu, X., Li, J., Zhang, Y., Shen, G., Ma, J. and Li, B., 2021. Updated global black carbon emissions from 1960 to 2017: improvements, trends, and drivers. Environmental Science & Technology, 55(12), pp.7869-7879.
+Yan, F., He, C., Kang, S., Chen, P., Hu, Z., Han, X., Gautam, S., Yan, C., Zheng, M., Sillanpää, M. and Raymond, P.A., 2019. Deposition of organic and black carbon: direct measurements at three remote stations in the Himalayas and Tibetan Plateau. Journal of Geophysical Research: Atmospheres, 124(16), pp.9702-9715.
+Yao T, Thompson L, Chen D, Chettri N., 2022. A Scientific Assessment of the Third Pole Environment.
+Zeng, X., Broxton, P. and Dawson, N., 2018. Snowpack change from 1982 to 2016 over conterminous United States. Geophysical Research Letters, 45(23), pp.12-940.
+
+Zhang, Y., Kang, S., Sprenger, M., Cong, Z., Gao, T., Li, C., Tao, S., Li, X., Zhong, X., Xu, M. and Meng, W., 2018. Black carbon and mineral dust in snow cover on the Tibetan Plateau. The Cryosphere, 12(2), pp.413-431.
+
+Zhou, J., Tie, X., Xu, B., Zhao, S., Wang, M., Li, G., Zhang, T., Zhao, Z., Liu, S., Yang, S. and Chang, L., 2018. Black carbon (BC) in a northern Tibetan mountain: effect of Kuwait fires on glaciers. Atmospheric Chemistry and Physics, 18(18), pp.13673-13685.
+Reviewer #2 (Remarks to the Author):
+
+The authors estimate that the decreasing trend of the snow water equivalent caused by climate change will slow down in the future especially at the Tibetan Plateau. This is mainly due to a decrease in Black Carbon emissions. The Earth System Model ELM, is used for the calculation. I find the authors have used appropriate techniques and made a thorough analysis and interpretation. They have presented the results in sufficient details and careful discuss the method they have used. I find the analysis is valid and the evidence for the conclusions are sufficiently strong. However, I find the study is rather narrow and would have liked to see a more comprehensive discussion on global climate aspects and maybe also a discussion on feedback mechanisms. The study has some significance, suggesting the findings could apply to other areas as well but also the presentation of the method will have some impact.
+
+We appreciate your comments and suggestions, and we have revised the manuscript accordingly!
+
+Firstly, we have reorganized the manuscript to strengthen and highlight the contributions of this study in the abstract and the introduction section, including 1. Quantify and understand the future evolution of LAP deposition and RF in snow; 2. Isolate the relative contribution of climate change and LAP evolution to future snowpack change; and 3. Quantify the uncertainty from the model configurations using the state-of-art E3SM Land Model (ELM). Please see our detailed response to Reviewer #1’s comments for details.
+
+Secondly, we have added more discussion in Line 338-353 of the revised manuscript on the potential impacts of our findings on environmental processes, socio-economic activities, and climate to broaden the impacts of our study. Specifically, we expect cascading impacts of a cleaner snow future on environmental processes, socio-economic activities, and climate. For example, the updated snowpack and snow phenology (i.e., evolution and duration) will potentially impact the mountain socio-ecological systems, e.g., the spring vegetation phenology (Wang et al., 2013) and thus terrestrial carbon cycle. The resulted increased availability of snow water resource may alleviate the future threats to the downstream snowmelt-dependent agricultural production caused by global warming (Biemans et al., 2019) and complicate future flood control and reservoir management. The increased snow cover may slow down the future glacier retreat (Painter et al., 2013). The cleaner snow future will also regulate the regional and global climate via snow-atmosphere coupling (Henderson et al.,
+2018). The air temperature will decrease with the increased snow cover. Due to the complex atmospheric feedback, the increased or decreased snowfall can alleviate or aggravate future snowpack loss under a warming climate. Conceivably, the resulted increase of snow cover over the TP will weaken surface heating, and vertical motion, intensify the westerly jet stream and eventually weaken the East Asian Summer Monsoon (Li et al., 2018; You et al., 2020). The snow cover change can also impact the magnitude, timing, and even sign of the South Asian Summer Monsoon and its precipitation (You et al., 2020).
+
+We stress that the far-reaching implications of reduced LAP pollution in climate change and the corresponding feedback mechanism need further analysis via coupled ESM experiments. We also urge more attentions on the future impacts of LAP on snow apart from climate change in Line 351-352 of the revised manuscript.
+
+References
+Biemans, H., Siderius, C., Lutz, A.F., Nepal, S., Ahmad, B., Hassan, T., von Bloh, W., Wijngaard, R.R., Wester, P., Shrestha, A.B. and Immerzeel, W.W., 2019. Importance of snow and glacier meltwater for agriculture on the Indo-Gangetic Plain. Nature Sustainability, 2(7), pp.594-601.
+Doherty, S.J., Dang, C., Hegg, D.A., Zhang, R. and Warren, S.G., 2014. Black carbon and other light-absorbing particles in snow of central North America. Journal of Geophysical Research: Atmospheres, 119(22), pp.12-807.
+Henderson, G.R., Peings, Y., Furtado, J.C. and Kushner, P.J., 2018. Snow–atmosphere coupling in the Northern Hemisphere. Nature Climate Change, 8(11), pp.954-963.
+Li, W., Guo, W., Qiu, B., Xue, Y., Hsu, P.C. and Wei, J., 2018. Influence of Tibetan Plateau snow cover on East Asian atmospheric circulation at medium-range time scales. Nature communications, 9(1), p.4243.
+Painter, T.H., Flanner, M.G., Kaser, G., Marzeion, B., VanCuren, R.A. and Abdalati, W., 2013. End of the Little Ice Age in the Alps forced by industrial black carbon. Proceedings of the national academy of sciences, 110(38), pp.15216-15221.
+You, Q., Wu, T., Shen, L., Pepin, N., Zhang, L., Jiang, Z., Wu, Z., Kang, S. and AghaKouchak, A., 2020. Review of snow cover variation over the Tibetan Plateau and its influence on the broad climate system. Earth-Science Reviews, 201, p.103043.
+Wang, T., Peng, S., Lin, X. and Chang, J., 2013. Declining snow cover may affect spring phenological trend on the Tibetan Plateau. Proceedings of the National Academy of Sciences, 110(31), pp.E2854-E2855.
+REVIEWER COMMENTS
+
+Reviewer #3 (Remarks to the Author):
+
+This paper investigated the future spatio-temporal characteristics of LAP mass, snow albedo reduction, and surface radiative forcing induced by BC and dust, and separated the relative contribution of future climate change and LAP evolution to snowpack changes. The paper provides valuable information for future snowpack loss mitigation and policy-maker. The authors have done substantial revision and I am satisfied with reply letter.
+
+The authors found that the projected LAP changes in snow cover over the Tibetan Plateau will alleviate future SWE loss due to climate change by 52.1±8.0% and 8.0±1.1% at the end of the century under SSP126 and SSP585, respectively, mainly due to reduced black carbon emission. Considering a large difference between two scenarios (green road and middle pathway), I think the significance of the work is for policy-makers that future green road pathway is great benefit for protection spring water supplies in Himalayan region and which is urgent to take an action. The authors should address this point in abstract and discussion section.
+
+BC and dust dataset collected from snowpits (or snow column) in the upper glacier (not those from summer aged snow in glacier surface which experience strong melt) could be also useful for comparison with simulated BC and dust. There are two recent works provide such data in the Tibetan Plateau.
+
+Yan F. et al., 2023. Dust dominates glacier darkening across majority of the Tibetan Plateau based on new measurements. Science of the Total Environment, 891, 164661. http://dx.doi.org/10.1016/j.scitotenv.2023.164661.
+
+Li Y., 2021. Black carbon and dust in the Third Pole glaciers: Reevaluated concentrations, mass absorption cross-sections and contributions to glacier ablation. Science of the Total Environment, 789: 147746. https://doi.org/10.1016/j.scitotenv.2021.147746.
+
+Several minor comments are as follow.
+
+1. In line 88-90 of the revised manuscript, “...using the state-of-the-art Energy Exascale Earth System Model (E3SM) Land Model (ELM) driven by meteorological and LAP changes simulated by a CMIP6 model (see Methods)”. What are the changes in meteorological field? Please clarify it.
+
+2. Line 100: Dust deposition increase under SSP126 and SSP585 in Himalaya and central Asia (Figure S2). Is this mainly due to anomaly atmospheric circulation or regional anthropogenic impacts in the future?
+3. In line 101-104, the statement of “...emissions sources ranging from the significant fossil fuel combustion, traditional biomass usage, to transport-related activity in the Asian regions” seems not appropriate, since the transport emission is also a kind of source sector of fossil fuel combustion.
+
+4. In line 203-205, the p values of the interannual variability of the average dust deposition are smaller than 0.05, which is inconsistent with those in and Fig. 1. Please check and revise it.
+
+5. In line 257-258, “… due to the reduced LAP deposition (Fig. 1 and S4) and snowpack under climate change…” should be “…due to the reduced BC deposition…”
+
+6. Line 346, I agree that the increased snow in spring may slow down the glacier retreat. One of reason is that spring heavier snow could protect summer glacier melt as indicated by Kang et al. (2009).
+
+7. Table S1, please add unit.
+
+8. The significance of trend of time series data was tested at the 95% confidence level (p<0.05). The p value was presented in Fig.1 only and should be added in other figures also.
+
+9. In the title of Fig.1, “… c,f,j,m their differences for (a-c,e-f) BC and (h-j,l-m) dust under SSP126 and SSP58” should be “…differences for (a-b, a-e) BC and (h-i, h-l) dust…”. Please check details carefully.
+
+10. “Consistent with Ref10, historically, dust plays a comparable role to BC in reducing snow albedo over high altitude regions of HMA (Fig. S7)” (line no. 244-245), but in Fig. S9, spatial patterns of historical snow albedo reduction caused by BC and dust are different obviously in this simulation.
+REVIEWER COMMENTS
+
+Reviewer #3 (Remarks to the Author):
+
+This paper investigated the future spatio-temporal characteristics of LAP mass, snow albedo reduction, and surface radiative forcing induced by BC and dust, and separated the relative contribution of future climate change and LAP evolution to snowpack changes. The paper provides valuable information for future snowpack loss mitigation and policy-maker. The authors have done substantial revision and I am satisfied with reply letter.
+
+Thanks for your constructive comments and suggestions. We have revised the manuscript carefully. Please see below for details.
+
+The authors found that the projected LAP changes in snow cover over the Tibetan Plateau will alleviate future SWE loss due to climate change by 52.1±8.0% and 8.0±1.1% at the end of the century under SSP126 and SSP585, respectively, mainly due to reduced black carbon emission. Considering a large difference between two scenarios (green road and middle pathway), I think the significance of the work is for policy-makers that future green road pathway is great benefit for protection spring water supplies in Himalayan region and which is urgent to take an action. The authors should address this point in abstract and discussion section.
+
+Thanks for the suggestions to improve the abstract and discussion section. Indeed, the contribution of future LAP changes can be very different under low and high emission scenarios. Cleaner snow can mitigate over half of the snowpack loss caused by climate change under the low emission scenario (i.e., SSP126). However, climate factors dominate snowpack loss under the worst emission scenario (i.e., SSP585). The change in the relative role of reduced LAP highlights the necessity of constraining global warming levels to mitigate snowpack loss. Compared to the high fossil-fuel development pathway (i.e., SSP585), a sustainable and green development pathway (i.e., SSP126) will alleviate future loss of water supply from snowmelt. We added more discussion on this in Line 258-265 of the revised manuscript.
+
+Besides, we have also added the corresponding description in the abstract “Our findings highlight a cleaner snow future and its benefits for future water supply from snowmelt especially under the sustainable development pathway of SSP126” of the revised manuscript.
+
+BC and dust dataset collected from snowpits (or snow column) in the upper glacier (not those from summer aged snow in glacier surface which experience strong melt) could be also useful for comparison with simulated BC and dust. There are two recent works provide such data in the Tibetan Plateau.
+Yan F. et al., 2023. Dust dominates glacier darkening across majority of the Tibetan Plateau based on new measurements. Science of the Total Environment, 891, 164661. http://dx.doi.org/10.1016/j.scitotenv.2023.164661.
+Li Y., 2021. Black carbon and dust in the Third Pole glaciers: Reevaluated concentrations, mass absorption cross-sections and contributions to glacier ablation. Science of the Total
+We appreciate your recommendation of the snowpit datasets for evaluating our simulations. We obtained the datasets from the two studies (illustrated in Figure R1a) and included them in Supplementary data 1. The model-observation comparison shows that overall, the ELM simulations are statistically in good agreements with the observations for the BC concentration especially in the snow column (Figure R1b,c). The correlation coefficients are 0.36 and 0.66, respectively, for BC concentration in the top snow layer and the snow column. About 78% of the simulated BC concentration in the top snow layer is within a factor of four of the observed concentrations, while that percentage is 83% for BC concentration in the snow column. Besides, the ELM simulated dust concentration in the snow column is well correlated to the snowpit measurements with a correlation coefficient of 0.74 (Figure R1d). About 74% of the simulated dust concentration in the snow column is within a factor of four of the observed concentrations. These results provide us with more confidence in using ELM to estimate the LAP darkening effects on snow. We have added these results in Line 416–433 and Supplementary Tet S1 of the revised manuscript.
+
+![Comparison of ELM-simulated and observed BC and dust concentration in snow across the NH. a. Spatial distribution of field snow samples. b,c Scatter plots between observed](page_256_613_1096_496.png)
+
+Figure R1| Comparison of ELM-simulated and observed BC and dust concentration in snow across the NH. a. Spatial distribution of field snow samples. b,c Scatter plots between observed
+and simulated BC concentration in the top snow layer and snow column. d. Scatter plots between observed and simulated dust concentration in the snow column. In (b,c,d), the dotted, dashed and solid lines are 1:1,f 1:4 (or 4:1) and 1:10 (or 10:1) ratio lines, and the correlation coefficient and p value are labeled.
+
+However, it should be noted that only the order-of-magnitude comparison is possible between climate model simulations and field measurements, considering that:
+
+1) Spatial mismatch: Our model simulations are at relatively coarse resolution (0.5 degree), while the field measurements are usually at a small point scale. Considering the strong dependence of snowpack on the local topography and microclimate conditions, the spatial representativeness of field measurements may be limited especially for summer when snow cover fraction is at its lowest and more heterogeneous. Such issue has been widely recognized by the community (Qian et al., 2011, 2015; Kang et al., 2020). High spatial resolution simulations and multi-site observation networks are needed to resolve this scale mismatch issue.
+
+2) Temporal mismatch: To utilize the snowpit data for model evaluation, we compared snowpit measurements over the TP available after 2014 (Li et al., 2021; Yan et al., 2023) with the climatological (2005-2014) average of the ELM historical simulations, since the ELM simulations after 2014 are driven by projected scenarios rather than the historically observed forcing. Furthermore, we compared the limited snowpit samples acquired during summer with the ELM simulated spring average values because the summer monthly outputs from ELM simulations show little snow cover during that period. Considering that the sensitivity of snowpack to local topography is not well resolved by our simulations at 0.5 degree resolution, and that the BC and dust in snow during spring may be representative of the same in summer, we are doing our best to make use of the point-scale snowpit data to evaluate the ELM simulations, focusing more on the spatial variations than the absolute values for the specific periods and seasons. As suggested by Qian et al. (2015), Yasunari et al. (2004) and Kang et al. (2020), more frequent snow impurity lifecycle measurements, e.g., at daily timescale are necessary for model validation in terms of different years and locations.
+
+3) Snow depth mismatch: The snow depth and sampling thickness of the measurements may vary case by case. It is challenging to accurately match the vertical location of LAP concentrations in the model-observation comparison, primarily due to the lack of the detailed measurement information.
+
+4) Uncertainties from field measurements: There are also some uncertainties in field measurements related to sample types, instrument errors, and measurement methods (Kang et al., 2020). Different studies measured snow samples with different surface snow types (e.g., fresh snow, aged snow, or granular ice) which can affect their inter-comparisons. BC concentrations in snow are primarily measured using laser-induced incandescence and thermal-optical methods. The former can determine the size distributions of BC particle, but its ability is limited by the instrument detection range (Lim et al., 2014). The accuracy of thermal-optical method can be affected by the presence of light-absorbing mineral dust in the snow samples (Li et al., 2017). As suggested by Kang et al. (2020), it is pressing to coordinate a LAP-in-snow measurement inter-comparison project to measure the same snow samples using different instruments/techniques,
+with the goal of providing an optimal estimate (with quantified uncertainty) of the LAP-in-snow concentrations.
+
+We have added more discussion on these uncertainties in model-observation intercomparison in Line 435-446 of the revised manuscript.
+
+Kang S, Zhang Y, Qian Y, Wang H. A review of black carbon in snow and ice and its impact on the cryosphere. Earth-Science Reviews 210, 103346 (2020).
+
+Li, C., et al. Re-evaluating black carbon in the Himalayas and the Tibetan Plateau: concentrations and deposition. Atmos. Chem. Phys. 17, 11899–11912 (2017).
+
+Li Y, et al. Black carbon and dust in the Third Pole glaciers: Reevaluated concentrations, mass absorption cross-sections and contributions to glacier ablation. Science of the Total Environment 789, 147746 (2021).
+
+Lim, S., et al. Refractory black carbon mass concentrations in snow and ice: method evaluation and inter-comparison with elemental carbon measurement. Atmospheric Measurement Techniques 7 3307-3324 (2014).
+
+Qian, Y., et al. Sensitivity studies on the impacts of Tibetan Plateau snowpack pollution on the Asian hydrological cycle and monsoon climate. Atmos. Chem. Phys. 11, 1929–1948 (2011).
+
+Qian Y, et al. Light-absorbing particles in snow and ice: Measurement and modeling of climatic and hydrological impact. Advances in Atmospheric Sciences 32, 64-91 (2015).
+
+Yan F, et al.. Dust dominates glacier darkening across majority of the Tibetan Plateau based on new measurements. Science of the Total Environment 891, 164661 (2023).
+
+Yasunari, Teppei J., et al. The GOddard SnoW impurity module (GOSWIM) for the NASA GEOS-5 earth system model: Preliminary comparisons with observations in Sapporo, Japan. Sola 10, 50-56 (2014).
+
+Several minor comments are as follow.
+1. In line 88-90 of the revised manuscript, “…using the state-of-the-art Energy Exascale Earth System Model (E3SM) Land Model (ELM) driven by meteorological and LAP changes simulated by a CMIP6 model (see Methods)”. What are the changes in meteorological field? Please clarify it.
+
+Sorry for the misleading expression in the original manuscript. We used meteorological forcing and LAP deposition data to drive the ELM simulations. We modified it as “using the state-of-the-art Energy Exascale Earth System Model (E3SM) Land Model (ELM) driven by meteorological forcing and LAP deposition data simulated by a CMIP6 model” in Line 63-65 of the revised manuscript.
+
+2. Line 100: Dust deposition increase under SSP126 and SSP585 in Himalaya and central Asia
+(Figure S2). Is this mainly due to anomaly atmospheric circulation or regional anthropogenic impacts in the future?
+
+Both climate change and human land use can contribute to the change of dust emission, transport, and deposition (Kok et al., 2023). For climate change, the future changes in soil moisture driven by precipitation, relative humidity, and surface wind have large impacts on the dust emission (Zhao et al., 2023). The change in atmospheric circulation and turbulent mixing can affect the long-range transportation of dust particles. The change in precipitation can also affect the wet deposition of dust. Future human land use change can contribute to the increase of dust emission by altering vegetation fraction (Tegen et al., 2004) and thus affect the dust deposition. We added the possible explanations in Line 101 and 267-269 of the revised manuscript. However, the complex attribution analysis on the dominant factors is beyond the scope of this study, which warrants further investigations.
+
+Kok JF, et al. Mineral dust aerosol impacts on global climate and climate change. Nature Reviews Earth & Environment 4, 71-86 (2023).
+
+Zhao Y, et al. Multi-model ensemble projection of global dust cycle by the end of 21st century using CMIP6 data. Atmos Chem Phys Discuss 2023, 1-28 (2023).
+
+Tegen I, Werner M, Harrison SP, Kohfeld KE. Relative importance of climate and land use in determining present and future global soil dust emission. Geophysical Research Letters 31, (2004).
+
+3. In line 101-104, the statement of “…emissions sources ranging from the significant fossil fuel combustion, traditional biomass usage, to transport-related activity in the Asian regions” seems not appropriate, since the transport emission is also a kind of source sector of fossil fuel combustion.
+
+We agree. We have revised it as “emissions sources ranging from the significant fossil fuel combustion and traditional biomass usage in the Asian regions” in Line 78 of the revised manuscript.
+
+4. In line 203-205, the p values of the interannual variability of the average dust deposition are smaller than 0.05, which is inconsistent with those in and Fig. 1. Please check and revise it.
+
+Thanks for catching this inconsistency. Here the p values should be larger than 0.05. We have modified this error as “and have a small, and insignificant increasing trend (MK test: p>0.05)” accordingly in 102-103 of the revised manuscript.
+
+5. In line 257-258, “… due to the reduced LAP deposition (Fig. 1 and S4) and snowpack under climate change…” should be “…due to the reduced BC deposition…”
+
+Thanks for pointing out this. We have modified it accordingly in Line 141-142 of the manuscript.
+6. Line 346, I agree that the increased snow in spring may slow down the glacier retreat. One of reason is that spring heavier snow could protect summer glacier melt as indicated by Kang et al. (2009).
+
+Good point! We have further explained the possible mechanisms in Line 350-352 of the revised manuscript: “The reduced snow loss may slow down the future glacier retreat, considering that spring heavy snow could suppress summer glacier melt56”.
+
+56Kang S, et al. Early onset of rainy season suppresses glacier melt: a case study on Zhadang glacier, Tibetan Plateau. Journal of Glaciology 55, 755-758 (2009).
+
+7. Table S1, please add unit.
+
+We have added the radiative forcing (RF) unit of (W m^{-2}) accordingly in Table S1 of the revised manuscript.
+
+8. The significance of trend of time series data was tested at the 95% confidence level (p<0.05). The p value was presented in Fig.1 only and should be added in other figures also.
+
+We have added the p values in all the figures related to the time-series analysis as well as the corresponding descriptions in the revised manuscript, except that in Figure S10, we introduced the p values in the figure caption because all the p-values of the temporal trends are smaller than 0.05.
+
+9. In the title of Fig.1, “… c,f,j,m their differences for (a-c,e-f) BC and (h-j,l-m) dust under SSP126 and SSP58” should be “…differences for (a-b, a-e) BC and (h-i, h-l) dust…”. Please check details carefully.
+
+To avoid the confusion, we have modified the figure caption as “their differences (calculated as Future - Historical) for BC and dust” in Line 111-112 of the revised manuscript.
+
+10. “Consistent with Ref10, historically, dust plays a comparable role to BC in reducing snow albedo over high altitude regions of HMA (Fig. S7)” (line no. 244-245), but in Fig. S9, spatial patterns of historical snow albedo reduction caused by BC and dust are different obviously in this simulation.
+
+Sorry for the misleading description. Indeed, our results show that the relative contributions of BC and dust to snow albedo reduction are regionally varied (Figs. S7 and S9). Sarangi et al. (2020) showed that the influence of dust on snow darkening is greater than that of BC at surface elevation above 4000 m over HMA. Our results also confirm the dominant role of dust over high-latitude regions of HMA. To make it clear, we have revised it as “Consistent with Ref10, historically, dust can play a greater role than BC in reducing snow albedo over high altitude regions of HMA (Fig. S7)” in Line 128-129 of the revised manuscript.
+Sarangi C, et al. Dust dominates high-altitude snow darkening and melt over high-mountain Asia. Nature Climate Change 10, 1045-1051 (2020).

0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc/preprint/preprint.md

@@ -0,0 +1,278 @@
+A cleaner snow future mitigates Northern Hemisphere snowpack loss from warming
+
+Dalei Hao ( dalei.hao@pnnl.gov )
+Pacific Northwest National Laboratory https://orcid.org/0000-0001-7154-6332
+Gautam Bisht
+Pacific Northwest National Laboratory
+Hailong Wang
+Pacific Northwest National Laboratory https://orcid.org/0000-0002-1994-4402
+Donghui Xu
+Pacific Northwest National Laboratory https://orcid.org/0000-0002-2859-2664
+Huilin Huang
+Pacific Northwest National Laboratory
+Yun Qian
+Pacific Northwest National Laboratory https://orcid.org/0000-0003-4821-1934
+L. Leung
+Pacific Northwest National Laboratory https://orcid.org/0000-0002-3221-9467
+
+Article
+
+Keywords:
+
+Posted Date: November 23rd, 2022
+
+DOI: https://doi.org/10.21203/rs.3.rs-2230945/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on October 2nd, 2023. See the published version at https://doi.org/10.1038/s41467-023-41732-6.
+A cleaner snow future mitigates Northern Hemisphere snowpack loss from warming
+
+Dalei Hao1*, Gautam Bisht1, Hailong Wang1, Donghui Xu1, Huilin Huang1, Yun Qian1 and L. Ruby Leung1*
+
+1Atmospheric Sciences and Global Change Division, Pacific Northwest National Laboratory, Richland, WA, USA
+
+Correspondence to: dalei.hao@pnnl.gov; ruby.leung@pnnl.gov
+
+Abstract
+Light-absorbing particles (LAP) such as black carbon and dust deposited on seasonal snowpack can result in snow darkening, earlier snowmelt, and regional climate change. However, the future deposition and surface radiative forcing of LAP in snow and their contributions to snowpack change remain unclear. Here, using Earth System Model simulations, we show significant reduction in black carbon deposition in the future. The reduced deposition decreases the December-May average LAP-induced radiative forcing in snow over the Northern Hemisphere from 1.3 W m\(^{-2}\) during 1995-2014 to 0.65 W m\(^{-2}\) and 0.49 W m\(^{-2}\) by 2081-2100 under the SSP126 and SSP585 scenarios, respectively. The reduced black carbon contamination in snow over the Tibetan Plateau will alleviate future snowpack loss due to climate change by 52.1±8.0% and 8.0±1.1% for the two scenarios. Our findings highlight a cleaner snow future and its benefits for future water availability from snowmelt.
+
+Main
+Seasonal snow plays a critical role in Earth’s energy budget and water cycle1, with snowmelt over mountainous regions providing an important source of freshwater for two billion people globally2,3. However, anthropogenic climate change is projected to reduce global snowfall and cause earlier snowmelt4. The high-mountain Asia (HMA), known as the water tower of Asia, has been experiencing an overall decrease in snow water equivalent (SWE) in the last 30 years5,6,7. The Western United States (WUS) is also expecting a low-to-no snow future, with a SWE decline of ~25% by 20508. Projections show a continuous decline in SWE for nearly all global high-mountain regions throughout the 21st century under both the low and high Representative Concentration Pathway (i.e., RCP2.6 and RCP8.5) greenhouse gas emission scenarios9. Future snowpack change under global warming will significantly alter downstream runoff as well as the amount and timing10 of freshwater supply from snowmelt for both humans and ecosystems.
+
+Light-absorbing particles (LAP), such as black carbon (BC) and dust, can darken snow surface, reduce snow albedo, and accelerate snowmelt processes11,12,13. BC originates from the incomplete burning of fossil fuels, biomass, or bio-fuels14, while dust can be produced from the natural wind erosion of soil or anthropogenic industrial emissions15. Snow darkening due to BC has significantly contributed to climate change and glacier retreats16. Dust was found to have larger darkening impacts than BC over HMA11. However, most climate models used in future snow projection neglect or oversimplify the effects of snow
+darkening due to BC and dust deposition9. A clean or dirty snow future could have a different impact on snowmelt and snow water resources. Future BC emissions are projected to decrease in both low and high emission scenarios, but the spatial and temporal patterns are expected to vary among different scenarios17. Under RCP8.5, most regions in the Northern Hemisphere (NH) are projected to show a significant decrease in BC deposition by 205018. The intensification of human land-use and drought induced by climate change will affect future dust emissions and depositions19. The projected 25 to 40% loss of biological soil crusts (i.e. biocrusts) will lead to around a 5-15% increase of global dust emission and deposition by 207020. Understanding the future changes in BC and dust depositions over snow-covered regions and how they affect snowpack is critical for constraining projections of downstream freshwater availability from snowmelt.
+
+Here we first use simulations from the Coupled Model Intercomparison Project Phase 6 (CMIP6) to analyze the evolution of BC and dust deposition from 2015 to 2100 over the NH. We explore two contrasting Shared Socioeconomic Pathway and Representative Concentration Pathway (SSP-RCP) scenarios (SSP126 and SSP585). We then examine the future spatio-temporal characteristics of LAP mass, snow albedo reduction, and surface radiative forcing (RF, a measure of the net change in surface radiative fluxes due to the change of a forcing agent) induced by BC and dust. We use the state-of-the-art Energy Exascale Earth System Model (E3SM) Land Model (ELM) driven by meteorological and LAP changes simulated by a CMIP6 model (see Methods). We also quantify the contributions of BC and dust evolution to future snowpack change to better understand their implications for future snow water resources. Our analyses mainly focus on December to May because the impacts of LAP on snow are generally larger in mid-latitude mountains than high-latitude regions13 and these regions have no or low snow cover from June to November
+
+Future evolution of BC and dust deposition
+
+Compared to the historical period (1995-2014), BC depositions over the entire snow-covered regions of the NH are projected to significantly decrease by 2081-2100 under both SSP126 and SSP585 (Fig. 1a-g). The historical BC deposition rates are higher in relatively lower latitude regions and show the largest values over the Southern border of the Tibetan Plateau (TP) and East Asia (Fig. 1a) because of the significant fossil fuel combustion, traditional biomass usage, and transport-related activity in the Asian regions17,21. The projected future BC deposition rates exhibit a similar spatial pattern as historical rates, but with significantly smaller magnitudes (Fig. 1b,e). Compared to the historical period, SSP126 shows larger BC deposition rate decreases than SSP585, especially in western Asia and eastern North America. There are significant decreasing trends in BC deposition from 2015 to 2100 over nearly all the snow-covered regions of the NH (Fig. 1c,f and Fig. S1a,b). The average BC deposition rates show a significant decreasing trend (p<0.05 based on the Mann-Kendall (MK) test) from 2015 to 2100, with higher decreasing rates from 2015-2040 (Fig. 1d). Compared to SSP126, SSP585 also shows a significant decreasing trend (p<0.05 from the MK test) but at a more uniform rate throughout the century (Fig. 1g). The relatively small standard deviations of the seven CMIP6 models that provide the deposition data for both BC and dust indicate similar decreasing trends across the models (shaded regions in Fig. 1d,g). The significant decrease of projected BC deposition can be attributed to socio-economic development and technological progress that have reduced BC emissions worldwide17.
+
+Compared to BC, overall future dust deposition is projected to be larger under both future scenarios from 2081-2100 than the historical period, especially over the TP (Fig. 1h-n). The spatial patterns of dust deposition rates are similar under the historical and future scenarios. The northern and western TP regions show large values because of their proximity to drylands, the main source of dust emission22. Most regions show no significant trends in dust deposition under SSP126, while many Asian regions show a
+significant increasing trend (MK test: p<0.05) under SSP585 (Fig. S1c,d). The average dust deposition rates over snow-covered NH regions shows large interannual variability and has a small increasing trend (MK test: p<0.05) from 2015 to 2100 under both scenarios (Fig. 1k,n). The large inter-model differences (shaded regions in Fig. 1k,n) can be attributed to different parameterizations of near-surface winds, soil erodibility, and/or vegetation evolution (prescribed vs dynamic vegetation) as well as diverse treatments of size of emitted dust particles in ESMs\(^{22,23}\). Despite the different and potentially opposite trends of future BC and dust depositions, the significant reduction of BC deposition is expected to have a larger effect on snow than dust changes.
+
+![Historical and future deposition rate of BC and dust. a,h Historical (1995-2014) and b,e,i,l future (2081-2100) spatial patterns of aerosol deposition rates and c,f,j,m their differences for (a-c,e-f) BC and (h-j,l-m) dust under SSP126 and SSP585. d,g,k,n Time series of the average deposition rate of BC and dust for snow-covered regions over the NH where the average SWE from December to May exceeds 5 mm. Historical and future deposition rates are calculated based on the ensemble mean of seven CMIP6 model outputs from December to May. In (a-c,e,f,h-j,l-m), grids with an average SWE from December to May smaller than 5 mm are masked. In (c,f,j,m), the black dots represent regions with statistically significant trends (p<0.05) using the MK test. In (d,g,k,n), the line and background shading represent the mean and standard deviation of deposition rates, respectively, based on the seven CMIP6 models. The p values from the MK test of statistical significance of the temporal trends from 2015-2100 are shown inside each panel; and the vertical dashed line indicates year 2015 when SSP scenarios start. We use ng m\(^{-2}\) s\(^{-1}\) and \( \mu \)g m\(^{-2}\) s\(^{-1}\) as BC and dust deposition rate units, respectively.](page_172_624_1207_627.png)
+
+Fig. 1| Historical and future deposition rate of BC and dust. a,h Historical (1995-2014) and b,e,i,l future (2081-2100) spatial patterns of aerosol deposition rates and c,f,j,m their differences for (a-c,e-f) BC and (h-j,l-m) dust under SSP126 and SSP585. d,g,k,n Time series of the average deposition rate of BC and dust for snow-covered regions over the NH where the average SWE from December to May exceeds 5 mm. Historical and future deposition rates are calculated based on the ensemble mean of seven CMIP6 model outputs from December to May. In (a-c,e,f,h-j,l-m), grids with an average SWE from December to May smaller than 5 mm are masked. In (c,f,j,m), the black dots represent regions with statistically significant trends (p<0.05) using the MK test. In (d,g,k,n), the line and background shading represent the mean and standard deviation of deposition rates, respectively, based on the seven CMIP6 models. The p values from the MK test of statistical significance of the temporal trends from 2015-2100 are shown inside each panel; and the vertical dashed line indicates year 2015 when SSP scenarios start. We use ng m\(^{-2}\) s\(^{-1}\) and \( \mu \)g m\(^{-2}\) s\(^{-1}\) as BC and dust deposition rate units, respectively.
+Future surface radiative forcings from BC and dust in snow
+
+We analyze the RF of BC, dust, and LAP (i.e., the sum of BC and dust) in snow simulated by ELM (see Methods). Northern Asia and TP have large RF of BC, dust, and LAP during the historical period (Fig. 2a,d,g). Due to reduced BC deposition, the projected BC RF over the whole NH decreases significantly by 2081-2100 (Fig. 2a-c) compared to the historical period (Fig. 1). However, the change of dust RF is sensitive to the emission scenarios, with an increase in SSP126 but a decrease in SSP585 (Fig. 2d-f). Due to the dominating change in BC RF, the LAP-induced RF significantly decreases over the NH under both scenarios (Fig. 2g-i; Table S1). Although BC has a larger RF than dust in the historical period, the significant reduction of BC deposition leads to a larger RF of dust than BC in the future (Table S1). These RF changes are consistent with the spatio-temporal distribution of LAP mass in the top snow layer (Fig. S2) and LAP-induced snow albedo reduction (Fig. S3).
+
+TP, a hotspot of climate change, is projected to experience a continuing decrease in LAP-induced RF from 2015-2100 (Fig. 3 and Fig. S4; Table S1). In the Control ELM simulations (see Methods), the historical (1995-2014) average LAP-induced RF over the TP is 5.1 W m^{-2}, while the future (2081-2100) average RF is 2.4 W m^{-2} under SSP126 (Fig. 3a). This is likely due to the reduced LAP deposition and snowpack under climate change. Compared to the historical period, assuming no change in future LAP deposition (see Methods) would result in a slightly decreasing average RF over the TP of 4.1 W m^{-2} during 2081-2100 under SSP126 due to the faster snowmelt and reduced snowpack in warmer temperatures. For SSP585, the average LAP-induced RF over the TP with and without future change of LAP deposition are 1.5 W m^{-2} and 2.3 W m^{-2}, respectively (Fig. 3b). Although ELM model configurations can affect the magnitude of simulated LAP-induced RF, their impacts on the relative difference caused by future LAP changes are small (Fig. 3a,b). Future LAP changes can account for 60.5±1.9% and 21.2±1.1% of the decrease of future LAP-induced RF relative to the historical period for SSP126 and SSP585, respectively. Temporally, the average BC-induced RF in the ELM simulations shows a significant decreasing trend (p<0.05 from the MK test) for both SSP126 and SSP585 scenarios. The trend is reduced under the assumption that future BC remains at the historical level (Fig. S4a). The dust-induced RF in SSP126 shows a slight increasing trend (p<0.05 from the MK test) that switches to a slight decreasing trend (p<0.05 from the MK test) if future dust deposition remains at the historical value (Fig. S4b). However, dust-induced RF shows a slight decreasing trend (p<0.05 from the MK test) in SSP585 that becomes stronger if future dust deposition remains unchanged (Fig. S4b). For both future scenarios, the LAP-induced RF has a significant decreasing trend that is larger than either individual trend from BC or dust (Fig. S4c). The significant reduction of future LAP-induced RF is expected to contribute to a slowdown of future snowpack loss.
+Fig. 2| Spatial patterns of historical and future surface radiative forcings (RF) of BC, dust and LAP (the sum of BC and dust) in snow covered regions over the NH. a,d,g Historical (1995-2014) spatial patterns of RF. b,c,e,f,h,i The difference between future (2081-2100) and historical RF under SSP126 and SSP585. Historical and future RF are calculated based on ELM outputs from December to May. In each panel, grids where the average SWE during December to May is smaller than 5 mm are masked.
+Fig 3| Historical and future average surface radiative forcing (RF) induced by LAP and SWE over the snow-covered TP regions (where the average SWE exceeds 5 mm in the historical period of 1995-2014) under different model configurations. a,b. The average RF during December to May under SSP126 and SSP585. c,d April SWE under SSP126 and SSP585. For each panel, Climate_hist+LAP_hist represents the historical (1995-2014) simulations with historical LAP depositions, while Climate_future+LAP_future and Climate_future+LAP_hist represent future (2081-2100) simulations with and without a future change of LAP depositions, respectively. The Climate_future+LAP_hist simulations used the historical average LAP depositions from 1995-2014. The horizontal axis labels represent different model configurations (see Methods), where Control has the ELM default settings and the others represent major adjustments made from the Control case. Specifically, PP assumes that the terrain is flat and neglects topographic effects on solar radiation; Koch assumes a non-spherical snow grain shape (Koch snowflake); extBC assumes external mixing between hydrophilic BC and snow grains; intDust assumes internal mixing between dust and snow grains; noLULCC has no land use and land cover change; MSE_high assumes high melt-water scavenging efficiency (MSE = 2, much higher than the default value of 0.2) of hydrophilic BC; and MSE_low assumes a low MSE (0.02) of hydrophilic BC. In (c,d), the contribution (\( \delta_{LAP} \)) of future LAP change that mitigates snowpack loss under each ELM configuration is noted as a percentage and is calculated as the ratio of the SWE difference (\( \Delta SWE_{LAP} \)) between Climate_future+LAP_future and Climate_future+LAP_hist to the SWE difference (\( \Delta SWE_{Climate} \)) between Climate_hist+LAP_hist and Climate_future+LAP_future. The geographical coverage of the TP is shown in Fig. 4.
+
+Contributions of LAP changes to future snowpack change
+
+We further quantify contributions of LAP changes to future snowpack changes using ELM (Fig. 4). Overall, the future SWE in April over the NH is projected to decrease compared to the historical period, especially under SSP585 (Fig. 4a-c and Fig. S5). WUS will have a significant decline in SWE under both low and high emission scenarios, identical to the findings of a low-to-no snow future noted in previous studies\(^{8,24}\). However, enhanced precipitation and sub-freezing temperatures will lead to an increase in SWE over the North American Arctic and high-latitude regions of Asia\(^{25}\). Consistent with previous findings\(^{26}\), TP is projected to have an overall decrease of seasonal snowpack throughout the 21st century. Due to its reliance on wintertime snowfall and initial cooler summertime temperatures, SWE in Karakoram, located in the northwestern TP, increases or stays relatively stable in the future\(^{27}\). We define
+the contribution of LAP change to snowpack change (\( \Delta \mathrm{SWE}_{\mathrm{LAP}} \)) as the difference between the projected SWE with and without future LAP change (see Methods). \( \Delta \mathrm{SWE}_{\mathrm{LAP}} \) is positive over the NH, especially under SSP126, and TP has the largest \( \Delta \mathrm{SWE}_{\mathrm{LAP}} \) (Fig. 4d-e). This highlights that future LAP change will contribute to a slowdown of warming-driven snowpack loss. The relative \( \Delta \mathrm{SWE}_{\mathrm{LAP}} \) (calculated as the ratio of \( \Delta \mathrm{SWE}_{\mathrm{LAP}} \) to projected SWE without LAP change) over the TP significantly increases (MK test: p<0.05) from 0 to over 10% (SSP126) and 20% (SSP585) during 2015-2100 (Fig. 4f-g). The future reduction of BC concentration dominates the change of relative \( \Delta \mathrm{SWE}_{\mathrm{LAP}} \) (Fig. S4f-g), accounting for most of the slowdown of SWE reduction.
+
+We further analyze the relative contribution of climate change (i.e., temperature and precipitation) and LAP change as well as the impacts of ELM model configurations (Fig. 3c-d and Fig. S6) on SWE. Compared to the historical period, SSP585 shows a stronger decrease of SWE than SSP126 regardless of model configuration (Fig. 3c,d). Although topography, snow grain shape, mixing states of LAP-snow, land use and land cover changes, and melt-water scavenging efficiency can affect the magnitude of SWE, they have small impacts on the relative change of SWE (Fig. 3c,d). For example, the larger snow albedo of non-spherical snow grain shape can lead to larger SWE\(^{28}\), but snow grain shape has little impact on the relative change of SWE. We further estimate climate change impacts (\( \Delta \mathrm{SWE}_{\mathrm{Climate}} \)) on SWE as the difference between the historical SWE and future SWE without LAP change. We define the contribution (\( \delta_{\mathrm{LAP}} \)) of future LAP change that mitigates snowpack loss as the ratio of \( \Delta \mathrm{SWE}_{\mathrm{LAP}} \) to \( \Delta \mathrm{SWE}_{\mathrm{Climate}} \) (see Methods). The spatial distribution of \( \delta_{\mathrm{LAP}} \) shows that TP is the most sensitive to LAP change (Fig. S6). Under different model configurations, \( \delta_{\mathrm{LAP}} \) of TP is 52.1±8.0% (SSP126) and 8.0±1.1% (SSP585) (Fig. 3c,d). These results suggest that under SSP585, climate change dominates the snowpack loss and only 8% of this loss can be mitigated by LAP change, while under SSP126, LAP change can offset around 52.1% of the impacts of climate change on the TP snowpack.
+Fig. 4| Future SWE in April and contributions of LAP change to future SWE. a Historical (1995-2014) and b,c future (2081-2100) spatial patterns of SWE in April under SSP126 and SSP585. d,e The differences (\( \Delta \mathrm{SWE}_{\mathrm{LAP}} \)) of future (2081-2100) April SWEs with and without LAP change under SSP126 and SSP585. f,g Time series of relative \( \Delta \mathrm{SWE}_{\mathrm{LAP}} \) (calculated as the ratio of \( \Delta \mathrm{SWE}_{\mathrm{LAP}} \) to projected SWE without LAP change) over snow-covered regions (where the average SWE exceeds 5 mm in the historical period) in the TP. In (a-e), grids with an average April SWE smaller than 5 mm in the historical period are masked. In (f,g), BC_{future}+Dust_{future}, BC_{future}+Dust_{hist}, and BC_{hist}+Dust_{future} represent different combinations of BC and dust depositions, where the subscripts of future and hist represent future and historical average depositions, respectively.
+
+Discussion
+
+Research has shown that LAP contribute to the melting of snow and the retreat of glacier^{21,29}. We show decreasing trends of LAP deposition, mass, and RF under both low (SSP126) and high (SSP585) emission scenarios using simulation results from CMIP6 and ELM. Our study highlights a cleaner snow future due
+to the significant reduction of BC deposition on snow surface. The projected cleaner snow will help alleviate future snowpack loss induced by warmer climates, especially over the TP. This is consistent with a recent study30 showing that BC deposition decrease since the 1980s has moderated the influence of climate change in the decline of snow cover over the French Alps and the Pyrenees. The compensating effects of decreasing BC will also contribute to the shift in snowmelt timing and substantially influence melt water runoff30. Considering the important role of the TP in Asia’s freshwater supply26, the increased SWE due to cleaner snow will be beneficial for the municipal, hydropower31 and agricultural32 sectors in Asian regions as well as vegetation growth33 and animal survival34. Cleaner snow can mitigate over half of the snowpack loss caused by climate change under the optimized emission scenario (i.e., SSP126). These results stress the importance of reducing combustion aerosol emissions by developing clean, renewable energy and negative-emission technologies, in addition to mitigating climate change.
+However, climate factors dominate snowpack loss under the worst emission scenario (i.e., SSP585). The change in the relative role of reduced LAP highlights the necessity of constraining global warming levels to mitigate snowpack loss.
+
+Beyond the significant decrease of RF of BC over the TP, the RF of dust will increase under SSP126. This is potentially due to increasing drought frequency, duration, and intensity linked to climate change as well as continued land use and land cover change35. The dust RF is projected to have a small decrease under SSP585 (Fig. 2). Consequently, dust is projected to account for a larger portion of LAP-induced RF than BC in the future under both SSP126 and SSP585, while BC accounts for more in the historical period (Fig. 2 and Fig. S4; Table S1). The increasing contributions of dust to snowmelt will be more significant over the high-altitude HMA, as previous work demonstrated that dust dominates high-altitude snow darkening and melt over HMA11. These highlight the importance of mitigating soil disturbance and stabilizing soil surface in dust source regions to reduce both natural and anthropogenic dust emissions36.
+
+However, there remain uncertainties in simulating the LAP darkening effects on snow. First, accurately characterizing the emission, transport, chemistry, and deposition of LAP under changing climate is challenging. The emission, cycling, and persistence of BC are still under-represented in ESMs37. Although reduced fossil fuel burning in developing countries can decrease future BC emissions, increased wildfire intensity and frequency due to climate change and land-use change may potentially increase the BC emissions38, 39. However, ESMs still have large uncertainties when representing fire ignition, suppression, spread, and particularly emission, given our incomplete understanding of the complex and interacting controls on wildfire activities40, 41. The uncertainties of wildfire simulations are expected to have only minor impacts on our results, as we focus on the snow season from December to May when wildfire activities are less frequent42. Although different ESMs show relatively similar magnitudes and spatio-temporal patterns of BC deposition (Fig. 1d,g), the uncertainties associated with the emission inventory, transport and deposition modeling in ESMs should be quantified and reduced43. Furthermore, while ESMs reproduce the global spatial patterns and seasonal variations of dust distribution, there remain large differences in deposition rates (Fig. 1k,n) among ESMs due to uncertainties in simulating the dust life cycle (i.e., emission, transport, deposition). Such inconsistencies have been widely reported and are attributed to uncertainties of simulated land surface properties and atmospheric states22, 23, 44, 45, 46. Prior work shows that ESMs underestimate the amount of coarse dust with diameter \( \geq 5 \mu m \)47. Biological soil crusts (biocrusts), which are rarely considered in ESMs, have been found to have large effects on regional and global dust cycling20. These limitations could lead to an underestimation of dust impurity effects in both historical period and future scenarios. However, our analyses are primarily based on relative differences rather than absolute values, mitigating the impact of the previously mentioned uncertainties.
+Advanced remote sensing observations, e.g., NASA’s Earth Surface Mineral Dust Source Investigation (EMIT), are promising for providing potential constraints of the sign and magnitude for projecting dust deposition under climate scenarios48.
+
+Second, representing the evolutions of LAP and their mixing states with snow affects estimates of the LAP darkening effect on snow. The Snow, Ice, and Aerosol Radiative (SNICAR) model used in ELM (see Methods) can model the snow darkening effects of BC and dust well and has been widely used in snow-related studies49. The Control ELM simulations assumed spherical snow grain shape, internal mixing of hydrophilic BC-snow, and external mixing of dust-snow (see Methods). Although non-spherical snow grain shape and the mixing state of LAP-snow can affect the magnitude of LAP-induced RF, their impacts on the relative contribution of future LAP change to RF is within ±2% (Fig. 3). More observations are needed to better constrain the irregular snow grain shape and space- and time-varying mixing states of LAP-snow in ESMs28. Although LAP scavenging processes via melting water can regulate LAP concentration after deposition, using either low or high melt-water scavenging efficiencies for hydrophilic BC leads to similar results in our simulations (Fig. 3). Apart from BC and dust, brown carbon also has large darkening effects on snow50. However, it is not represented in most ESMs due to the large variations and high uncertainties of its chemical composition and optical properties. Snow algae also play important roles in snow melt and glacier retreat51, but nearly all ESMs neglect snow algae effects on snow. Although snow algae blooming and distributions have been successfully implemented in the Minimal Advanced Treatments of Surface Interaction and Runoff land surface model (MATSIRO)52, more observations and modeling are needed to evaluate and improve MATSIRO performance before further applications. Better constraining the projections LAP impacts will benefit from the long-term field measurements and hyperspectral remote sensing satellite missions, e.g., the Surface Biology and Geology mission led by NASA53.
+
+Third, projecting future SWE change is sensitive to meteorological forcings and model configurations. Precipitation and temperature largely determine the snowfall and snowmelt rates, producing large effects on snowpack4. However, temperature and precipitation of ESM simulations still have systematic biases54. The Community Earth System Model Version 2 (CESM2) forcing data used in this study shows an overall consistent trend of future precipitation and air temperature projections over the TP with the ensemble mean of the used CMIP6 models (Fig. S7a-d). Snowpack simulations are affected by the representations of various snow processes55. For example, topography affects the solar radiation received at the surface and snow processes56, while land use and land cover change can also affect the snow accumulation and melting processes57. Despite their importance, both of these factors have small impacts on our analysis (Fig. 3). Although model uncertainties may influence SWE projections, our sensitivity experiments with different ELM model configurations show only small impacts on the relative contributions of LAP. Our previous study showed that ELM can well capture the snow distribution in the TP, compared to the MODIS remote sensing data, supporting the reliability of the results28. ELM simulations can reproduce the spatio-temporal pattern, interannual variability and elevation gradient for different snow properties over the WUS58. The simulations are in line with Snow Telemetry field measurements, MODIS remote sensing products, and data assimilation products. The projected SWE trends of ELM simulations under both SSP126 and SSP585 are consistent with the ensemble mean of the seven CMIP6 models (Fig. S7). These model sensitivity experiments, model evaluations, and intercomparisons provide confidence in simulating future SWE change using ELM.
+With a focus on the future evolution of LAP deposition and RF over the NH, we identify a cleaner snow future and reduced snowpack loss, especially over the TP. The broader implications of reduced LAP pollution in climate change need further analysis via coupled ESM experiments.
+Methods
+
+CMIP6 simulations
+
+Monthly aerosol deposition data during 1995-2100, including deposition rates of BC and dust from seven ESMs participating in CMIP6 (Table S2), are used in the study. We select the CMIP6 models that provide data for both BC and dust deposition with the variant of "r1i1p1f1". Two future scenarios of SSP126 and SSP585 are included in the analysis. All the data are remapped to a spatial resolution of 0.94° × 1.25° in latitude and longitude and are aggregated to the annual scale by calculating average values during December to May.
+
+Only CESM2 provides aerosol deposition data with different BC and dust categories (i.e., hydrophilic BC, hydrophobic BC, dust particle size). The other six models just provided the total deposition rates of BC and dust. Therefore, we used the historical and future atmospheric CO₂ concentration, meteorological forcing data (e.g., precipitation, air temperature, humidity, wind speed, downward solar radiation, and longwave radiation), and aerosol deposition data from CESM2 to drive the ELM to further investigate future trends of LAP mass in snow.
+
+E3SM Land Model
+
+The E3SM is a state-of-the-art fully-coupled ESM supported by the U.S. Department of Energy, which aims to improve and enhance actionable predictions of Earth system variability and change59. ELM, the land component of E3SM, originated from the Community Land Model Version 4.5 (CLM4.5)60. ELM can mechanistically simulate snow processes from snow accumulation to snow evolution to snow melt based on a multi-layer scheme. ELM can also prognostically simulate the change of LAP concentrations at different snow layers after deposition and how these changes affect snow albedo. Specifically, ELM uses a hybrid mode of SNICAR and the delta-Eddington adding-doubling radiative transfer solver to calculate the shortwave radiative characteristics of snow at different spectral bands61. The new SNICAR scheme in ELM treats both external mixing and internal mixing (within-hydrometeor) of BC and snow grains as well as size-dependent BC optical properties62. We recently extended ELM’s ability to consider both external mixing and internal mixing of between dust and snow grains as well as the impact of non-spherical snow grain shape on snow albedo28. In addition, we implemented a new parameterization that considers sub-grid topographic effects on solar radiation56. The model enhancements allow ELM to be used to investigate uncertainties related to model configurations on future snow projections. Our previous study showed that ELM can well capture the spatio-temporal distribution and interannual variability of snow properties and timing compared to field measurements, remote sensing observations, and data assimilated products28, 58. These confirm the effectiveness of simulating snow processes in ELM.
+
+Simulating the mass and RF of LAP in snow
+
+To analyze future trends of the mass and RF of LAP, we conducted a series of offline ELM simulations at 0.5° spatial resolution over the NH from 1950 to 2100. The simulations were driven by CESM historical and future (SSP126 and SSP585) meteorological forcing and aerosol deposition data. We used the prescribed satellite phenology mode and downscaled the forcing data temporally and spatially to half-hourly and 0.5° resolution with bilinear interpolation methods. For each simulation, ELM was run at a half-hour time step with a monthly output frequency. The first 45 years (i.e., 1950-1994) were treated as model spin-up time and the remaining 106 years (i.e. 1995-2100) are used in the analysis.
+To quantify the contribution of future LAP changes, we conducted a suite of simulations from 2015 to 2100 under the SSP126 and SSP585 scenarios with separate configurations of: 1) monthly LAP deposition averaged in the historical period (1995-2014) (i.e., without interannual variation and trends in LAP); 2) future projected LAP deposition (i.e., with interannual variation and trends in LAP); 3) historical average BC deposition and future projected dust deposition; and 4) future projected BC deposition and historical average dust deposition. These simulations were carried out with the default ELM model settings, except that the simulations considered topographic effects on solar radiation.
+
+To quantify uncertainties associated with model configurations, we also conducted ELM simulations with different model configurations related to snow grain shape, mixing state of LAP-snow, land use and land cover change, topographic effects on solar radiation, and melt-water scavenging efficiency of hydrophilic BC (Table S3).
+
+Quantifying contributions of LAP changes to future snowpack loss
+
+Based on the ELM simulations, the difference (\( \Delta SWE_{Climate} \)) between the historical (1995-2014) SWE and projected future SWE (2081-2100) without LAP changes is used to characterize the impact of climate change (i.e., temperature and precipitation) (see equation 1). The SWE difference (\( \Delta SWE_{LAP} \)) between future SWE with and without LAP changes is used to represent the impact of future LAP changes (see equation 2). We further define the ratio of \( \Delta SWE_{LAP} \) to \( \Delta SWE_{Climate} \) as the relative contribution of future LAP changes to the slowdown in snowpack loss (\( \delta_{LAP} \)), as shown in equation 3.
+
+\[
+\Delta SWE_{Climate} = SWE_{Climate_{his}} + LAP_{hist} - SWE_{Climate_{future}+LAP_{hist}}
+\] (1)
+
+\[
+\Delta SWE_{LAP} = SWE_{Climate_{future}+LAP_{future}} - SWE_{Climate_{future}+LA_{hist}}
+\] (2)
+
+\[
+\delta_{LAP} = \frac{\Delta SWE_{LAP}}{\Delta SWE_{Climate}}
+\] (3)
+
+The mean and standard deviation of \( \delta_{LAP} \) under different ELM configurations are used to represent the mean LAP effect and the corresponding uncertainty.
+
+Statistical analysis
+
+All analyses are conducted using MATLAB R2019b (MathWorks Inc.). We use the non-parametric Mann-Kendall (MK) Tau test and Sen’s slope to detect the monotonic trend of time series data. Trends with p<0.05 are considered to be statistically significant in this study. Specifically, we use the ‘ktau’ function63 in this study.
+
+Data availability
+
+Except for the CESM2 data, the aerosol deposition data in CMIP6 simulations can be freely downloaded from https://esgf-node.llnl.gov/search/cmip6/. All CESM2 data can be acquired using NCAR’s data sharing service. The ELM outputs from this study are openly available at https://github.com/daleihao/snow_SSP.
+
+Code availability
+
+The ELM code is publicly available at https://github.com/daleihao/E3SM. Codes to generate all results and plot all figures are available at https://github.com/daleihao/snow_SSP.
+Acknowledgements
+
+This research has been supported by the U.S. Department of Energy (DOE), Office of Science, Biological and Environmental Research (BER) program, Earth System Model Development program area, as part of the Climate Process Team projects. H.W. acknowledges support by the Regional and Global Model Analysis program area, as part of the HiLAT project. This research was conducted at Pacific Northwest National Laboratory (PNNL), which is operated for the U.S. DOE by Battelle Memorial Institute under contract DEAC05-76RL01830. This research used resources from Cori and Perlmutter supercomputers of the National Energy Research Scientific Computing Center (NERSC), a User Facility supported by the Office of Science of the U.S. DOE under contract no. DE-AC02-15 05CH11231. The reported research also used DOE BER Earth and Environmental Systems Modeling program’s Compy computing cluster located at PNNL. We thank Beth Mundy and Ben Bond-Lamberty for their valuable suggestions.
+
+Author contributions
+
+D.H. and G.B. conceived the study. D.H. processed the data, conducted the simulations, performed the analysis and drafted the original manuscript. All authors made suggestions to the design of the study and the analysis of the results, and contributed to improving the manuscript.
+
+Competing interests
+
+The authors declare no competing interests.
+
+References
+
+1. Flanner MG, Shell KM, Barlage M, Perovich DK, Tschudi MA. Radiative forcing and albedo feedback from the Northern Hemisphere cryosphere between 1979 and 2008. Nature Geoscience 2011, 4(3): 151-155.
+
+2. Biemans H, Siderius C, Lutz AF, Nepal S, Ahmad B, Hassan T, et al. Importance of snow and glacier meltwater for agriculture on the Indo-Gangetic Plain. Nature Sustainability 2019, 2(7): 594-601.
+
+3. Simpkins G. Snow-related water woes. Nature Climate Change 2018, 8(11): 945-945.
+
+4. Barnett TP, Adam JC, Lettenmaier DP. Potential impacts of a warming climate on water availability in snow-dominated regions. Nature 2005, 438(7066): 303-309.
+
+5. Smith T, Bookhagen B. Changes in seasonal snow water equivalent distribution in High Mountain Asia (1987 to 2009). Science Advances 2018, 4(1): e1701550.
+
+6. Bormann KJ, Brown RD, Derksen C, Painter TH. Estimating snow-cover trends from space. Nature Climate Change 2018, 8(11): 924-928.
+7. Kraaijenbrink PDA, Stigter EE, Yao T, Immerzeel WW. Climate change decisive for Asia’s snow meltwater supply. Nature Climate Change 2021, **11**(7): 591-597.
+
+8. Siirila-Woodburn ER, Rhoades AM, Hatchett BJ, Huning LS, Szinai J, Tague C, *et al.* A low-to-no snow future and its impacts on water resources in the western United States. Nature Reviews Earth & Environment 2021, **2**(11): 800-819.
+
+9. Hock R, Rasul G, Adler C, Cáceres B, Gruber S, Hirabayashi Y, *et al.* High Mountain Areas. InIPCC Special Report on the Ocean and Cryosphere in a Changing Climate; Pörtner, H.-O., Roberts, DC. Masson-Delmotte, V., Zhai, P., Tignor, M., Poloczanska, E., Mintenbeck, K …. 
+
+10. Xu D, Ivanov VY, Li X, Troy TJ. Peak Runoff Timing Is Linked to Global Warming Trajectories. Earth's Future 2021, **9**(8): e2021EF002083.
+
+11. Sarangi C, Qian Y, Rittger K, Ruby Leung L, Chand D, Bormann KJ, *et al.* Dust dominates high-altitude snow darkening and melt over high-mountain Asia. Nature Climate Change 2020, **10**(11): 1045-1051.
+
+12. Hadley OL, Kirchstetter TW. Black-carbon reduction of snow albedo. Nature Climate Change 2012, **2**(6): 437-440.
+
+13. Skiles SM, Flanner M, Cook JM, Dumont M, Painter TH. Radiative forcing by light-absorbing particles in snow. Nature Climate Change 2018, **8**(11): 964-971.
+
+14. Bond TC, Doherty SJ, Fahey DW, Forster PM, Berntsen T, DeAngelo BJ, *et al.* Bounding the role of black carbon in the climate system: A scientific assessment. Journal of Geophysical Research: Atmospheres 2013, **118**(11): 5380-5552.
+
+15. Tanaka TY, Chiba M. A numerical study of the contributions of dust source regions to the global dust budget. Global and Planetary Change 2006, **52**(1): 88-104.
+
+16. Hansen J, Nazarenko L. Soot climate forcing via snow and ice albedos. Proceedings of the National Academy of Sciences 2004, **101**(2): 423-428.
+
+17. Gidden MJ, Riahi K, Smith SJ, Fujimori S, Luderer G, Kriegler E, *et al.* Global emissions pathways under different socioeconomic scenarios for use in CMIP6: a dataset of harmonized emissions trajectories through the end of the century. Geosci Model Dev 2019, **12**(4): 1443-1475.
+
+18. Ménégoz M, Krinner G, Balkanski Y, Cozic A, Boucher O, Ciais P. Boreal and temperate snow cover variations induced by black carbon emissions in the middle of the 21st century. The Cryosphere 2013, **7**(2): 537-554.
+19. Neff JC, Ballantyne AP, Farmer GL, Mahowald NM, Conroy JL, Landry CC, et al. Increasing eolian dust deposition in the western United States linked to human activity. Nature Geoscience 2008, **1**(3): 189-195.
+
+20. Rodriguez-Caballero E, Stanelle T, Egerer S, Cheng Y, Su H, Canton Y, et al. Global cycling and climate effects of aeolian dust controlled by biological soil crusts. Nature Geoscience 2022.
+
+21. Li C, Bosch C, Kang S, Andersson A, Chen P, Zhang Q, et al. Sources of black carbon to the Himalayan–Tibetan Plateau glaciers. Nature Communications 2016, **7**(1): 12574.
+
+22. Zhao A, Ryder CL, Wilcox LJ. How well do the CMIP6 models simulate dust aerosols? Atmos Chem Phys 2022, **22**(3): 2095-2119.
+
+23. Evan AT, Flamant C, Fiedler S, Doherty O. An analysis of aeolian dust in climate models. Geophysical Research Letters 2014, **41**(16): 5996-6001.
+
+24. Fyfe JC, Derksen C, Mudryk L, Flato GM, Santer BD, Swart NC, et al. Large near-term projected snowpack loss over the western United States. Nature Communications 2017, **8**(1): 14996.
+
+25. Krasting JP, Broccoli AJ, Dixon KW, Lanzante JR. Future Changes in Northern Hemisphere Snowfall. Journal of Climate 2013, **26**(20): 7813-7828.
+
+26. Yao T, Thompson L, Chen D, Chettri N. *A Scientific Assessment of the Third Pole Environment*, 2022.
+
+27. Kapnick SB, Delworth TL, Ashfaq M, Malyshev S, Milly PCD. Snowfall less sensitive to warming in Karakoram than in Himalayas due to a unique seasonal cycle. Nature Geoscience 2014, **7**(11): 834-840.
+
+28. Hao D, Bisht G, He C, Bair E, Huang H, Dang C, et al. Improving snow albedo modeling in E3SM land model (version 2.0) and assessing its impacts on snow and surface fluxes over the Tibetan Plateau. Geosci Model Dev Discuss 2022, **2022**: 1-31.
+
+29. Ramanathan V, Carmichael G. Global and regional climate changes due to black carbon. Nature Geoscience 2008, **1**(4): 221-227.
+
+30. Réveillet M, Dumont M, Gascoin S, Lafaysse M, Nabat P, Ribes A, et al. Black carbon and dust alter the response of mountain snow cover under climate change. Nature Communications 2022, **13**(1): 5279.
+
+31. Li D, Lu X, Walling DE, Zhang T, Steiner JF, Wasson RJ, et al. High Mountain Asia hydropower systems threatened by climate-driven landscape instability. Nature Geoscience 2022, **15**(7): 520-530.
+32. Qin Y, Abatzoglou JT, Siebert S, Huning LS, AghaKouchak A, Mankin JS, et al. Agricultural risks from changing snowmelt. Nature Climate Change 2020, **10**(5): 459-465.
+
+33. Shen M, Piao S, Dorji T, Liu Q, Cong N, Chen X, et al. Plant phenological responses to climate change on the Tibetan Plateau: research status and challenges. National Science Review 2015, **2**(4): 454-467.
+
+34. Luo Z, Jiang Z, Tang S. Impacts of climate change on distributions and diversity of ungulates on the Tibetan Plateau. Ecological Applications 2015, **25**(1): 24-38.
+
+35. Kok JF, Storelvmo T, Karydis V, Adebiyi AA, Mahowald NM, Evan A, et al. The impacts of mineral dust aerosols on global climate and climate change. 2022.
+
+36. Painter TH, Deems JS, Belnap J, Hamlet AF, Landry CC, Udall B. Response of Colorado River runoff to dust radiative forcing in snow. Proceedings of the National Academy of Sciences 2010, **107**(40): 17125-17130.
+
+37. Coppola AI, Wagner S, Lennartz ST, Seidel M, Ward ND, Dittmar T, et al. The black carbon cycle and its role in the Earth system. Nature Reviews Earth & Environment 2022.
+
+38. Sullivan A, Baker E, Kurvits T. Spreading Like Wildfire: The Rising Threat of Extraordinary Landscape Fires. 2022.
+
+39. Jones MW, Abatzoglou JT, Veraverbeke S, Andela N, Lasslop G, Forkel M, et al. Global and Regional Trends and Drivers of Fire Under Climate Change. Reviews of Geophysics 2022, **60**(3): e2020RG000726.
+
+40. Kloster S, Lasslop G. Historical and future fire occurrence (1850 to 2100) simulated in CMIP5 Earth System Models. Global and Planetary Change 2017, **150**: 58-69.
+
+41. Hantson S, Arneth A, Harrison SP, Kelley DI, Prentice IC, Rabin SS, et al. The status and challenge of global fire modelling. Biogeosciences 2016, **13**(11): 3359-3375.
+
+42. Giglio L, Randerson JT, van der Werf GR. Analysis of daily, monthly, and annual burned area using the fourth-generation global fire emissions database (GFED4). Journal of Geophysical Research: Biogeosciences 2013, **118**(1): 317-328.
+
+43. Kang S, Zhang Y, Qian Y, Wang H. A review of black carbon in snow and ice and its impact on the cryosphere. Earth-Science Reviews 2020, **210**: 103346.
+
+44. Huneeus N, Schulz M, Balkanski Y, Griesfeller J, Prospero J, Kinne S, et al. Global dust model intercomparison in AeroCom phase I. Atmos Chem Phys 2011, **11**(15): 7781-7816.
+45. Wu C, Lin Z, Liu X. The global dust cycle and uncertainty in CMIP5 (Coupled Model Intercomparison Project phase 5) models. Atmos Chem Phys 2020, 20(17): 10401-10425.
+
+46. Aryal YN, Evans S. Global Dust Variability Explained by Drought Sensitivity in CMIP6 Models. Journal of Geophysical Research: Earth Surface 2021, 126(6): e2021JF006073.
+
+47. Adebiyi AA, Kok JF. Climate models miss most of the coarse dust in the atmosphere. Science Advances 2020, 6(15): eaaz9507.
+
+48. Green RO, Mahowald N, Ung C, Thompson DR, Bator L, Bennet M, et al. The Earth Surface Mineral Dust Source Investigation: An Earth Science Imaging Spectroscopy Mission. 2020 IEEE Aerospace Conference; 2020 7-14 March 2020; 2020. p. 1-15.
+
+49. Flanner MG, Arnheim JB, Cook JM, Dang C, He C, Huang X, et al. SNICAR-ADv3: a community tool for modeling spectral snow albedo. Geosci Model Dev 2021, 14(12): 7673-7704.
+
+50. Brown H, Wang H, Flanner M, Liu X, Singh B, Zhang R, et al. Brown Carbon Fuel and Emission Source Attributions to Global Snow Darkening Effect. Journal of Advances in Modeling Earth Systems 2022, 14(4): e2021MS002768.
+
+51. Ganey GQ, Loso MG, Burgess AB, Dial RJ. The role of microbes in snowmelt and radiative forcing on an Alaskan icefield. Nature Geoscience 2017, 10(10): 754-759.
+
+52. Onuma Y, Yoshimura K, Takeuchi N. Global Simulation of Snow Algal Blooming by Coupling a Land Surface and Newly Developed Snow Algae Models. Journal of Geophysical Research: Biogeosciences 2022, 127(2): e2021JG006339.
+
+53. Cawse-Nicholson K, Townsend PA, Schimel D, Assiri AM, Blake PL, Buongiorno MF, et al. NASA's surface biology and geology designated observable: A perspective on surface imaging algorithms. Remote Sensing of Environment 2021, 257: 112349.
+
+54. Chai Y, Yue Y, Slater LJ, Yin J, Borthwick AGL, Chen T, et al. Constrained CMIP6 projections indicate less warming and a slower increase in water availability across Asia. Nature Communications 2022, 13(1): 4124.
+
+55. Krinner G, Derksen C, Essery R, Flanner M, Hagemann S, Clark M, et al. ESM-SnowMIP: assessing snow models and quantifying snow-related climate feedbacks. Geosci Model Dev 2018, 11(12): 5027-5049.
+
+56. Hao D, Bisht G, Gu Y, Lee WL, Liou KN, Leung LR. A parameterization of sub-grid topographical effects on solar radiation in the E3SM Land Model (version 1.0); implementation and evaluation over the Tibetan Plateau. Geosci Model Dev 2021, 14(10): 6273-6289.
+57. Varhola A, Coops NC, Weiler M, Moore RD. Forest canopy effects on snow accumulation and ablation: An integrative review of empirical results. Journal of Hydrology 2010, **392**(3): 219-233.
+
+58. Hao D, Bisht G, Rittger K, Stillinger T, Bair E, Gu Y, *et al*. Evaluation of snow processes over the Western United States in E3SM land model. *EGUsphere* 2022, **2022**: 1-38.
+
+59. Leung LR, Bader DC, Taylor MA, McCoy RB. An Introduction to the E3SM Special Collection: Goals, Science Drivers, Development, and Analysis. *Journal of Advances in Modeling Earth Systems* 2020, **12**(11): e2019MS001821.
+
+60. Golaz J-C, Caldwell PM, Van Rockel LP, Petersen MR, Tang Q, Wolfe JD, *et al*. The DOE E3SM Coupled Model Version 1: Overview and Evaluation at Standard Resolution. *Journal of Advances in Modeling Earth Systems* 2019, **11**(7): 2089-2129.
+
+61. Dang C, Zender CS, Flanner MG. Intercomparison and improvement of two-stream shortwave radiative transfer schemes in Earth system models for a unified treatment of cryospheric surfaces. *The Cryosphere* 2019, **13**(9): 2325-2343.
+
+62. Wang H, Easter RC, Zhang R, Ma P-L, Singh B, Zhang K, *et al*. Aerosols in the E3SM Version 1: New Developments and Their Impacts on Radiative Forcing. *Journal of Advances in Modeling Earth Systems* 2020, **12**(1): e2019MS001851.
+
+63. Burkey J. Mann-Kendall Tau-b with Sen's Method (enhanced). 2022.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SM.pdf

0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/preprint/preprint.md

@@ -0,0 +1,500 @@
+Cell cycle-regulated release of a replication inhibition complex in Vibrio cholerae
+
+Marie-Eve Val
+marie-eve.kennedy-val@pasteur.fr
+
+Institut Pasteur, Université Paris Cité, CNRS UMR3525   https://orcid.org/0000-0001-7097-9072
+
+Theophile Niault
+Institut Pasteur
+
+Ariel Talavera
+Université libre de Bruxelles
+
+Eric LeCam
+Institut Gustave Roussy, Université Paris Saclay, CNRS UMR9019
+
+Ole Skovgaard
+Roskilde University   https://orcid.org/0000-0002-4860-1847
+
+Florian Fournes
+DMF, UNIL
+
+Léa Wagner
+Institut Pasteur, Université Paris Cité, CNRS UMR3525
+
+Hedvig Tamman
+Université libre de Bruxelles   https://orcid.org/0000-0003-4453-7814
+
+Andrew Thompson
+SOLEIL Synchrotron
+
+Abel Garcia Pino
+Université libre de Bruxelles
+
+Didier Mazel
+Institut Pasteur, Unité Plasticité du Génome Bactérien, UMR3525, CNRS   https://orcid.org/0000-0001-6482-6002
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+
+Version of Record: A version of this preprint was published at Nature Communications on January 8th, 2025. See the published version at https://doi.org/10.1038/s41467-024-55598-9.
+Cell cycle-regulated release of a replication inhibition complex in Vibrio cholerae
+
+Théophile Niault1,2, Ariel Talavera3, Eric Le Cam4, Ole Skovgaard5, Florian Fournes1, Léa Wagner1, Hedvig Tamman3, Andrew Thompson6, Abel Garcia Pino3*, Didier Mazel1* and Marie-Eve Val1*
+
+1 Institut Pasteur, Université Paris Cité, CNRS UMR3525, Unité Plasticité du Génome Bactérien, Département Génomes et Génétique, Paris 75015, France.
+2 Sorbonne Université, Collège Doctoral, Paris 75005, France
+3 Cellular and Molecular Microbiology, Faculté des Sciences, Université libre de Bruxelles (ULB), Boulevard du Triomphe, Brussels, Belgium.
+4 Genome Integrity and Cancer UMR 9019 CNRS, Université Paris Saclay, Gustave Roussy, Villejuif 94805, France
+5 Department of Science, Systems and Models, Roskilde University, Roskilde DK-4000, Denmark.
+6 SOLEIL Synchrotron, Saint-Aubin - BP48, Gif sur Yvette, France
+* Corresponding authors
+ABSTRACT
+
+Over 10% of bacteria have expanded their genomes through the domestication of megaplasmids, which subsequently evolved into secondary chromosomes encoding core functions. A fundamental challenge of this genomic expansion is coordinating the replication of multiple replicons within a single cell cycle. In *Vibrio cholerae*, the replication of the secondary chromosome (Chr2) is intricately linked to the replication of the primary chromosome (Chr1) via a unique checkpoint sequence, *crtS*. This sequence binds to the initiator of Chr2, RctB. While *crtS* replication on Chr1 triggers Chr2 replication, the specific molecular dynamics remain elusive. To investigate this, we conducted a comprehensive genome-wide analysis of RctB binding patterns in *V. cholerae* across various cell cycle stages. Our findings show that RctB primarily binds to sites inhibiting replication initiation at the Chr2 origin (*ori2*). This inhibitory effect is counteracted when *crtS* replicates on Chr1, causing a shift in RctB binding to sites that activate replication at *ori2*. Structural analyses support the formation of diverse oligomeric states of RctB, coupled to the allosteric effect of DNA, which determine *ori2* accessibility. We propose a synchronization model where, upon replication, *crtS* locally destabilizes the RctB inhibition complex, releasing the Chr2 replication origin.
+INTRODUCTION
+
+Replication initiation is a crucial step in the bacterial life cycle, subject to complex regulatory controls to adapt to fluctuating growth conditions \(^{1,2}\). This complexity increases further for bacteria with multipartite genomes, which require an additional layer of control to replicate multiple chromosomes simultaneously \(^{3}\). Although substantial evidence across a variety of bacterial species supports the idea that the replication of multiple chromosomes must be coordinated \(^{4-9}\), *Vibrio cholerae* is the only species where a specific mechanism synchronizing the replication of two chromosomes has been identified \(^{10-12}\). In this pathogen, a replication checkpoint sequence, called *crtS*, ensures the synchronous replication and termination of the primary (Chr1) and secondary chromosome (Chr2) within a single replication cycle \(^{12}\). The objective of the present study is to elucidate the molecular mechanisms by which *crtS* orchestrates this synchronization process.
+
+*V. cholerae* Chr2 originates from an iteron-type plasmid, evident from its replication origin (*ori2*) and initiator (RctB) \(^{13}\). RctB is divided into four structural regions (I – IV) \(^{14}\) (Supplementary Fig. 1a). Its two central domains (II, III) are structurally similar to the Rep iteron plasmid initiators domains (WH1, WH2) while domains (I, IV) are unique to RctB \(^{14}\). Domains, I, II and III, interact with DNA via Helix-Turn-Helix (HTH) motifs and are essential for *ori2* replication \(^{14}\). The C-terminal domain (IV) is required to down-regulate Chr2 initiation and to coordinate Chr1 and Chr2 replication \(^{15,16}\). RctB recognizes three types of double-stranded DNA sites: iterons, 29/39m and *crtS*. To promote replication initiation at *ori2*, RctB binds to an array of six regularly spaced iterons (six 12-bp repeats), which prompts DNA unwinding at an adjacent AT-rich region known as the DNA Unwinding Element (DUE) (Supplementary Fig. 1b) \(^{17}\). At this location, RctB engages with single strand of the open DUE on six regularly spaced direct repeats (5'-ATCA) \(^{18}\). A nucleoprotein complex, composed of RctB, IHF and DnaA, enables the recruitment of the replicative helicase DnaB, similar to what occurs in iteron plasmids \(^{19}\). Iterons sites each contain a GATC motif which is specifically recognized and methylated by the Dam methylase. Iteron sites must be methylated on both DNA strand for RctB to bind \(^{20}\). RctB recognizes another type of sites, named for their size as 29-mer or 39-mer (29/39m), which differ from iterons in both sequence and function (Supplementary Fig. 1b). The 29/39m sites play a regulatory role in inhibiting *ori2* initiation \(^{21}\). Two 39m sites and one 29m site flanking the minimal *ori2* (containing the six iteron array and the DUE) participates in RctB-mediated iterons handcuffing (pairing via initiator bridges) \(^{21,22}\). The 29m site is also found in the *rctB* gene promoter and contributes to *rctB* self-repression \(^{22}\). Additionally, RctB binds to a unique 62 bp sequence on Chr1, named *crtS* \(^{11,12,23}\). This sequence is crucial for synchronizing the replication of Chr1 and Chr2 \(^{12}\). The replication of *crtS* triggers the initiation of Chr2, and its position relative to Chr1 origin (*ori1*) sets the replication timing for Chr2 \(^{12}\). If *crtS* is deleted, *V. cholerae* undergoes lethal loss of Chr2 due to under-initiation at *ori2* \(^{12}\). Although the action of *crtS* has been
+indirectly linked to the inhibitory function of the 29/39m sites \(^{16}\), the exact molecular dynamics that guide the coordination of Chr1 and Chr2 replication by \(crtS\) remains largely undefined.
+
+In this study, we conducted a comprehensive, genome-wide analysis of RctB binding, unveiling key elements of the \(crtS\)-triggered mechanism for Chr2 replication initiation at various stages of the V. cholerae cell cycle. We identified novel RctB binding sites and established a dynamic pattern of RctB binding throughout the cell cycle. Our findings reveal that RctB predominantly binds to the 29/39m sites within ori2, effectively preventing the initiation of Chr2 replication for a large portion of the cell cycle.
+
+Using transmission electron microscopy (TEM), we observed that RctB forms large nucleoprotein complexes at ori2, linking the 29/39m sites together. We show that RctB Domain IV can mediate alternative dimerization interfaces, suggesting their role in the assembly of oligomeric bridging structures when bound to 29/39m sites. Following \(crtS\) replication, we observed a marked shift in RctB binding preferences at ori2, transitioning towards the iterons and the DUE, thereby enabling Chr2 replication initiation. Using an integrative approach combining ChIP-seq, live cell microscopy and structural studies, we propose a model of the firing of Chr2 replication in which one \(crtS\) duplication event directly triggers the activation of one ori2 by destabilizing an inhibition complex.
+
+RESULTS
+
+Genome-wide binding profile of RctB in V. cholerae
+
+We conducted a ChIP-seq of RctB in V. cholerae to investigate its genome-wide binding profile within an exponentially growing population. In comparison to a previous ChIP-on-chip study \(^{11}\), we found that RctB binds to more regions across both chromosomes. These include 8 regions on Chr1 and 6 regions on Chr2, one of which encompasses ori2 (Fig. 1a, Supplementary Table 1). Most of these regions contain either iterons-like sites with a 5'-TGATCA inverted repeat or 29/39m-like sites with a 5'-TTACGG motif, as revealed by MEME analysis \(^{24}\) (Fig. 1b). We also identified three regions that do not share any sequence similarity with these motifs (Supplementary Table 1), including \(crtS\) in the intergenic region between VC0764 and VC0765 on Chr1 (Fig. 1a, Supplementary Fig. 2) \(^{23}\). Given that our ChIP-seq was executed on non-synchronized, exponentially growing bacterial cultures, the heights of the peaks offer valuable insights into the frequency with which RctB binds at various sites throughout the cell cycle. To assess the impact of the newly discovered RctB binding sites on Chr2 replication, we used a plasmid-based reporter system, pORI2, that exclusively relies on ori2 for replication \(^{23}\). Each RctB binding regions (spanning 250bp on each side and centered around the peak) was inserted into the lacZ gene of Escherichia coli chromosome, and pORI2 copy number was measured using quantitative digital PCR (dPCR). Our results show that \(crtS\) stands out as the only strong enhancer of ori2 initiation (Fig. 1c). Mutations in three of the most prominent binding sites (VC0643, VC01643, and VC1042-VC1043) had
+minimal effects on Chr2 replication in V. cholerae; although the 39m-like site within VC0643 did display a weak, yet statistically significant, negative effect on Chr2 copy number (Fig. 1d). Given RctB’s ability to repress its own transcription by binding to its promoter \(^{22}\), we explored whether RctB could repress other genes. We monitored the transcription levels of VC1042 and VC1803 genes in V. cholerae, both of which have an RctB binding site within their promoters. RT-dPCR analyses revealed minimal to no differences in the expression of these genes, regardless of RctB presence (Fig. 1e). Although the RctB binding in the VC1803 promoter exhibited a weak, yet statistically significant, positive regulatory effect on VC1803 transcription. VC1803 is a component of Vibrio pathogenicity island 2 (VPI-2) and contains a Cro/C1 repressor-like HTH domain profile commonly found in transcriptional repressors of temperate bacteriophages, suggesting potential gene regulation by RctB within the pathogenicity island. In conclusion, while crtS plays a pivotal role in regulating Chr2 replication, the functional relevance of the newly discovered RctB binding sites on Chr1 remains uncertain. RctB appears to regulate its own transcription exclusively, in contrast to the initiator DnaA, which acts more broadly as a general transcription factor \(^{25}\).
+
+RctB binds predominantly to 29/39m inhibitory sites within ori2
+Upon closer examination of the ori2 region, we observed that RctB ChIP signal was about 10-fold higher than at other chromosomal loci (Supplementary Fig. 2). RctB showed a strong preference for binding to the 29/39m sites (Fig. 2a). This predominant occupancy of RctB at inhibitory sites suggests that initiation of Chr2 replication is hindered for most of the cell cycle. To understand RctB binding dynamics in non-replicating cells, we performed a ChIP-seq of RctB during stationary phase. The RctB binding pattern at ori2 mirrored that observed during the exponential phase (Supplementary Fig. 3), suggesting that RctB binding to 29/39m inhibits initiation at ori2 during both replicative and non-replicative phases. In both conditions, the signal at the 29m site was approximately 3-fold higher than that at the two 39m sites. We hypothesized that this discrepancy might arise from the ParB2 protein, essential for Chr2 segregation, competing with RctB for binding to both 39m sites, as previously reported \(^{26}\). Confirming our suspicion, our ChIP-seq of ParB2 on ori2 (Supplementary Fig. 4) revealed that ParB2 does indeed compete with RctB at both 39m sites and not at the 29m site. To further investigate the binding requirements of RctB at ori2, we performed a ChIP-seq analysis of RctB in a strain deleted for dam methylase (\(Δdam#4\)). RctB cannot bind to iterons in this mutant because they are not methylated, and so there is no initiation at ori2 \(^{20,27}\). The \(Δdam#4\) mutant is viable due to the integration of Chr2 into Chr1, enabling its replication from ori1 as the DnaA initiator does not require Dam methylation \(^{27}\) (Supplementary Fig. 5a). In the \(Δdam#4\) mutant, the RctB binding pattern at ori2 was identical to that of the wild-type strain (Supplementary Fig. 5b). This result demonstrates that RctB molecules can form
+a complex at ori2 strictly through their interaction with 29/39m sites, without any engagement with iterons. It also confirms that RctB binding to crtS is independent of Dam as shown in 23.
+
+RctB binding to 29/39m sites generates DNA loops in ori2 and prevents its binding to iterons
+To directly visualize the complex formed by RctB bound to 29/39m on ori2, we used positive-staining transmission electron microscopy (TEM) 28. An unmethylated DNA fragment containing ori2 was chosen to allow RctB to bind exclusively to the 29/39m sites. Our observations revealed that RctB formed large nucleoprotein complexes bridging the 29m and 39m sites together, causing looping within the origin (Fig. 2b, white arrows). Such loops in ori2 could create a steric hindrance preventing the initiation of replication by interfering with RctB binding to the iteron array and the proper unwinding of the DNA strands. To test this hypothesis, we performed a ChIP-seq of RctB in a mutant with the C21 > A mutation at the 29m site (29m*_{C/A}) known to disrupt RctB binding 16. The binding pattern of RctB at ori2 was drastically altered (Fig. 2c). RctB still bound to both 39m sites but not to the 29m. Additionally, we observed a clear binding of RctB to the array of six iterons and the adjacent DUE. Given that initiation of ori2 replication requires RctB interaction with both the iteron array and the DUE 18, our observations suggest that preventing RctB binding to 29m effectively alleviates the constraint on ori2 initiation.
+
+RctB^{IV} harbors two dimerization interfaces
+The C-terminal region of RctB, named domain IV (RctB^{IV}), is critical to downregulate Chr2 copy number through its involvement directly or indirectly in handcuffing activities and binding to 29/39m 15,16,29. It is also crucial for crtS-mediated control of Chr2 replication 16. In vitro, RctB is able to self-oligomerize when bound to DNA through interactions mediated by domain IV 16,23. Thus, we hypothesized that the coordination of Chr1 and Chr2 replication likely depends on the ability of RctB to oligomerize on ori2 via domain IV. While RctB has remained refringent to structural studies, likely due to the presence of intrinsically unstructured regions in the protein, the structure of domain I and domains II-III have been solved 14,30. Therefore, we used a truncated version of RctB to determine the structure of domain IV 14,30 and generated the stable fragment RctB^{IV} (amino acids 532 to 658) that could be used in structural biology. The structure of RctB^{IV} reveals that it is organized into two structural subdomains connected by an extended linker or hinge region (Fig. 3a). The N-terminal subdomain consists of a 3-stranded β-sheet stacking two α-helices and the C-terminal subdomain has 4 α-helices in a bundle (Fig. 3b). In the crystal, RctB^{IV} retains a dimeric association as predicted by Size Exclusion Chromatography (SEC) (Supplementary Fig. 6). This dimerization is primarily mediated by a β-sheet (β3) spanning residue 545 to 549 and involves an interface of 392.4 Å^2 (Fig. 3a-b, primary interface). The crystal lattice contacts of the C-terminal subdomain suggest the presence of a secondary interaction interface involving 315.4 Å^2, which could mediate further oligomerization via this region (Fig. 3c, secondary interface). A DALI 31
+search confirmed the C-terminal subdomain as a protein recognition domain capable of forming transient homodimeric interfaces, with Z-scores between 3.7 and 5.4. This secondary interface closely resembles a representative structural homologue detected by DALI (a putative transcription regulator from Neisseria gonorrhoeae) \(^{32}\). These findings suggest that RctB\(^{IV}\) may associate in multiple ways and potentially form higher-order oligomeric structures when recruited to DNA. The presence of multiple intermolecular interfaces in RctB\(^{IV}\) inferred from the X-ray structure, together with class averages of RctB obtained by cryo electron microscopy (Fig. 3d) are consistent with the presence of dimeric and tetrameric forms RctB. These findings suggest that RctB\(^{IV}\) may associate in multiple ways and potentially form higher-order oligomeric structures when recruited to DNA.
+
+Domain IV dimerization interfaces mediate RctB oligomerization
+
+To further characterize these interfaces, we used the bacterial two-hybrid assay in *E. coli* \(^{33}\). We performed a structure-guided point mutation analysis to probe the potential interfaces involved in the oligomerization of RctB\(^{IV}\). Given that RctB has a separate dimerization interface in domain II \(^{14}\), we substituted residue D314 with a proline (D\(_{314}P\)) to prevent domain II-mediated interactions that would obscure RctB\(^{IV}\)-mediated interactions (as done in \(^{16}\)). In Fig. 3e, blue colonies (D\(_{314}P/D_{314}P\)) indicated that RctB could interact with other RctB through another domain, as already reported to be via domain IV \(^{16}\). Our present results showed that substitutions in the β3 strand of RctB\(^{IV}\) caused the formation of white colonies (D\(_{314}P/D_{314}P\)-I\(_{546}G\)-I\(_{546}G\)), indicating that β3 is important for RctB interactions while substitutions that destabilize the α1 helix appear to have little impact on dimerization as evidenced by the formation of blue colonies (D\(_{314}P/D_{314}P\)-A\(_{565}P\)). Deletion of the linker connecting the two subdomains (591-596), gives blue colonies (D\(_{314}P/D_{314}P\)-Δ\(_{591-596}\)) but to a lesser extent than the A\(_{565}P\) substitution, suggesting that RctB self-association may be sensitive to local structural rearrangements between the two subdomains. The I\(_{625}P\) and L\(_{651}P\) substitutions in two different α-helices (α3 and α6) of the C-terminal subdomain also abrogate RctB interactions confirming our previous results \(^{16}\). Thus, RctB\(^{IV}\) can form homodimers via two different regions: at a primary interface formed between the β3 sheets and at the secondary interface between α-helices. Collectively these results suggest that RctB\(^{IV}\) could be an anchor not only for dimerization but also for the formation of transient oligomeric structures when bound to DNA or even to bridge two chromosomal regions.
+
+RctB oligomerization is required for the binding to inhibitory sites at *ori2*
+
+To better understand the role of RctB\(^{IV}\)-mediated interactions on the genome-wide *in vivo* binding of RctB, we performed ChIP-seq of RctB-L\(_{651}P\) (Fig. 3f). These analyses revealed a completely different binding profile to *ori2* compared with RctB wt (Fig. 2a). RctB-L\(_{651}P\) binding to *ori2* is fully shifted to the iterons array and the DUE and is absent from the 29/39m sites. However, we observed the same binding
+profile at crtS between RctB-L651P and RctB wt (Supplementary Fig. 7a). This suggests that, although not involved in DNA binding, RctBIV likely mediates the differential recognition of 29/39m inhibitory sites relative to iterons or crtS, probably via the formation of alternative oligomeric structures. As a control, we performed ChIP-seq of RctB-L651P in a Δdam context. As expected, in the absence of Dam methylation, no binding of RctB-L651P to ori2 was observed, although RctB still bound to crtS (Supplementary Fig. 7b). Based on these results, we propose that intermolecular domain IV mediated RctB interactions are crucial for the inhibition of replication at ori2 through the bridging of 29/39m sites.
+
+crtS exclusively affects the RctB inhibition complex at ori2
+
+The triggering mechanism of ori2 replication by crtS remains poorly understood. In a ΔcrtS strain, Chr2 is under-initiated, resulting in cell filamentation and lower Chr2 copy number than Chr1 12. To better understand the consequences of the loss of crtS, we conducted a ChIP-seq analysis of RctB in that strain (Fig. 4a). In this background, RctB also predominantly binds to the 29/39m inhibitory sites at ori2, but the ChIP signal was approximately 10-fold higher, indicating a much stronger inhibition of replication in the absence of crtS. Based on our new findings, we propose that crtS could trigger the initiation of Chr2 replication by destabilizing the nucleoprotein complex that impedes ori2 initiation. When RctB carried the L651P mutation, there was no difference in its binding profile at ori2, irrespective of the presence or absence of crtS, indicating that crtS solely impacts the binding of RctB to the 29/39m (Supplementary Fig. 7a). Given that RctB binds to other 39m-like sites outside of ori2 (Supplementary Table 1), we hypothesized that the absence of crtS could enhance the binding of RctB to these sites as well. However, we observed no increase in RctB binding to the other 39m in the ΔcrtS mutant (Supplementary Fig. 8). This suggests that crtS has an exclusive activity on ori2. To investigate this further, we examined RctB binding on crtS-containing DNA fragments by TEM. The binding of RctB on crtS was restricted to the site and induced a kink in the DNA (Supplementary Fig. 9a). When we mixed DNA substrates containing unmethylated ori2 and crtS with RctB, we detected sporadic RctB-mediated contacts between ori2 and crtS (Supplementary Fig. 9b), suggesting the possibility of direct interactions between crtS and the 29/39m inhibitory sites in ori2. These findings align with the high frequency of contacts between chromosomal regions proximal to crtS and ori2, observed in vivo by chromosome conformation capture 12.
+
+The coupled replication of crtS and activation of ori2 is stoichiometric
+
+Past studies 12,23,34 demonstrated that introducing an additional crtS copy into Chr1 increases Chr2 copy number. To delve deeper into this relationship, we focused on strains with two crtS sites. First, we inserted one or two copies of crtS into the attTn7 site (located near VC0487 on Chr1) and deleted the endogenous crtS site. As anticipated, the strain with two crtS exhibited a higher Chr2 copy number
+compared to the strain with a single crtS when measured on stationary phase culture (1.7 vs. 1) (Supplementary Fig. 10a). However, when comparing the ChIP-seq profiles of RctB at ori2 between these strains, we did not observe any significant differences; in both cases, RctB mainly binds to the 29/39m sites (Supplementary Fig. 10b). This observation indicates that having two copies of crtS does not alter the binding of RctB to activate replication at ori2. Consequently, any activation of ori2 after crtS replication probably occurs transiently. To gain a temporal perspective, we used fluorescence microscopy to observe a V. cholerae mutant with two distant crtS sites \(^{12}\). This strain contains the endogenous site (crtS_{wt}) and a second site added next to ori1 (crtS_{ori1})—thus delaying the replication timing of both crtS sites (Fig. 4b-d). We followed the dynamics of both crtS and ori2 loci using three different DNA binding protein fluorescent proteins and their cognate sites inserted in the genome: ParB_{Pl}-CFP (\(parS_{Pl}\) site near crtS_{ori1}), ParB_{MT1}-yGFP (\(parS_{MT1}\) site near crtS_{wt}), LacI-RFP (\(lacO\) array near ori2) \(^{35}\). Compared to the wild type, which contains only one or two foci of ori2, the two-crtS containing mutant has up to four ori2 foci (Fig. 4b). We tracked the longitudinal position of ori2 foci relative to the Chr1 sites in growing cells (Fig. 4d vs. Fig. 4c) and observed that the two-crtS containing cells are born with two ori2 foci. As cell size increases, we observed a sequential increase in the number of ori2 foci, going from two to three and then to four ori2. These results suggest that the early replication of crtS_{ori1} triggers the replication of only one of the two ori2. The replication of the second site crtS_{wt} activates the initiation of the second ori2.
+
+We took a closer look at the stoichiometry by Marker Frequency Analysis (MFA) on three exponentially growing strains with different crtS contents: the wild type (crtS_{wt}), one mutant with crtS relocated at ori1 (crtS_{ori1}), and the two-crtS containing mutant (crtS_{wt}, crtS_{ori1}). First, we confirm that the location of crtS on Chr1 controls the time of initiation of Chr2 and thereby the copy number of Chr2, as in crtS_{wt} it is 0.484 and in crtS_{ori1} it is 0.601 (Fig. 4e, left and center) as observed before \(^{12}\). Then, the strain with two crtS sites (crtS_{wt} and crtS_{ori1}) have an ori2 copy number of 0.900 approximating the sum of ori2 copy number in the single crtS strains: crtS_{wt} and crtS_{ori1} (0.484 + 0.601). These findings align with a model in which one crtS duplication event specifically activates one single ori2. This model is illustrated in Supplementary movie 1 (crtS_{wt}), movie 2 (crtS_{ori1}) and movie 3 (crtS_{wt}, crtS_{ori1}).
+
+crtS replication induces a temporal shift in RctB binding at ori2
+
+Based on these results, we hypothesized that to activate replication at ori2, RctB should interact with the iterons in a transient way, while remaining bound to the inhibitory 29/39m sites for most of the cell cycle. Our previous approach did not allow us to capture short-lived events. Therefore, we implemented the ChIP-seq of RctB in a replication-synchronized population to monitor the RctB binding profile over the course of the cell cycle. Similar to E. coli, an exponentially growing population of V. cholerae can be synchronized using serine hydroxamate (SHX, a serine analogue that chemically stimulates the stringent
+response) \(^{36-38}\). During treatment with SHX, replication forks can still progress but the initiation of new rounds of replication is hindered. Transfer of these cells to a fresh medium (without SHX) allows a theoretical synchronous restart of replication. Indeed, previous work showed that after SHX treatment, *V. cholerae* Chr1 re-initiates first and then Chr2 \(^{37}\) presumably due to control by *crtS* \(^{12}\). In our hands, the addition of SHX to 1.5mg/mL final concentration to an exponentially growing culture of *V. cholerae* caused an instantaneous growth arrest (Fig. 5a). To ensure completion of ongoing replication, we measured the ratio of loci at the origin and terminus of Chr1 (ori1/ter1) by dPCR (Fig. 5b). After 120min of SHX treatment, all cells successfully completed replication with a ratio of ori1/ter1 = 1 meaning that Chr1 was fully replicated. Once SHX was removed, the cells restart to replicate within a time window of 20-30 minutes, similar to what was observed \(^{38}\). While synchronization by SHX is not as perfect in *V. cholerae* as it is in other bacteria, it provided a reasonable approximation to a synchronized population. Thus, we decided to use it to perform ChIP-seq of RctB at different time points after removal of SHX (15, 30, 45, 60 minutes) to follow the binding cycle of RctB.
+
+Using reads from the input file (control without immunoprecipitation), we performed MFA to track replication fork progression on both chromosomes (Fig. 5c-5f, top panel). Within the 15 minutes after removal of SHX (Fig. 5c), replication has not restarted, as shown by the flat MFA profile of Chr1 and Chr2, RctB predominantly binds to 29/39m on *ori2*. After 30 minutes (Fig. 5d), Chr1 starts replicating and at 45 minutes (Fig. 5e) the *crtS* site seems to be just replicated (the Ter region of Chr1 still appears not replicated), while Chr2 replication has not yet started. At both time points, RctB ChIP-seq patterns at *ori2* is nearly unchanged (Fig. 5d-5e). At 60 minutes (Fig. 5f), Chr2 is finally replicating. On the ChIP-seq profile, two additional peaks appear: one on the six-iteron array and one on the DUE (Fig. 5f, black arrows). Here, we were able to capture a release of RctB from the inhibitory sites (29/39m) towards the activating sites (iteron) and towards the DUE. This temporal shift in the binding pattern of RctB from inhibitory to activating sites on *ori2* underscored the opening of *ori2* that resulted in the initiation of Chr2 replication. As expected, *ori2* activation follows *crtS* replication on Chr1. We did not observe fluctuations in the binding of RctB at *crtS* before or after the passage of the replication fork, nor during stationary versus exponential phase of growth (Supplementary Fig. 3 & 11). Thus, RctB binds similarly to *crtS* throughout the cell cycle.
+
+DISCUSSION
+
+Our work provides cell-cycle-integrated insights into the synchronized replication of Vibrio’s two chromosomes. We identified an inhibition complex at *ori2* that prevents replication initiation for a significant portion of the cell cycle. We also highlighted the pivotal role of *crtS* in temporarily disrupting this complex, enabling Chr2 replication. Based on the binding dynamics of RctB throughout the cell cycle, we propose a model for the regulated control of Chr2 replication. In this model, *ori2* is inherently
+bistable, with the replication of crtS acting as a decisive ON switch. Before crtS replication, ori2 is sequestered by an inhibition complex, with RctB bridging the 29/39m sites, and inducing DNA loops that effectively keep ori2 in an OFF state. The replication of crtS destabilizes this complex, causing transient exposure of ori2 to an ON state compatible with replication initiation. This shift promotes the recruitment of RctB to the six iterons, the DUE, and ultimately, the initiation of Chr2 replication.
+
+A crucial feature of the stabilization of the OFF state is the oligomerization properties of RctB via the domain IV which defines the formation of the 29/39m bridges and preclude Chr2 replication. In a broader context, analogous types of nucleoprotein complexes have been observed intermolecularly in the regulation of iteron plasmid copy number, involving the initiators RepA_{PPS10}^{39} and P_{R6K}^{40}. In our model, we propose that initiation silencing occurs through intramolecular DNA looping at the origin. This is akin to the mechanism in the F plasmid, where RepE initiators establish a bridge between the iterons of the negative control element (incC) and those of the origin (oriF) that further inhibit the formation of an open complex^{41}. Vibrio distinguishes itself by having tailored specific 29/39m sites to serve this regulatory role.
+
+The replication of crtS triggers a shift in RctB binding, favoring its binding with the iterons and DUE, conditions that are compatible with ori2 initiation. Further mechanistic research is needed to explore the molecular interactions between RctB and DNA, as well as how crtS disrupts the inhibition complex. However, our model is consistent with the bistable state of ori2 being influenced by crtS replication, the structural properties of RctB and the dramatic changes in the DNA structure and organization induced by RctB. Given that crtS specifically influences RctB binding at ori2—without affecting other genomic sites—and considering the stoichiometric relationship between crtS replication and ori2 activation, our data strongly suggest a direct physical interaction that leads to the activation of ori2 by crtS. This is consistent with our previous observations of frequent inter-chromosomal interactions proximal to crtS and ori2^{12}. Moreover, we observed that the regulatory control exerted by crtS over ori2 is lost when a single 29m or 39m site is mutated^{16}. This synergistic effect implies that crtS likely disrupts the 29/39m bridging, rather than affecting each site individually. We propose that each newly duplicated crtS site have RctB bound to it. Both crtS-RctB complexes can each interact with one 29/39m-RctB complex. These interactions destabilize the DNA loop which inhibit ori2 initiation. Once ori2 is exposed, the iteron array and DUE become accessible, so that free RctB molecules can bind and initiate replication (Fig. 6).
+
+In contrast to plasmids, where initiators bind solely to iterons, RctB capacity to interact with multiple sites offers a more refined control over replication initiation. This multifaceted binding ability is determined by factors such as methylation-dependent binding to iterons^{20}, domain IV-oligomerization-dependent binding to 29/39m (this study), and Lrp-enhanced binding to crtS^{42,43}. All these factors introduce dynamic binding behaviors throughout the cell cycle, ensuring that Chr2 replication is finely tuned and synchronized with other cellular processes.
+METHODS
+
+Strains and growth conditions
+Vibrio cholerae N16961 El tor (7PET) is the strain studied here. As a heterologous system for dPCR experiments Escherichia coli K12 MG1655 was also used. Different E.coli strains were used for cloning DH5α, TOP10, Π3813 and β3914\(^{44}\) for conjugation. LB (Luria Broth) was used as standard rich medium for both E. coli and V. cholerae.
+
+Strains and plasmids constructions
+Both V. cholerae and E. coli strains were modified using conditionally replicative suicide vectors as described in \(^{45}\). These vectors were generated by Gibson assembly \(^{46}\), and in cases requiring specific point mutations, they were introduced via PCR using oligonucleotides containing the desired mutations at the 5' end. For ChIP-seq experiment (on RctB and on ParB2), we replaced the wild-type alleles on the chromosome with a C-terminal 3xFLAG through allelic exchange. For the dPCR experiments, we introduced RctB secondary binding sites into the lacZ gene of MG1655, we took an extending of 250 base pairs on each side of the peak detected from ChIP-seq analysis. Subsequently, these modified strains were transformed with a pORI2 plasmid, following the procedure outlined in \(^{23}\). Plasmids and strains used in this study are listed in Supplementary Tables 2 and 3.
+
+High Resolution ChIP-seq
+ChIP-seq were performed as described in \(^{47}\) in V. cholerae expressing protein-3xFLAG under its native promoter. We confirmed that 3xFLAG did not impair protein function by dPCR measurement of Chr2 copy number. Cells were grown in 100mL Mueller Hinton (MH) media with shaking at 37°C until they reach mid-log growth phase (OD\(_{600}\) 0.5). Then cells were fixed by crosslinking with 1% formaldehyde for 30 min at room temperature under agitation, followed by quenching with 0.5M glycine for 15 min. Bacteria were harvested, washed two times in ice-cold Tris-Buffered saline (20 mM Tris/HCl pH 7.5, 150 mM NaCl) and pellets were stored at -80°C. Pellets were lyzed in buffer containing lysozyme and protease inhibitor pill (cOmplete Protease Inhibitor, Roche). Chromatin was then fragmented by sonication (Covaris S220) for in milliTUBE (1ml with AFA Fiber). Correct DNA fragmentation centered around 300bp was confirmed using 1.8% agarose gel electrophoresis. Following sonication, protein-DNA complexes were immunoprecipitated with anti-FLAG magnetic beads (M8823 Sigma). Correct immunoprecipitation was confirmed by western blot using anti-FLAG antibodies (F7425 Sigma). Once recovered, immunoprecipitated DNA was de-crosslink at 65°C overnight. Then DNA was blunt-ended, A-tailed, ligated to adaptors, amplified by PCR (13 cycles) and purified using TruSeq ChIP Library Preparation Kit (illumina) and AMPureXP beads (Beckman Coulter). ChIP-seq libraries quality was
+measured by Agilent Bioanalyzer 2100 instrument using a DNA High Sensitivity chip. Finally, libraries were sequenced using a MiniSeq Mid Output Kit on a Miniseq sequencing machine (Illumina). Paired-end reads from two independent RctB ChIP-seq experiments were mapped to Vibrio cholerae N16961 reference genome (Chr1: CP028827.1 and Chr2: CP028828.1) with Bowtie2 using galaxy.pasteur platform. Peak calling was performed using MACS2, where aligned reads were significantly enriched compared to a control (strain without FLAG). Identified peaks were checked by hand, using IGV genome browser to confirm correct peak assignment. MEME-ChIP (version 5.4.1) online browser was used for motif enrichment finding. For normalization, the IP and INPUT files were first adjusted to the same number of reads using R. Then the coverage value for each position in the IP file was divided by the average coverage value of a 2kb sliding window from the INPUT file. Data available at the European Nucleotide Archive (Supplementary Information).
+
+The genomic replication pattern of the different strains was examined using a computational approach based on Marker Frequency Analysis (MFA). The reads from the INPUT sample (not immunoprecipitated DNA) of the ChIP-seq data were mapped to V. cholerae N16961 reference genome (Chr1: CP028827.1 and Chr2: CP028828.1) using BOWTIE2, the resultant mapping was saved in two separate Binary Alignment Map (BAM) files corresponding to the two chromosomes. Subsequently, a R script (available on demand) was used for data processing and visualization. The depth of coverage was extracted from the BAM files using the Rsamtools package. The genome was divided into 10 kbp bins, and the average coverage for each bin was computed. Coverage values were then normalized with respect to the average coverage of the termination (ter) region, which is located at the midpoint of the first chromosome (+/- 5kb). Coverage values more than three standard deviations from the mean were considered outliers and removed from the dataset. For visualization, a ggplot2-based plot was generated, showing the normalized coverage for each bin against the genomic position in Mbp. Linear regression analysis was performed separately on each half of the chromosomes, and the best-fit lines were superimposed on the plot (Supplementary Fig. 12).
+
+Marker Frequency Analysis
+Grow conditions, DNA extraction, sequencing and analysis for MFA was done as described previously\(^{12}\). Data available at the European Nucleotide Archive (Supplementary Information).
+
+Digital PCR quantification (dPCR)
+Quantifications of (ori1 and ori2) in V. cholerae and (oriC and pORI2) in E. coli were performed as described in \(^{16}\) using multiplex dPCR (Stilla Technologies). dPCR was performed directly on cell lysate: 1mL of overnight culture was washed twice with 1mL PBS, pellets were then frozen at -20°C and resuspended in 200uL PBS, samples were then boiled for 10min. Samples were centrifuge for 10min,
+supernatant was recovered, and DNA content was measured using Qubit device (ThermoFisher). PCR reactions were performed with 0.1 ng of DNA using the PerfeCTaMultiPlex qPCR ToughMix (Quantabio) on a Sapphire chip (Stilla Technologies). Digital PCR was conducted on a Naica Geode and Image acquisition on the Naica Prism3 reader. Images were analyzed Crystal Miner software (Stilla Technologies). The dPCR run was performed using the following steps: droplet partition (40°C, atmospheric pressure AP to +950mbar, 12 min), initial denaturation (95°C, +950 mbar, for 2 min), followed by 45 cycles at (95°C for 10 s and 60°C for 30 s), droplet release (down 25°C, down to AP, 33 min).
+
+Digital PCR quantification of RNA (RT-dPCR)
+Quantification of RctB mRNA was performed in V. cholerae from exponentially growing cultures (OD600 0.5) using multiplex digital RT-dPCR (Stilla Technologies) 48. Primers and probes are listed in Supplementary Table 4. V. cholerae RNA was prepared as described in 49. PCR reactions were performed with 1 ng of RNA using the qScriptTM XLT One-Step RT-qPCR ToughMix® (Quantabio). RT-dPCR and data analysis were conducted as described above for dPCR. The RT-dPCR run was performed in the following steps: droplet partition (40°C, AP to +950 mbar, 12 min), cDNA synthesis (50°C, +950 mbar, 10 min), initial denaturation (95°C, +950 mbar, for 1 min), followed by 45 cycles at (95°C for 10 s and 60°C for 15 s), droplet release (down 25°C, down to AP, 33 min). Expression values were normalized to the expression of the housekeeping gene gyrA as described in 49.
+
+SHX Vibrio cholerae synchronization
+V. cholerae expressing RctB-3xFLAG from its natural promoter was streaked from a -80°C stock onto an MH agar plate. An isolated colony was then picked and cultured overnight in M9 media containing 0.2% casamino acids and 0.4% glucose at 30°C. Cells from this overnight culture were subsequently diluted 100-fold in fresh media and grown in 40 mL of M9 media with 0.2% casamino acids and 0.4% glucose at 30°C. When the cells reached an OD600nm of 0.5, they were treated with SHX (DL-serine hydroxamate, Sigma) at a final concentration of 1.5 mg/mL for 120 minutes. This treatment prevented new rounds of initiation and allowed cells to complete ongoing replication. Cells stalled in G1 were then centrifuged at 6,000g for 15 minutes, and the pellets were resuspended in 40 mL of fresh M9 media before being returned to 30°C for growth. Cells were fixed with 1% formaldehyde at various time points: 15, 30, 45, and 60 minutes after the wash. Finally, samples were processed according to the ChIP-seq protocol.
+
+Transmission electron microscopy
+The ori2 DNA probe (1864bp) containing the three 29/39mer sites was produced by PCR from pFF149 plasmid using MV109/MV226 primers (Supplementary Table 4). The crtS probe (944bp) was produced
+by PCR from a pUC18::crtS plasmid using MV577/MV485 primers. DNA fragments were then purified using AMPure XP beads (Beckman Coulter). RctB, DnaK and DnaJ proteins were produced and purified like in PMC6212839. RctB was first treated with DnaK/J chaperones to enhance RctB activity. For this, RctB (1200nM) was incubated with DnaK (100nM) and DnaJ (100nM) in 20uL of 10 mM Tris-HCl pH 7.5, 50 mM NaCl, 1mM MgCl2 buffer for 10min at 4°C. Then, for the DNA-protein complexes formation 2uL of treated RctB was mixed with 4nM of DNA probes (ori2 probe, crtS probe or both) in a 10 mM Tris-HCl pH 7.5, 50 mM NaCl, 1mM MgCl2 buffer (20uL final volume) for 10min at room temperature. For observation of DNA-protein complexes, 5 µL drop of the incubation was deposited on a 600-mesh copper grid previously covered with a thin carbon film and preactivated by glow discharge in the presence of amylamine (Sigma-Aldrich, France). The grids were rinsed and positively stained with 2% (w/v) aqueous uranyl acetate, carefully dried with filter paper, and observed in annular dark-field mode in lossless filtered imaging using a Zeiss 902 transmission electron microscope. Images were captured at 85,000× magnification with a Veleta MegaviewIII CCD camera and analyzed with iTEM software (Olympus Soft Imaging Solution).
+
+Expression and purification of RctB and RctBIV for crystallization and biophysical studies
+
+RctB and were RctBIV cloned into a pET-24b plasmid downstream an N-terminal 10xHis_SUMO tag. BL21 E.coli cells were transformed with these constructs for expression and purification. Transformed cells were first grown at 37°C until reaching an OD ~0.6 then expression was induced with 0.25 mM Isopropyl β-D-1-thiogalactopyranoside (IPTG) overnight at 16°C. Afterwards cells were harvested by centrifugation at 3000g.
+
+For purification, the cell lysate supernatants were loaded onto a Ni-Sepharose column. In both cases, proteins were eluted using a step gradient of imidazole. The fractions containing the protein of interest were then loaded into a Citiva 60/160 Seperdex 200 size exclusion chromatography column, equilibrated in 50 mM Hepes pH 7.5, 500 mM NaCl, 500 mM KCl, 2 mM MgCl2 and 1 mM TCEP. The 10xHis-SUMO tag was removed by incubating overnight the protein with the protease Ulpl in a 1:100 molar ratio. The tag-free RctB and RctBIV were then isolated from the cleaved tag and Ulpl by loading the mixture into a Co-sepharose column and collecting the flow-through that contained the tag-free samples. For further studies RctB and RctBIV were extensively dialyzed against 50 mM Hepes pH 7.5, 200 mM NaCl, 2 mM MgCl2 and 1 mM TCEP.
+
+RctBIV Analytical Size Exclusion Chromatography (SEC)
+
+Analytical SEC of RctBIV was carried out on a Superdex 200 10/300 column (Citiva) equilibrated in 50 mM Hepes, 200mM NaCl, 2 mM MgCl2 and 1 mM TCEP. The flow rate was 1 ml/min and the injection volume 0.5 ml. The calibration of the column was performed by running the Biorad gel filtration standards (Cat.
+No. 151-1901), containing a mixture of bovine thyroglobulin (670 kDa), bovine γ-globulin (158 kDa), chicken ovalbumin (44 kDa), equine myoglobin (17 kDa) and vitamin B12 (1.35 kDa).
+
+RctBIV crystallization and structure determination
+The screening of crystallization conditions of RctBIV was carried out using the sitting-drop vapor-diffusion method with highly pure samples of RctBIV at a concentration of 8 mg mL⁻¹. The drops were set up in Swiss (MRC) 96-well two-drop UVP sitting-drop plates using the Mosquito HTS system (TTP Labtech). Drops of 0.1 μL protein and 0.1 μL precipitant solution were equilibrated to 80 μL precipitant solution in the reservoir. Commercially available screens PACT premier, LMB and SG1 (Molecular Dimensions) were used to test crystallization conditions. The condition resulting in protein crystals (PACTpr screen position E3: 200 mM NaI, 20 % PEG 3350) were repeated as 2 μL drops. Crystals were harvested using suitable cryo-protecting solutions (consisting of the mother liquor supplemented with 20% to 25% of glycerol) and vitrified in liquid N₂ for transport and storage before X-ray exposure.
+
+X-ray diffraction data was collected at the SOLEIL synchrotron (Gif-sur-Yvette, Paris, France) on the Proxima 1 (PX1) beamlines using an Eiger-X 16M detector. Native crystals typically diffracted to 1.8 Å – 2.0 Å resolution. We screened a collection of heavy atoms for experimental phasing and observed that the combination of quick soaks (under 30s) with NaBr at 1.5M, and longer soaks (20 min to 30 min) in AgNO₃ (10 mM) and GdCl₃ (10 mM) provided sufficient isomorphous signal for experimental phasing. For this, data were collected at the peak and inflexion point of the Br edge and Ag and Gd K-edge. After phase improvement with Pirate and density modification with Parrot, the maps were used initially used in Buccaneer for model building and then as starting point for the MR-Rosetta ⁵⁰ suit which completed one third of the structure. Further cycles of automated model building with AutoBuild from the Phenix package ⁵¹ extended the model to 90%. After several iterations of manual building with Coot ⁵² and maximum likelihood refinement as implemented in Buster/TNT ⁵³, the model was extended to cover all the residues (R/Rfree of 19.2%/26.7 %). Supplementary Table 5 details all the X-ray data collection and refinement statistics.
+
+Single particle cryo-electron microscopy (cryo-EM)
+For the cryo-EM experiments RctB was purified in 50 mM Hepes, 200mM NaCl, 2 mM MgCl₂ and 1 mM TCEP. RctB (3 μl) was applied to gold Quantifoil grids (UltrAuFoil 1.2/1.3 Au 300) under 100 % relative humidity at 10 °C at 0.25 mg/ml. Two grids were blotted for 2 and 3 seconds, respectively, and plunged into liquid ethane using a Vitrobot IV (ThermoFisher Scientific). Particles were imaged from the aforementioned grids on a Glacios microscope (ThermoFisher Scientific) operating at 200 kV with a Falcon 4i direct electron detector. In total, 3800 micrographs, with 0.96 Å/pixel, were collected using a defocus range from -1 μm to -3 μm and total dose of 50 e⁻/Ų. The micrographs movies were motion
+corrected using MotionCor2 \(^{54}\), and contrast transfer function, CTF, were estimated using CTFFind-4.1 \(^{55}\). Particles were picked using Topaz \(^{56}\) and subjected to reference free 2D classification using cryoSPARC \(^{57}\). Several rounds of 2D classifications rendered a total of 133108 particles.
+
+Live-cell Fluorescence Microscopy
+The genes encoding for the CFP-ParB\(_{p1}\), yGFP-ParB\(_{pMT1}\) and LacI-RFP-T fluorescent DNA binding proteins were inserted into the *V. cholerae* chromosome \(^{35,58,59}\). Their cognate binding sites were inserted near *ori1* (*parS\(_{p1}\)*), near VC783 (*parS\(_{pMT1}\)*), and near *ori2* (*lacO* array). For microscopy observations, cultures were prepared following the procedures described in \(^{12}\). Initially, strains were streaked onto MH plates from a -80°C stock and then grown overnight in MH rich liquid media. The following day, cultures were diluted at a ratio of 1:1000 in M9 with 1% fructose and grown to exponential phase (OD\(_{600} = 0.2\)). A 2\( \mu \)L aliquot of this culture was spotted on an agar pad (M9, 1% fructose, 1% agarose), for microscopy observation. Cell imaging was performed using an Axio Observer 7 inverted videomicroscope (Zeiss). Analysis of the acquired snapshots was conducted using Fiji software with the MicrobeJ plugin \(^{60}\).
+
+Bacterial Two Hybrid
+BTH101 cells were co-transformed with plasmids encoding T18 and T25 fused to RctB variants following the method described in \(^{33,61}\) and plated on LB agar plates containing kanamycin (25\( \mu \)g/mL) and carbenicillin (100\( \mu \)g/mL). Approximatively 500 co-transformants were pooled, resuspended in PBS, diluted to OD\(_{600}=1\) and spotted (10\( \mu \)L) onto LB agar plates containing kanamycin (25\( \mu \)g/mL), carbenicillin (100\( \mu \)g/mL), Xgal (40\( \mu \)g/mL) and IPTG (0.5mM). Plates were incubated for 48h at 30°C, followed by an additional 24h at 4°C. Blue spots indicate an interaction between the tested proteins and white spots indicate no interaction. Empty pKT25 and pUT18C vectors were used as negative control, and pKT25T-Zip/pUT18C-Zip pair was used as positive control.
+
+ACKNOWLEDGEMENTS
+We acknowledge Olivier Espeli for his critical discussions on the project and valuable advice, which were instrumental in resolving numerous issues. We would like to express our gratitude to Jakub Czarnecki, Julia Bos, Zeynep Baharoglu and Dhruba Chattoraj for their engaging and insightful discussions. For technical assistance and help with the analysis of the ChIP-seq data we are grateful to Jean-Yves Bouet (CBI), Stéphane Duigou, Christophe Possoz, F.-X. Barre (I2BC), Thomas Cokelaer, and Etienne Kornobis from the Bioinformatics and Biostatistics HUB (IP), as well as Marc Monot and Juliana Pipoli Da Fonseca from the Biomics Sequencing Platform (IP). We thank Tristan Piolot and Julien Dumont from the Orion technical core (CIRB) for their support with the widefield microscope. We thank Matthijn Vos, Eduard Baquero Salazar and Stéphane Tachon for their support at the Nanoimaging Core facility (IP). We thank
+Ankur Dalia, Tove Atlung and Flemming Hansen for generously sharing and/or helping access fluorescent DNA-binding proteins needed for three-color fluorescent microscopy in V. cholerae.
+
+FUNDING
+
+This work was supported by the Institut Pasteur; Institut National de la Santé et de la Recherche Médicale (INSERM); Centre National de la Recherche Scientifique [CNRS-UMR 3525]; French National Research Agency, Jeunes Chercheurs [ANR-19CE12-0001]; Laboratoires d’Excellence [ANR-10-LABX62-IBEID] ; Fondation pour la Recherche Médicale [Equipe FRM EQU202103012569] ; Fonds National de Recherche Scientifique (FNRS J.0065.23F; FNRS-EQP UN.025.19; and PDR T.0090.22 to AGP); ERC (CoG DiStRes, n° 864311 to AGP) and Fonds Jean Brachet and the Fondation Van Buuren (AGP). T.N. was supported by the Ministère de l’Enseignement Supérieur et de la Recherche and [ANR-19-CE12-0001]; F.F. was supported by [ANR-10-LABX-62IBEID] and [ANR-19-CE12-0001].
+FIGURES
+
+Figure 1: Chr2 initiator RctB binds preferentially to inhibitory sites on ori2.
+
+a. RctB ChIP signal plotted on the circular map of V. cholerae N16961 using shinyCircos \(^{62}\) with reference sequences of Chr1 (CP028827.1) and Chr2 (CP028828.1) \(^{63}\), genes annotation (VCNNNNN and VCANNNN) is based on the Heidelberg et al. genome \(^{64}\). The RctB ChIP signal is shown in blue. Significant enrichment peaks are marked with a blue dot with the proximal genes indicated in red (39m-like motifs), green (iteron-like motifs), light blue (crtS) and purple (no similarity to known motifs). b. ChIP peak pattern analysis using the MEME suite \(^{24}\). Two significantly enriched motifs correspond to known RctB binding sequences: iterons (left) and 39m (right). c. pORI2 copy number (pORI2/oriC) in E. coli containing different binding regions of RctB in attTn7 site. d. Chr2 copy number relative to Chr1 (ori2/ori1) in non-replicating V. cholerae. For VC0643 and VC1643 contained in CDS RctB binding sites have been inactivated without perturbing the amino acid sequence, for VC1042-1043 the intergenic region containing RctB has been deleted. e. Expression of VC1042 and VC1803 genes relative to the housekeeping gene gyrA in the N16961 wild-type strain (WT), and MCH2 (\(ΔrctB\)) V. cholerae mutant carrying fused chromosomes \(^{45}\). To report the statistical significance of the results, a Student's t-test was performed, with non-significant differences denoted by "ns" and statistically significant differences denoted by an asterisk (*), with a significance level set at p < 0.05
+a
+
+Chr1
+2.961 kbp
+
+Chr2
+1.072 kbp
+
+b
+
+iteron-like
+39m-like
+
+c
+
+pORI2 copy number
+VC1042-VC1043
+crIS
+VC1803
+VC0643
+VC1624
+VC1643
+VC1757
+VC2373
+
+d
+
+Chr2 copy number
+VC1042-aph-VC1043
+VC0643 inactivated
+VC1643 inactivated
+
+e
+
+VC1042 relative expression
+wt
+ΔrctB
+
+VC1803 relative expression
+wt
+ΔrctB
+
+Figure 1
+Figure 2 : RctB predominantly binds to 29/39m sites, forming a nucleoprotein complex that introduces DNA loops into ori2 preventing binding to iterons and DUE.
+
+a. ChIP-seq of RctB in V. cholerae N16961 focused on ori2. The y-axis represents the normalized ChIP signal of RctB (IP/input coverage), the x-axis displays a 2Kbp window centered on ori2, with genomic coordinates of CP028828.1. The genetic context and RctB binding sites are depicted above the plot (Supplementary Fig. 1b for more details on ori2). b. TEM observations of nucleoprotein complexes formed by RctB binding to an un-methylated ori2 DNA (1864bp). DNA-bound RctB forms loops within ori2 connecting the 29m and 39m sites (white arrows). c. ChIP-seq of RctB on ori2 in a mutant having a point mutation in the 29m site (TTGGAACATATAGTGATATTA[<A>]GGTAAAGTG) preventing RctB binding at this site. Same legend as 2a.
+a
+
+![Normalized coverage IP/input vs genomic coordinates Chr2 (bp) for wt](page_184_67_1092_357.png)
+
+b
+
+RctB + ori2 (unmethylated)
+
+![Transmission electron microscopy images of DNA structures](page_184_420_1092_180.png)
+
+c
+
+![Normalized coverage IP/input vs genomic coordinates Chr2 (bp) for 29m*](page_184_670_1092_357.png)
+
+Figure 2
+Figure 3 : RctB domain IV structure reveals two dimerization interface involved in 29/39m-mediated inhibition. **a.** Crystal structure of the RctB\textsuperscript{IV} dimer. Each monomer is composed of an αβ N-terminal subdomain (dark blue) connected to a four α-helices bundle (light blue) via a hinge region. **b.** Topological representation of RctB\textsuperscript{IV} dimer. The primary dimer interface (blue shadow) is formed by the extension of the central β-sheet of each monomer through β3 and the antiparallel interactions of the stacking α1. **c.** Crystallographic tetramer formed through lattice contacts mediated by the C-terminal subdomain of neighboring dimers. Important residues contributing to the primary interface are shown in dark green. The secondary interface (yellow shadow) with the I625 and L651 functionally relevant residues shown in red. **d.** 2D class averages of RctB obtained from cryo-EM **e.** Bacterial-Two-hybrid of RctB-D314P (DP) vs. RctB domain IV mutants. Empty (negative control), Zip (positive control). **f.** ChIP-seq of the RctB-L651P mutant at *ori*2. Same legend as for Fig. 2a.
+a
+
+C-terminal subdomain
+N-terminal subdomain
+hinge
+primary interface
+90°
+
+b
+
+C-terminal subdomain
+N-terminal subdomain
+hinge
+primary interface
+N-terminal subdomain
+
+c
+
+secondary interface
+1548
+1546
+L651
+1625
+90°
+
+d
+
+dimer
+tetramer
+
+e
+
+<table>
+  <tr>
+    <th></th>
+    <th>Empty</th>
+    <th>Zip</th>
+    <th>D314P</th>
+    <th>D314P</th>
+    <th>D314P</th>
+    <th>D314P</th>
+    <th>D314P</th>
+    <th>D314P</th>
+    <th>D314P</th>
+  </tr>
+  <tr>
+    <th>pKT25</th>
+    <td>Empty</td>
+    <td>Zip</td>
+    <td>D314P</td>
+    <td>D314P</td>
+    <td>D314P</td>
+    <td>D314P</td>
+    <td>D314P</td>
+    <td>D314P</td>
+    <td>D314P</td>
+  </tr>
+  <tr>
+    <th>pUT18C</th>
+    <td>Empty</td>
+    <td>Zip</td>
+    <td>D314P</td>
+    <td>D314P L651P</td>
+    <td>D314P I546G-I548G</td>
+    <td>D314P A565P</td>
+    <td>D314P Δ591-596</td>
+    <td>D314P I625P</td>
+    <td>D314P</td>
+  </tr>
+</table>
+
+f
+
+parA2  39m  39m  6x iteron DUE 29m  rctB* (L651P)
+
+Normalized coverage IP/Input
+genomic coordinates Chr2 (bp)
+
+Figure 3
+Figure 4: Stoichiometric relationship between crtS replication and ori2 initiation.
+
+a. ChIP-seq of RctB at ori2 in a V. cholerae ΔcrtS mutant compared to wt. b. Violin plot showing the distribution of cells with 1, 2, 3, and 4 foci. n=Total number of cells analyzed. c-d. Live fluorescence microscopy in V. cholerae with 1 crtS at the native locus (c) and 2 crtS at the native locus and near ori1 (d). Binding sites for fluorescent proteins were inserted near ori1, VC783 (near crtS) and ori2. The x-axis represents cell length (\( \mu m \)). The y-axis represents the longitudinal position of ori1, VC783, and ori2 foci within the cells relative to the old pole (0 being the old pole and 1 the new pole). The old pole of the cell was defined as the closest pole to one ori1 focus. n=Total number of cells analyzed. p : percentage of cells. (e) MFA of exponentially growing V. cholerae cultures using a corrected reference sequence of Chr1 \(^{12}\). For comparison, MFA from wt (crtS_{wt}) and mutant with relocated crtS near ori1 (crtS_{ori1}) from \(^{12}\) are displayed. Sequencing data from mutant with two crtS sites (crtS_{wt} – crtS_{ori1}) are from this study. Log2 of the number of reads starting at each base (normalized against reads from a stationary phase wt control) is plotted against their relative position on Chr1 (in blue) and Chr2 (in yellow). Positions of ori1 and ori2 are set to 0. Any window containing repeated sequences is omitted; thus, the large gap observed in the right arm of Chr2 consists of filtered repeated sequences. Blue dots (Chr1) and yellow dots (Chr2) indicate the averages of 1000-bp windows; black dots indicate the averages of 10,000-bp windows. Dark blue, light blue, orange, and yellow lines indicate ori1, ter1, ori2, and ter2 numbers of reads, respectively. The dashed black lines indicate the number of reads of the loci where the crtS sites are located.
+a
+
+![Panel showing normalized coverage of input and control samples across genomic coordinates on Chr2](page_0_0_1049_384.png)
+
+b
+
+![Panel comparing fold changes between wild-type and crtS mutant strains](page_0_384_1049_384.png)
+
+c
+
+![Panel showing circular diagrams and scatter plots for ori1 and ori2 in different conditions](page_0_768_523_1024.png)
+
+d
+
+![Panel showing circular diagrams and scatter plots for ori1 (crtSswt) and ori2 in different conditions](page_523_768_523_1024.png)
+
+e
+
+![Panel showing read distribution plots for different conditions](page_0_1792_1583_384.png)
+
+Figure 4
+Figure 5: RctB binding pattern on ori2 shift toward iterons after crtS replication. a. V. cholerae growth curve in M9 glucose (0.2%) casamino acids (0.4%) at 30°C. At OD_{600}=0.5, DL-serine hydroxamate (SHX) was added to 1.5mg/mL final concentration to prevent new rounds of initiation^{38}. After 120min of SHX treatment, cells were washed with fresh M9 medium in order to restart replication. b. ori1/ter1 ratio measurement by dPCR at different time points during SHX treatment and removal. c, d, e, f. ChIP-seq experiments on a synchronized population at different time points after SHX removal (R+15min, R+30min, R+45min, R+60min). Top panel: Marker frequency analysis of input samples, coverage on y-axis, genomic coordinates of Chr1 (CPO28827.1) and Chr2 (CPO28828.1) on x-axis. Curves smoothed with 10kb sliding window. Bottom panel: ChIP-seq RctB at ori2. Same legend as Fig. 2a.
+a
+
+Serine hydroxamate
+Wash
+SHX (1.5mg/mL)
+
+OD600nm
+time (min)
+
+b
+
+SHX (1.5mg/mL)
+Wash
+Ratio on/tea1
+time (min)
+
+c
+Chr1    crtS    Chr2
+coverage Input
+genomic coordinates (kbp)
+Normalized coverage IP/input
+genomic coordinates Chr2
+15min
+Normalized coverage IP/input
+genomic coordinates Chr2
+30min
+Normalized coverage IP/input
+genomic coordinates Chr2
+45min
+Normalized coverage IP/input
+genomic coordinates Chr2
+f
+Chr1    crtS    Chr2
+coverage Input
+genomic coordinates (kbp)
+Normalized coverage IP/input
+genomic coordinates Chr2
+60min
+Normalized coverage IP/input
+genomic coordinates Chr2
+
+Figure 5
+Figure 6: Model for the Cell-Cycle-Integrated Control of Chr2 Replication in V. cholerae
+
+1. Prior crtS replication, ori2 is sequestered by an inhibition complex, formed by RctB multimers bridging 29/39m sites. RctB oligomerization domain IV is involved in the formation of this nucleoprotein complex.
+
+2. Upon crtS replication, we hypothesize that the second copy of crtS destabilizes the inhibition complex, possibly at the domain IV secondary interface. 3. Ori2 is thus released and the iteron array becomes accessible for RctB binding. This leads to the opening of the DUE, as shown \(^{18}\).
+1. Prior crtS replication
+Inhibition complex at ori2
+RctB multimerization through domain IV primary and secondary interfaces
+
+Chr1
+Chr2
+activatory site (iteron) | ori2
+inhibitory site (29/39m)
+triggering site (crtS)
+
+RctB
+Full length    domain IV    Multimerization forms
+
+2. crtS is replicated
+Destabilization of the Inhibition complex
+
+3. Chr2 initiation
+ori2 release
+RctB binding to activating sites
+DNA unwinding
+
+Figure 6
+REFERENCES
+
+1 Reyes-Lamothe, R. & Sherratt, D. J. The bacterial cell cycle, chromosome inheritance and cell growth. Nature reviews. Microbiology **17**, 467-478, doi:10.1038/s41579-019-0212-7 (2019).
+2 Beaufay, F., Coppine, J. & Hallez, R. When the metabolism meets the cell cycle in bacteria. *Current opinion in microbiology* **60**, 104-113, doi:10.1016/j.mib.2021.02.006 (2021).
+3 Fournes, F., Val, M. E., Skovgaard, O. & Mazel, D. Replicate Once Per Cell Cycle: Replication Control of Secondary Chromosomes. *Frontiers in microbiology* **9**, 1833, doi:10.3389/fmicb.2018.01833 (2018).
+4 Du, W. L. *et al.* Orderly Replication and Segregation of the Four Replicons of Burkholderia cenocepacia J2315. *PLoS genetics* **12**, e1006172, doi:10.1371/journal.pgen.1006172 (2016).
+5 Frage, B. *et al.* Spatiotemporal choreography of chromosome and megaplasmids in the Sinorhizobium meliloti cell cycle. *Molecular microbiology* **100**, 808-823, doi:10.1111/mmi.13351 (2016).
+6 Deghelt, M. *et al.* G1-arrested newborn cells are the predominant infectious form of the pathogen Brucella abortus. *Nature communications* **5**, 4366, doi:10.1038/ncomms5366 (2014).
+7 Dubarry, N., Willis, C. R., Ball, G., Lesterlin, C. & Armitage, J. P. In Vivo Imaging of the Segregation of the 2 Chromosomes and the Cell Division Proteins of Rhodobacter sphaeroides Reveals an Unexpected Role for MipZ. *mBio* **10**, doi:10.1128/mBio.02515-18 (2019).
+8 Xie, B. B. *et al.* Evolutionary Trajectory of the Replication Mode of Bacterial Replicons. *mBio* **12**, doi:10.1128/mBio.02745-20 (2021).
+9 Ren, Z. *et al.* Conformation and dynamic interactions of the multipartite genome in Agrobacterium tumefaciens. *Proceedings of the National Academy of Sciences of the United States of America* **119**, doi:10.1073/pnas.2115854119 (2022).
+10 Rasmussen, T., Jensen, R. B. & Skovgaard, O. The two chromosomes of Vibrio cholerae are initiated at different time points in the cell cycle. *The EMBO journal* **26**, 3124-3131, doi:10.1038/sj.emboj.7601747 (2007).
+11 Baek, J. H. & Chattoraj, D. K. Chromosome I controls chromosome II replication in Vibrio cholerae. *PLoS genetics* **10**, e1004184, doi:10.1371/journal.pgen.1004184 (2014).
+12 Val, M. E. *et al.* A checkpoint control orchestrates the replication of the two chromosomes of Vibrio cholerae. *Science advances* **2**, e1501914, doi:10.1126/sciadv.1501914 (2016).
+13 Egan, E. S. & Waldor, M. K. Distinct replication requirements for the two Vibrio cholerae chromosomes. *Cell* **114**, 521-530, doi:10.1016/S0092-8674(03)00611-1 (2003).
+14 Orlova, N. *et al.* The replication initiator of the cholera pathogen's second chromosome shows structural similarity to plasmid initiators. *Nucleic acids research* **45**, 3724-3737, doi:10.1093/nar/gkw1288 (2017).
+15 Jha, J. K., Demarre, G., Venkova-Canova, T. & Chattoraj, D. K. Replication regulation of Vibrio cholerae chromosome II involves initiator binding to the origin both as monomer and as dimer. *Nucleic acids research* **40**, 6026-6038, doi:10.1093/nar/gks260 (2012).
+16 Fournes, F. *et al.* The coordinated replication of Vibrio cholerae's two chromosomes required the acquisition of a unique domain by the RctB initiator. *Nucleic acids research* **49**, 11119-11133, doi:10.1093/nar/gkab903 (2021).
+17 Niault, T., Czarnecki, J., Lambérioux, M., Mazel, D. & Val, M. E. Cell cycle-coordinated maintenance of the Vibrio bipartite genome. *EcoSal Plus* **11**, doi:10.1128/ecosalplus.esp-0008-2022 (2023).
+18 Chatterjee, S., Jha, J. K., Ciaccia, P., Venkova, T. & Chattoraj, D. K. Interactions of replication initiator RctB with single- and double-stranded DNA in origin opening of Vibrio cholerae chromosome 2. *Nucleic acids research*, doi:10.1093/nar/gkaa826 (2020).
+19 Wegrzyn, K. E., Gross, M., Uciechowska, U. & Konieczny, I. Replisome Assembly at Bacterial Chromosomes and Iteron Plasmids. *Frontiers in molecular biosciences* **3**, 39, doi:10.3389/fmolb.2016.00039 (2016).
+Demarre, G. & Chattoraj, D. K. DNA adenine methylation is required to replicate both Vibrio cholerae chromosomes once per cell cycle. PLoS genetics 6, e1000939, doi:10.1371/journal.pgen.1000939 (2010).
+Venkova-Canova, T. & Chattoraj, D. K. Transition from a plasmid to a chromosomal mode of replication entails additional regulators. Proceedings of the National Academy of Sciences of the United States of America 108, 6199-6204, doi:10.1073/pnas.1013244108 (2011).
+Venkova-Canova, T., Saha, A. & Chattoraj, D. K. A 29-mer site regulates transcription of the initiator gene as well as function of the replication origin of Vibrio cholerae chromosome II. Plasmid 67, 102-110, doi:10.1016/j.plasmid.2011.12.009 (2012).
+de Lemos Martins, F., Fournes, F., Mazzuoli, M. V., Mazel, D. & Val, M. E. Vibrio cholerae chromosome 2 copy number is controlled by the methylation-independent binding of its monomeric initiator to the chromosome 1 crtS site. Nucleic acids research 46, 10145-10156, doi:10.1093/nar/gky790 (2018).
+Bailey, T. L., Johnson, J., Grant, C. E. & Noble, W. S. The MEME Suite. Nucleic acids research 43, W39-49, doi:10.1093/nar/gkv416 (2015).
+Menikpurage, I. P., Woo, K. & Mera, P. E. Transcriptional Activity of the Bacterial Replication Initiator DnaA. Frontiers in microbiology 12, 662317, doi:10.3389/fmicb.2021.662317 (2021).
+Venkova-Canova, T., Baek, J. H., Fitzgerald, P. C., Blokesch, M. & Chattoraj, D. K. Evidence for two different regulatory mechanisms linking replication and segregation of vibrio cholerae chromosome II. PLoS genetics 9, e1003579, doi:10.1371/journal.pgen.1003579 (2013).
+Val, M. E. et al. Fuse or die: how to survive the loss of Dam in Vibrio cholerae. Molecular microbiology 91, 665-678, doi:10.1111/mmi.12483 (2014).
+Benureau, Y. et al. Method combining BAC film and positive staining for the characterization of DNA intermediates by dark-field electron microscopy. Biol Methods Protoc 5, bpaa012, doi:10.1093/biomethods/bpaa012 (2020).
+Jha, J. K., Ghirlando, R. & Chattoraj, D. K. Initiator protein dimerization plays a key role in replication control of Vibrio cholerae chromosome 2. Nucleic acids research 42, 10538-10549, doi:10.1093/nar/gku771 (2014).
+Jha, J. K. et al. The DnaK Chaperone Uses Different Mechanisms To Promote and Inhibit Replication of Vibrio cholerae Chromosome 2. mBio 8, doi:10.1128/mBio.00427-17 (2017).
+Holm, L., Laiho, A., Toronen, P. & Salgado, M. DALI shines a light on remote homologs: One hundred discoveries. Protein science : a publication of the Protein Society 32, e4519, doi:10.1002/pro.4519 (2023).
+Das, D. et al. The structure of the first representative of Pfam family PF09836 reveals a two-domain organization and suggests involvement in transcriptional regulation. Acta Crystallogr Sect F Struct Biol Cryst Commun 66, 1174-1181, doi:10.1107/S1744309109022672 (2010).
+Karimova, G., Pidoux, J., Ullmann, A. & Ladant, D. A bacterial two-hybrid system based on a reconstituted signal transduction pathway. Proceedings of the National Academy of Sciences of the United States of America 95, 5752-5756, doi:10.1073/pnas.95.10.5752 (1998).
+Ramachandran, R., Ciaccia, P. N., Filsuf, T. A., Jha, J. K. & Chattoraj, D. K. Chromosome 1 licenses chromosome 2 replication in Vibrio cholerae by doubling the crtS gene dosage. PLoS genetics 14, e1007426, doi:10.1371/journal.pgen.1007426 (2018).
+Dalia, A. B. & Dalia, T. N. Spatiotemporal Analysis of DNA Integration during Natural Transformation Reveals a Mode of Nongenetic Inheritance in Bacteria. Cell 179, 1499-1511 e1410, doi:10.1016/j.cell.2019.11.021 (2019).
+Ferullo, D. J., Cooper, D. L., Moore, H. R. & Lovett, S. T. Cell cycle synchronization of Escherichia coli using the stringent response, with fluorescence labeling assays for DNA content and replication. Methods 48, 8-13, doi:10.1016/j.ymeth.2009.02.010 (2009).
+Kemter, F. S. et al. Synchronous termination of replication of the two chromosomes is an evolutionary selected feature in Vibrionaceae. PLoS genetics 14, e1007251, doi:10.1371/journal.pgen.1007251 (2018).
+38 Kemter, F. S. et al. Stringent response leads to continued cell division and a temporal restart of DNA replication after initial shutdown in Vibrio cholerae. Molecular microbiology 111, 1617-1637, doi:10.1111/mmi.14241 (2019).
+39 Molina-Garcia, L. et al. Functional amyloids as inhibitors of plasmid DNA replication. Scientific reports 6, 25425, doi:10.1038/srep25425 (2016).
+40 McEachern, M. J., Bott, M. A., Tooker, P. A. & Helinski, D. R. Negative control of plasmid R6K replication: possible role of intermolecular coupling of replication origins. Proceedings of the National Academy of Sciences of the United States of America 86, 7942-7946, doi:10.1073/pnas.86.20.7942 (1989).
+41 Zzaman, S. & Bastia, D. Oligomeric initiator protein-mediated DNA looping negatively regulates plasmid replication in vitro by preventing origin melting. Molecular cell 20, 833-843, doi:10.1016/j.molcel.2005.10.037 (2005).
+42 Ciaccia, P. N., Ramachandran, R. & Chattoraj, D. K. A Requirement for Global Transcription Factor Lrp in Licensing Replication of Vibrio cholerae Chromosome 2. Frontiers in microbiology 9, 2103, doi:10.3389/fmicb.2018.02103 (2018).
+43 Doan, A. et al. The replication enhancer crtS depends on transcription factor Lrp for modulating binding of initiator RctB to ori2 of Vibrio cholerae. Nucleic acids research, doi:10.1093/nar/gkad1111 (2023).
+44 Le Roux, F., Binesse, J., Saulnier, D. & Mazel, D. Construction of a Vibrio splendidus mutant lacking the metalloprotease gene vsm by use of a novel counterselectable suicide vector. Applied and environmental microbiology 73, 777-784, doi:10.1128/AEM.02147-06 (2007).
+45 Val, M. E., Skovgaard, O., Ducos-Galand, M., Bland, M. J. & Mazel, D. Genome engineering in Vibrio cholerae: a feasible approach to address biological issues. PLoS genetics 8, e1002472, doi:10.1371/journal.pgen.1002472 (2012).
+46 Gibson, D. G. et al. Enzymatic assembly of DNA molecules up to several hundred kilobases. Nature methods 6, 343-345, doi:10.1038/nmeth.1318 (2009).
+47 Diaz, R. E., Sanchez, A., Anton Le Berre, V. & Bouet, J. Y. High-Resolution Chromatin Immunoprecipitation: ChIP-Sequencing. Methods in molecular biology 1624, 61-73, doi:10.1007/978-1-4939-7098-8_6 (2017).
+48 Madic, J. et al. Three-color crystal digital PCR. Biomolecular detection and quantification 10, 34-46, doi:10.1016/j.bdq.2016.10.002 (2016).
+49 Lo Scrudato, M. & Blokesch, M. The regulatory network of natural competence and transformation of Vibrio cholerae. PLoS genetics 8, e1002778, doi:10.1371/journal.pgen.1002778 (2012).
+50 Terwilliger, T. C. et al. phenix.mr_rosetta: molecular replacement and model rebuilding with Phenix and Rosetta. J Struct Funct Genomics 13, 81-90, doi:10.1007/s10969-012-9129-3 (2012).
+51 Afonine, P. V. et al. Towards automated crystallographic structure refinement with phenix.refine. Acta crystallographica. Section D, Biological crystallography 68, 352-367, doi:10.1107/S0907444912001308 (2012).
+52 Emsley, P. & Cowtan, K. Coot: model-building tools for molecular graphics. Acta crystallographica. Section D, Biological crystallography 60, 2126-2132, doi:10.1107/S0907444904019158 (2004).
+53 Smart, O. S. et al. Exploiting structure similarity in refinement: automated NCS and target-structure restraints in BUSTER. Acta crystallographica. Section D, Biological crystallography 68, 368-380, doi:10.1107/S0907444911056058 (2012).
+54 Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nature methods 14, 331-332, doi:10.1038/nmeth.4193 (2017).
+55 Rohou, A. & Grigorieff, N. CTFFIND4: Fast and accurate defocus estimation from electron micrographs. Journal of structural biology 192, 216-221, doi:10.1016/j.jsb.2015.08.008 (2015).
+56 Bepler, T. et al. Positive-unlabeled convolutional neural networks for particle picking in cryo-electron micrographs. Nature methods 16, 1153-1160, doi:10.1038/s41592-019-0575-8 (2019).
+57 Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nature methods **14**, 290-296, doi:10.1038/nmeth.4169 (2017).
+58 David, A. *et al.* The two Cis-acting sites, parS1 and oriC1, contribute to the longitudinal organisation of Vibrio cholerae chromosome I. PLoS genetics **10**, e1004448, doi:10.1371/journal.pgen.1004448 (2014).
+59 Woldringh, C. L., Hansen, F. G., Vischer, N. O. & Atlung, T. Segregation of chromosome arms in growing and non-growing Escherichia coli cells. Frontiers in microbiology **6**, 448, doi:10.3389/fmicb.2015.00448 (2015).
+60 Ducret, A., Quardokus, E. M. & Brun, Y. V. MicrobeJ, a tool for high throughput bacterial cell detection and quantitative analysis. Nature microbiology **1**, 16077, doi:10.1038/nmicrobiol.2016.77 (2016).
+61 Battesti, A. & Bouveret, E. The bacterial two-hybrid system based on adenylate cyclase reconstitution in Escherichia coli. Methods **58**, 325-334, doi:10.1016/j.ymeth.2012.07.018 (2012).
+62 Yu, Y., Ouyang, Y. & Yao, W. shinyCircos: an R/Shiny application for interactive creation of Circos plot. Bioinformatics **34**, 1229-1231, doi:10.1093/bioinformatics/btx763 (2018).
+63 Matthey, N., Drebes Dorr, N. C. & Blokesch, M. Long-Read-Based Genome Sequences of Pandemic and Environmental Vibrio cholerae Strains. Microbiology resource announcements **7**, doi:10.1128/MRA.01574-18 (2018).
+64 Heidelberg, J. F. *et al.* DNA sequence of both chromosomes of the cholera pathogen Vibrio cholerae. Nature **406**, 477-483, doi:10.1038/35020000 (2000).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• Supplementarymovie1.mp4
+• Supplementarymovie2.mp4
+• Supplementarymovie3.mp4
+• NiaultSupplInformationFigures.pdf

0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/peer_review/peer_review.md

@@ -0,0 +1,466 @@
+Peer Review File
+
+A minority of final stacks yields superior amplitude in single-particle cryo-EM
+
+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/.
+Reviewers' Comments:
+
+Reviewer #1:
+Remarks to the Author:
+The manuscript "Not final yet: a minority of final stacks yields superior amplitude in single-particle cryo-EM" from Zhu et al approaches the important question in cryo-EM of the number of images required to get high resolution structures for a protein of a given size, for which practical experience and theoretical limits seem at odds with each other. However, the paper in it's current state could be improved to make the method clearer and the impact of the results more significant.
+
+First of all, from the initial description it is not abundantly clear what the CryoSieve algorithm is. First it seems to be reconstruction algorithm independent, as it indicates that it uses 'a cryo-EM single-particle reconstruction software selected by the user'. While this is nice, it begs the reader to ask if different reconstruction software produce different results with this algorithm. Furthermore it leads one to wonder if the iterations must be done by hand or if CryoSieve has been setup to work automatically with one software or another (cryoSPARC, Relion, or cisTEM perhaps). The existence of such a pipeline would be made somewhat clearer with the sharing of the software itself which it has been declared "will be open-source upon publication and is also available upon request during the review process".
+
+The second aspect of the algorithm which is very unclear is despite a mathematical formalism preseented for the cryoSieve score, it is very unclear how various important parameters within this score are chosen per dataset and per iteration within each dataset. Two clear examples are 1) the choice of high pass filter per iteration and 2) the score for which particles are chosen to be retained or dropped. These seem to vary as a function of dataset, as described in the methods, and it is unclear if this has been done in a systematic way. These choices also make it very unclear how this method can be compared to the other methods (random, NCC, AGC, and cisTEM). I think the algorithm would be much clearer if these parameter choices per iteration were explicitly described within the mathematical formalism presented, and if the other methods (especially the 'random' method to which a clear comparison can be made) can also be described within this same formalism so that it is much clearer what is actually being done. In this light, the cisTEM method has been clearly separated from the NCC and AGC methods, and I think keeping them together would improve the currency of the arguments of this paper.
+
+Another aspect of this algorithm that is not clear is how it relates to sieve-based strategies in machine learning, or if indeed has no relation to these, and the name was not chosen in relation to other sieve-based algorithms. For example see "Universal sieve-based strategies for efficient estimation using machine learning tools" arXiv:2003.01856v2 from Qiu, Luedtke, and Carone (2020).
+
+Despite the lack of clarity in the description of the algorithm - the results of the CryoSieve method presented in this paper are enlightening - that similar resolution data can be achieved with 26.2%-32.8% of the data is quite interesting. But what the consequences of this result are could be discussed in much greater detail. The theoretical experiments show that CryoSieve is dropping particles with increased radiation damage (according to one model of simulated radiation damage), which is interesting - but is this the only aspect of the images the method is picking up on? Because the CryoSieve score depends so critically on the high pass filter - are there lessons to be learned regarding how to pick particles or collect data so that we can actually increase our resolution with the same number of particles in the future? Or even guidance for experimentalists for how to collect fewer of these 'futile particles'?
+
+It is impressive to see the comparisons of the CryoSieve results to the theoretical limits, however the language in the text needs clarification 'the TRPA1 dataset fell short by approximately 52' should be '52-fold' I believe. Otherwise this is quite misleading. That the pfCRT dataset matches the theoretical limit very closely is very exciting and it would be worthwhile to hear more discussion about why the
+authors think this dataset outperforms the results of the others and how one might achieve similarly good results for the other datasets either through further optimization of cryoSieve or changes in the data collection procedure (or some third approach).
+
+Overall, while this manuscript shows valuable results within the context of understanding which particles aren't contributing information to high resolution reconstructions of molecules in single particle cryoEM, I believe it would need some significant clarifications and improvements before it can be accepted for publication by this journal.
+
+Minor points:
+
+It is advisable to cite EMPIAR: Iudin A, Korir PK, Somasundharam S, Weyand S, Cattavitello C, Fonseca N, Salih O, Kleywegt GJ, Patwardhan A (2023). "EMPIAR: the Electron Microscopy Public Image Archive." Nucleic Acids Res., 51, D1503-D1511. https://doi.org/10.1093/nar/gkac1062.
+
+In section 2.1 the sentence "We have demonstrated that the CryoSieve score can identify particles with incorrect pose parameters or components in the high-frequency range through theoretical analysis and simulation verification." should include references to the sections which do these theoretical analyses and simulation verifications.
+
+In Figure 1 - it seems like a substantial coincidence that four datasets use 26.2% of the data while two datasets use 32.8% of the data to achieve the final high resolution results while culling particles with the CryoSieve method - is this an artifact of the way iterations are chosen or a typo? If it is an artifact of the method it would be worthwhile to clarify this with the more in depth description of the method requested above.
+
+Reviewer #2:
+Remarks to the Author:
+The manuscript by Zhu, et al. describes their work using high-frequency signals to sort particles for finding a minimal “finest subset” for 3D reconstruction in cryo-EM. The work aims to address a crucial question in cryo-EM: that is, how to get the minimum number of particles required to reach a specific resolution. To do so, they developed a CryoSieve procedure and applied it to six EMPIAR data sets with resolutions ranging from 4.11 to 3.04 Å. They show that with the CryoSieve procedure, they can identify and remove the majority of particles while maintaining a sufficient number of particles to reach a similar or higher resolution as evaluated by FSC curves and Q scores. The work is interesting in the sense that it provides a practical way to sort particles. With the six data sets selected, their technique is sound, and the results support their main claim. However, as a new method, it must be validated using a wide range of data sets to show its broad applicability. This is particularly true because many data sets can be downloaded from EMPIAR. Below are my comments for the authors to consider.
+
+Major:
+1. The authors developed a CryoSieve procedure for particle sorting to further remove particles after consensus refinement. They selected six EMAPIR data sets with resolutions between 4.11 and 3.04 Å. To show its broader applicability, the authors should expand their tests on additional data sets. 1) They should include data sets better than 3 Å, for example selecting data sets between 2-3 Å, and data sets better than 2.0 Å. As the resolution of cryo-EM has reached atomic resolutions at about 1.2 Å, it would be interesting to see if CryoSieve works on these very high-quality datasets. The analysis will help better understand the performance of CryoSieve. 2) The signal over noise of particles depends on the particle size. To evaluate the quality of the work and performance of CryoSieve, it is suggested to evaluate the performance of CryoSieve in dealing with particles of different sizes, i.e. molecular weights as kDa. Cryo-EM can allow structure determination of particles with a molecular
+weight of about 50 kDa or lower. The authors should evaluate the performance of CryoSieve on such small particles and add molecular weight to Tables 1 or 2.
+
+2. One challenge in cryo-EM is data heterogenicity. In solution, macromolecules are equilibrium in many conformational states which are captured during the vitrification process. Cryo-EM data analysis is essentially a triage process to filter out conformational states and radiation damage. The author claims that CryoSieve can remove radiation-damaged particles. How can they exclude the possibility that the particles they removed could be particles belonging to minor conformational states which are slightly different from the consensus model? The authors used simulated particles to show the effectiveness of CryoSieve in removing radiation-damaged particles. They need to demonstrate the effectiveness using experimental data.
+
+Minor:
+1. P. 3, line 89: The authors used “high-resolution amplitude” for sorting particles in Fourier space. Have the authors sorted particles based on the high-resolution phase? It would be interesting to compare phase-based sorting with amplitude-based sorting.
+
+2. Page 6, lines 151-156. The authors describe the 2D and 3D classification work used in reference 26. Such a statement does not bring in new information here and should be deleted.
+
+3. Page 6, lines 164-166. The authors claim that CryoSieve can remove over half of the particles with unreliable high-frequency signals without negatively affecting the final reconstruction. However, it’s not clear what’s the criterion/threshold to define “unreliable high-frequency signals”. Besides, after removing the “unreliable high-frequency signals”, have the authors observed improved cryo-EM densities or structural features that were blurred or missing in the published maps?
+
+4. Page 6, lines 168-171. The authors compared CryoSieve with two other sorting methods of NCC and AGC. The authors should discuss in more detail why their method is better than the other two. Did the authors observe the preferred orientation issue while sorting particles based on high-frequency signals? In cryo-EM, nonalignment classification is routine and effective for the classification of heterogeneous data. The nonalignment classification can sort and remove particles, in the meanwhile can identify additional conformational states. The authors should compare the performance of CryoSeieve with the nonalignment classification in terms of removing particles while maintaining the resolution.
+
+5. Page 6, lines 181-183. The authors used “Einstein-from-noise” to justify the removal of the deposited Euler angles. This statement is not accurate because there is no evidence that the published reconstructions/Eular angles suffered from the “Einstein-from-noise” issue. The authors should revise the sentence to say “to remove bias in the published maps”.
+
+6. Page 6, lines 185-187. The authors should include all metrics, in addition to FSC and Q-score, that they have used to evaluate the maps before and after the CryoSieve procedure. As B factors are important for evaluating data quality, they should plot B factors with respect to the number of interactions in Figure 2 and Supplementary Figure 2.
+
+7. Page 7, Figure 1 caption: Did the authors apply the same B-factor to sharpen the maps before and after CryoSieve? In addition to the sharpened maps, the authors may compare the non-sharpened maps in a Supplementary Figure. For a better comparison, they should include the contour levels that were used to draw maps before and after CryoSeive.
+
+8. Page 7, lines 194-195. If cisTEM reports a per-particle score, the authors should explain why the score can’t be used as a particle sorting criterion.
+
+9. Page 8, lines 211-214. Why did CryoSieve remove a substantial number of high-resolution 2D
+particles in TRPA1, but not in the other five data sets (Figure 2)? The authors should perform 2D class averaging on additional data sets (see major #1).
+
+10. Page 8, section 2.4: Using simulated data, the authors claim that CryoSieve can effectively detect radiation-damaged particles better than NCC and cisTEM. The authors should also compare CryoSieve performance with the AGC method and nonalignment classification method. In addition, they should use experimental data, not just simulated data to show its effectiveness in the treatment of experimental radiation damage.
+
+11. Page 9, Table 2: For B-factor calculation, the authors should use the Rosenthal and Henderson’s B-factor method instead of values from the cryoSPARC auto-processing.
+
+12. Page 11, lines 330-345. The discussion on sample preparation is off-topic to the work and should be removed or revised in the context of CryoSieve.
+
+13. The authors should have shared their code as an attachment for a better evaluation of the work.
+Authors’ Response to Reviews of
+
+Not final yet: a minority of final stacks yields superior amplitude in single-particle cryo-EM
+
+Jianying Zhu, Qi Zhang, Hui Zhang, Zuoqiang Shi, Mingxu Hu and Chenglong Bao
+Submitted to Nature Communications, NCOMMS-23-22170-T
+
+RC: Reviewers’ Comment.    AR: Authors’ Response,   □ Manuscript Text
+
+We sincerely thank the valuable suggestions and comments from the reviewers. We list our point-to-point replies in the following context and hope that the revision can address the concerns.
+
+Response to Referee #1
+
+RC: The manuscript "Not final yet: a minority of final stacks yields superior amplitude in single-particle cryo-EM" from Zhu et al approaches the important question in cryo-EM of the number of images required to get high resolution structures for a protein of a given size, for which practical experience and theoretical limits seem at odds with each other. However, the paper in it’s current state could be improved to make the method clearer and the impact of the results more significant.
+
+AR: Thanks for your support of this paper. We have made substantial changes in the revised version and hope that it can clearly show the significance of CryoSieve.
+
+RC: (1-1) First of all, from the initial description it is not abundantly clear what the CryoSieve algorithm is. First it seems to be reconstruction algorithm independent, as it indicates that it uses ‘a cryo-EM single-particle reconstruction software selected by the user’. While this is nice, it begs the reader to ask if different reconstruction software produce different results with this algorithm.
+
+AR: Thank you for your insightful comments and constructive suggestions. We have added a Supplementary Figure to clarify the performance comparison between using Relion and CryoSPARC in the reconstruction algorithm/module.
+
+Our statement in the initial submission was based on our observation that the reconstructed density outputs from mainstream software were nearly indistinguishable, as evidenced by the high correlations among them (refer to Supplementary Fig. 2b). However, upon further investigation and in light of your valuable feedback, we were surprised to discover that using Relion in combination with CryoSieve yielded significantly superior results compared to CryoSPARC (see Supplementary Fig. 2a). We have accordingly updated our manuscript to reflect this new insight.
+
+The observed performance difference when switching from Relion to CryoSPARC can be clarified as follows: It is a common observation among cryo-EM image processing researchers that the range of reconstructed maps from CryoSPARC differs from those from RELION. Specifically, CryoSPARC applies a multiplication factor to the amplitude of the reconstructed density (see Supplementary Fig. 2c). This action changes the scale between the reconstructed density map and the corresponding particles. Such a modification introduces a scale variation, which significantly impacts the computed CryoSieve score, rendering it less effective. The score is given by:
+
+\[
+g_i = |H b_i|_2^2 - |H(b_i - \tilde{A}_i x)|_2^2.
+\]
+In essence, replacing \( x \) with \( x' = \alpha X \), where \( \alpha \neq 1 \) (notably, in CryoSPARC, \( \alpha \) is much greater than 1), disrupts the score. Also, estimating the scalar \( \alpha \) in CryoSPARC is difficult due to the inaccessibility of the source code.
+
+We have revised our manuscript as
+
+Given that \( g_d \) relies on the accurate amplitude of the reconstructed density map \( x^{(k)} \), CryoSPARC is not the optimal choice for reconstruction (Supplementary Figure 2).
+
+RC: (1-2) Furthermore it leads one to wonder if the iterations must be done by hand or if CryoSieve has been setup to work automatically with one software or another (CryoSPARC, Relion, or cisTEM perhaps). The existence of such a pipeline would be made somewhat clearer with the sharing of the software itself which it has been declared "will be open-source upon publication and is also available upon request during the review process".
+
+AR: Thank you for your comment. We would like to clarify that users are not required to perform iterations manually. CryoSieve automates this process, with each iteration encompassing both reconstruction and sieving. The software takes the path of the reconstruction module from RELION as an input option and utilizes it in the iterative sieving process. This procedure is illustrated in a flow chart, available in Supplementary Figure 3. Moreover, CryoSieve is now open-sourced and available on GitHub https://github.com/mxhulab/cryosieve. A comprehensive user guide can be found on the project’s homepage, providing detailed instructions and assistance for users. We have revised our manuscript as
+
+A flow chart scheme is provided in Supplementary Figure 3.
+
+and
+
+CryoSieve is now open-sourced and available on GitHub (https://github.com/mxhulab/cryosieve). A detailed tutorial can also be found on its homepage. Moreover, datasets used in this manuscript, along with the expected outputs after running CryoSieve, have been deposited on GitHub and can be accessed via CryoSieve’s homepage.
+
+RC: (2) The second aspect of the algorithm which is very unclear is despite a mathematical formalism presented for the cryoSieve score, it is very unclear how various important parameters within this score are chosen per dataset and per iteration within each dataset. Two clear examples are 1) the choice of high pass filter per iteration and 2) the score for which particles are chosen to be retained or dropped. These seem to vary as a function of dataset, as described in the methods, and it is unclear if this has been done in a systematic way. These choices also make it very unclear how this method can be compared to the other methods (random, NCC, AGC, and cisTEM). I think the algorithm would be much clearer if these parameter choices per iteration were explicitly described within the mathematical formalism presented, and if the other methods (especially the 'random' method to which a clear comparison can be made) can also be described within this same formalism so that it is much clearer what is actually being done.
+
+AR: Thanks for your comment. We have given the parameter settings for CryoSieve and the other comparative algorithms in Supplementary Material VI. For CryoSieve, the high-pass cutoff frequency increases linearly across iterations. Additionally, for CryoSieve, NCC, cisTEM, and random, a fixed retention ratio of 80% was maintained in our experiments.
+
+We have revised our manuscript as
+The high-pass cutoff frequency of CryoSieve increases linearly across iterations.
+
+The parameter settings for CryoSieve and the other comparative algorithms were listed in Supplementary Material VI.
+
+along with adding Supplementary Material VI.
+
+RC: (3) In this light, the cisTEM method has been clearly separated from the NCC and AGC methods, and I think keeping them together would improve the currency of the arguments of this paper.
+
+AR: Thanks for your suggestion. We have integrated the cisTEM results from the supplementary section and combined them with results from other algorithms, namely CryoSieve, NCC, AGC, and non-alignment classification, using random as the baseline for comparison. It is important to emphasize that while cisTEM can report a score for each individual particle image, this score is provided after its 3D refinement. The pose parameters of the particles undergo re-estimation or refinement during cisTEM’s 3D refinement process. Given the differences in alignment and other image processing workflows between cisTEM and CryoSPARC, a direct comparison between cisTEM and CryoSieve may not be fair.
+
+RC: (4) Another aspect of this algorithm that is not clear is how it relates to sieve-based strategies in machine learning, or if indeed has no relation to these, and the name was not chosen in relation to other sieve-based algorithms. For example see "Universal sieve-based strategies for efficient estimation using machine learning tools" arXiv:2003.01856v2 from Qiu, Luedtke, and Carone (2020).
+
+AR: Thank you for your comment. CryoSieve is a software designed to filter out non-essential particles by leveraging high-frequency distance. However, its method does not directly relate to the sieve-based estimations discussed in the suggested paper. In that context, a sieve estimator employs a sequence of simpler models (the sieves) to approximate a complex model. While both utilize the concept of a ‘sieve,’ the application and methodology differ significantly.
+
+RC: (5-1) Despite the lack of clarity in the description of the algorithm - the results of the CryoSieve method presented in this paper are enlightening - that similar resolution data can be achieved with 26.2%-32.8% of the data is quite interesting.
+
+AR: Thank you for your support.
+
+RC: (5-2) But what the consequences of this result are could be discussed in much greater detail. The theoretical experiments show that CryoSieve is dropping particles with increased radiation damage (according to one model of simulated radiation damage), which is interesting - but is this the only aspect of the images the method is picking up on?
+
+AR: Thank you for your suggestion. We have simulated orientation, translation and CTF parameter errors in the TRMP8 dataset, and found out that CryoSieve is capable of efficiently removing particles achieving a high accuracy of over 90%. We organized the result as Supplementary Material III. However, non-alignment classification seems to achieve comparable accuracy in cases of the simulated orientation, translation and CTF parameter error (also in Supplementary Material III). Therefore, these type of errors are unlikely to present in the final stacks.
+
+Generally, we cannot definitively determine the full range of image features that CryoSieve identifies. Current experimental evidence indicates that CryoSieve preferentially eliminates particles experiencing radiation damage. However, it’s unlikely that radiation-damaged particles account for all the eliminations. This question might be better addressed by analyzing their coordinates in the micrographs considering not only the X-Y
+axis but the Z-axis as well. Factors such as air-water interference or charging effects could play roles. This is an intriguing area that warrants further investigation.
+
+We have revised our manuscript as
+
+It’s worth noting that CryoSieve can efficiently remove particles with incorrect pose and CTF parameter estimations, achieving a high accuracy of over 90% (Supplementary Material III). However, These particles are also removed by the non-alignment classification approach (Supplementary Material III), making them unlikely to be present in the final stacks.
+
+and add Material III in Supplementary.
+
+RC: (5-3) Because the CryoSieve score depends so critically on the high pass filter - are there lessons to be learned regarding how to pick particles or collect data so that we can actually increase our resolution with the same number of particles in the future? Or even guidance for experimenters for how to collect fewer of these 'futile particles'?
+
+AR: Current particle-picking strategies, encompassing both template-based and deep-learning-based methods, predominantly rely on the low-frequency information of the target biological macromolecule. Given that the CryoSieve score is contingent on the high-pass filter, this highlights the value of maintaining a high retention level in the high-frequency range when evaluating particle quality. While attempts aim to assess particle quality during the data acquisition phase, minimizing the collection of these "futile particles" remains a persistent, unresolved challenge in the cryo-EM field, necessitating further research.
+
+An additional insight derived from CryoSieve suggests substantial potential for improving sample preparation techniques to reduce the proportion of futile particles. Following the submission of this paper, we collaborated with Professor Hongwei Wang, a renowned expert in sample separation. Our collaboration revealed that the conditions during sample preparation play a pivotal role in limiting the presence of futile particles.
+
+RC: (6) It is impressive to see the comparisons of the CryoSieve results to the theoretical limits, however the language in the text needs clarification 'the TRPA1 dataset fell short by approximately 52' should be '52-fold' I believe. Otherwise this is quite misleading.
+
+AR: Thank you for your support and pointing that out. We have revised it.
+
+RC: (7) That the pfCRT dataset matches the theoretical limit very closely is very exciting and it would be worthwhile to hear more discussion about why the authors think this dataset outperforms the results of the others and how one might achieve similarly good results for the other datasets either through further optimization of cryoSieve or changes in the data collection procedure (or some third approach).
+
+AR: Thank you for your feedback. It is excited to see that pfCRT is nearing the theoretical limit. Notably, during its data collection, a Cs corrector and energy filter were used in tandem. Only a few facilities possess both these devices, and it’s possible that the pfCRT dataset benefited from this combination. However, whether the advancements in the Cs corrector, energy filter, and sample preparation contribute to obtaining datasets closer to the theoretical limit and to what extent they help still largely requires further research.
+
+Given that this matter requires further investigation and is beyond the scope of this work, we choose to avoid making strong claims about how to reach the theoretical limit in this manuscript.
+
+RC: (8) Overall, while this manuscript shows valuable results within the context of understanding which particles aren’t contributing information to high resolution reconstructions of molecules in single particle cryoEM, I believe it would need some significant clarifications and improvements before it can be accepted
+for publication by this journal.
+
+AR: Thank you for your valuable comments. We hope the revised manuscript can address your concerns.
+
+RC: (9) It is advisable to cite EMPIAR: Iudin A, Korir PK, Somasundharam S, Weyand S, Cattavitello C, Fonseca N, Salih O, Kleywegt GJ, Patwardhan A (2023). “EMPIAR: the Electron Microscopy Public Image Archive.” Nucleic Acids Res., 51, D1503-D1511. https://doi.org/10.1093/nar/gkac1062.
+
+AR: Thank you for your suggestion. We added this citation.
+
+RC: (10) In section 2.1 the sentence "We have demonstrated that the CryoSieve score can identify particles with incorrect pose parameters or components in the high-frequency range through theoretical analysis and simulation verification." should include references to the sections which do these theoretical analyses and simulation verifications.
+
+AR: We have added the reference section in the revision. In Supplementary Material I and III, we carry out two types of analysis:
+
+• Assuming that noise in particles follows a Gaussian distribution, we have shown that, with high probability, the CryoSieve score is an ideal indicator of particle image quality, distinguishing it from typical cryo-EM damage or artifact, including high-frequency random phasing and inaccurate estimation of imaging parameters such as rotation angle, in-plane translation, and CTF parameters.
+
+• When we use simulated datasets for particle sieving, CryoSieve score exhibits remarkable accuracy in removing particles with incorrect pose and CTF parameter estimations, achieving a precision rate of over 90%.
+
+We have revised our manuscript as
+
+<table>
+  <tr>
+    <td>We have demonstrated that the CryoSieve score can identify particles with incorrect pose parameters or components in the high-frequency range through theoretical analysis and simulation verification. Assuming that noise in particles follows a Gaussian distribution, we have shown that, with high probability, the CryoSieve score is an ideal indicator of particle image quality, distinguishing it from typical cryo-EM damage or artifacts (Supplementary Material I). Furthermore, when simulating radiation damage as high-frequency random phasing, the CryoSieve score exhibits remarkable accuracy in selecting particles even with a very low signal-to-noise ratio (approximately 0.001), achieving a precision rate of around 90% (Supplementary Material I).</td>
+  </tr>
+  <tr>
+    <td>We have demonstrated that the CryoSieve score can identify particles with incorrect pose parameters or components in the high-frequency range through theoretical analysis and simulation verification (Supplementary Material I and III). Specifically, assuming that noise in particles follows a Gaussian distribution, we have shown that, with high probability, the CryoSieve score is an ideal indicator of particle image quality, distinguishing it from typical cryo-EM damage or artifacts (Supplementary Material I). Furthermore, the CryoSieve score exhibits remarkable accuracy in removing particles with incorrect pose and CTF parameter estimations, achieving a high accuracy of over 90% (Supplementary Material III).</td>
+  </tr>
+</table>
+
+RC: (11) In Figure 1 - it seems like a substantial coincidence that four datasets use 26.2% of the data while two datasets use 32.8% of the data to achieve the final high resolution results while culling particles with the CryoSieve method - is this an artifact of the way iterations are chosen or a typo? If it is an artifact of the
+method it would be worthwhile to clarify this with the more in depth description of the method requested above.
+
+AR: In all experiments, we have set the retention ratio for each iteration, a hyperparameter of CryoSieve, to 80%. If the finest subset is in iteration 5, its ratio would be \(0.8^5 = 32.8\%\). If the finest subset is in iteration 6, the ratio becomes \(0.8^6 = 26.2\%\). The iteration in which the finest subset appears is determined based on comprehensive metrics, including FSC, Q-score and Rosenthal-Henderson B-factor, as illustrated in Figure 2.
+Response to Referee #2
+
+RC: The manuscript by Zhu, et al. describes their work using high-frequency signals to sort particles for finding a minimal “finest subset” for 3D reconstruction in cryo-EM. The work aims to address a crucial question in cryo-EM: that is, how to get the minimum number of particles required to reach a specific resolution. To do so, they developed a CryoSieve procedure and applied it to six EMPIAR data sets with resolutions ranging from 4.11 to 3.04 Å. They show that with the CryoSieve procedure, they can identify and remove the majority of particles while maintaining a sufficient number of particles to reach a similar or higher resolution as evaluated by FSC curves and Q scores. The work is interesting in the sense that it provides a practical way to sort particles. With the six data sets selected, their technique is sound, and the results support their main claim. However, as a new method, it must be validated using a wide range of data sets to show its broad applicability. This is particularly true because many data sets can be downloaded from EMPIAR. Below are my comments for the authors to consider.
+
+AR: Thanks for your support. We appreciate your suggestions for improving our paper.
+
+RC: (1-1) The authors developed a CryoSieve procedure for particle sorting to further remove particles after consensus refinement. They selected six EMAPIR data sets with resolutions between 4.11Å and 3.04Å. To show its broader applicability, the authors should expand their tests on additional data sets. 1) They should include data sets better than 3 Å, for example selecting data sets between 2-3 Å, and data sets better than 2.0 Å. As the resolution of cryo-EM has reached atomic resolutions at about 1.2 Å, it would be interesting to see if CryoSieve works on these very high-quality datasets. The analysis will help better understand the performance of CryoSieve.
+
+AR: Thank you for the thoughtful suggestion. We introduced an additional dataset of human apoferritin (EMPAIR-10200) at 1.9Å for further investigation. Impressively, CryoSieve was able to filter out 79% of the particles from the final stack, enhancing the resolution from 1.89Å to 1.81Å (based on half-maps resolution). Additionally, the number of particles in the finest subsets exceeded the theoretical limit by only a slight margin of 8%.
+
+To achieve atomic resolutions around 1.2Å, it becomes necessary to consider high-order aberrations during image processing and density map reconstruction. However, the current version of CryoSieve does not implement high-order aberrations correction. The potential for CryoSieve to operate at these cutting-edge resolutions is intriguing, and we plan to explore this in future research.
+
+The seventh is from human apoferritin (EMPAIR-10200).
+
+<table>
+  <tr>
+    <th colspan="2">Out of the six datasets examined, two (pfCRT and TSHR-Gs) were found to be close to their theoretical limits (Table 2, column E, emphasized by bold font).</th>
+  </tr>
+  <tr>
+    <th colspan="2">Out of the eight datasets examined, three (pfCRT, TSHR-Gs and apoferritin) were found to be close to their theoretical limits (Table 2, column E, emphasized by bold font).</th>
+  </tr>
+</table>
+
+RC: (1-2) The signal over noise of particles depends on the particle size. To evaluate the quality of the work and performance of CryoSieve, it is suggested to evaluate the performance of CryoSieve in dealing with particles of different sizes, i.e. molecular weights as kDa. Cryo-EM can allow structure determination of particles with a molecular weight of about 50 kDa or lower.
+
+AR: Thank you for your valuable suggestion. We have included the streptavidin dataset (EMPAIR-10269) in Tables 1, 2 and Figures 1, 2, 3, which has a molecular weight of 52kDa. CryoSieve managed to remove 67.2%
+of the particles from the final stack, improving the resolution from 3.15Å to 2.99Å (based on half-maps resolution).
+
+RC: (1-3) The authors should evaluate the performance of CryoSieve on such small particles and add molecular weight to Tables 1 or 2.
+
+AR: Thank you for your suggestion. We’ve incorporated a streptavidin dataset (52kDa) and applied CryoSieve for particle sieving. The results have been added to Tables 1 and 2. Additionally, we’ve included the molecular weight for each entry in Table 1.
+
+RC: (2) One challenge in cryo-EM is data heterogenicity. In solution, macromolecules are equilibrium in many conformational states which are captured during the vitrification process. Cryo-EM data analysis is essentially a triage process to filter out conformational states and radiation damage. The author claims that CryoSieve can remove radiation-damaged particles. How can they exclude the possibility that the particles they removed could be particles belonging to minor conformational states which are slightly different from the consensus model?
+
+AR: Thank you for your valuable comments. The main goal of our manuscript is to develop a numerical method capable of identifying the smallest subset within the final stack without losing the resolution, which does not specifically address the challenges associated with heterogeneity. Our experiments demonstrate that if particles contain radiation damage or parameter estimation errors (such as orientation, translation, or CTF), CryoSieve can accurately and robustly identify them. Nonetheless, in practical applications, we cannot guarantee that the particles discarded do not contain information about other conformations. We can only ensure that they are unnecessary for the reconstructed density map. In the future, it is promising to delve into the heterogeneity problem by integrating CryoSieve with classification techniques.
+
+RC: (3) The authors used simulated particles to show the effectiveness of CryoSieve in removing radiation-damaged particles. They need to demonstrate the effectiveness using experimental data.
+
+AR: Thank you for your suggestion. In the revision, we have transitioned from utilizing simulated particles to employing experimental data.
+
+To verify the possibility of this conjecture, we acquired micrograph movie stacks of the proteasome using a Titan Krios 300keV cryo-EM equipped with a K3 direct electron detection camera. The defocus range was set between 0.5μm and 1.5μm. Each stack comprised 32 frames with a total electron dose of 50e−1−2. The electron dose was uniformly distributed across all frames. Particles were picked from identical positions using averages from frames 5–14, 10–19, 15–24, and 20–29. Consequently, we constructed a dataset consisting of 183,464 particles that represented four different levels of absorbed electron doses.
+
+This experiment demonstrates the robustness and practical applicability of CryoSieve in filtering out radiation-damaged particles. Specifically, CryoSieve is capable of sieving out the majority of particles heavily damaged by high radiation exposure, with significantly higher accuracy compared to other particle sorting algorithms such as NCC, cisTEM, AGC, and non-alignment classification.
+
+Since experimental data may be influenced by various factors, including incorrect pose, among others, we have moved the simulated particles generated via InSilicoTEM to the supplementary section.
+
+RC: (4) P. 3, line 89: The authors used “high-resolution amplitude” for sorting particles in Fourier space. Have the authors sorted particles based on the high-resolution phase? It would be interesting to compare phase-based sorting with amplitude-based sorting.
+AR: We conducted an experiment where we replaced the chosen criteria in CryoSieve with the high-resolution phase residual, defined as the phase difference between the particle and the reference projection above the high-pass threshold. We utilized the TRPM8 dataset for this experiment. The high-pass threshold of each iteration and the retention ratio of each iteration were consistent with those in the TRPM8 experiment. The FSC resolution of remaining particles using high-resolution phase residual as criterion drops as iteration progresses. The results suggest that the high-resolution phase residual may not be a suitable criterion for particle sorting.
+
+We have revised our manuscript as
+
+Furthermore, the amplitude information within the CryoSieve score proves to be vital, given that the phase residual is ineffective as a metric for particle selection (Supplementary Figure 4).
+
+and added Supplementary Figure 4.
+
+RC: (5) Page 6, lines 151-156. The authors describe the 2D and 3D classification work used in reference 26. Such a statement does not bring in new information here and should be deleted.
+
+AR: Thank you for your suggestion. We have deleted it.
+
+RC: (6-1) Page 6, lines 164-166. The authors claim that CryoSieve can remove over half of the particles with unreliable high-frequency signals without negatively affecting the final reconstruction. However, it’s not clear what’s the criterion/threshold to define “unreliable high-frequency signals”.
+
+AR: Thank you for your comment. We have revised the statement as
+
+These results indicate that CryoSieve can effectively eliminate over half of the particles with unreliable high-frequency signals without negatively affecting the final reconstruction. Therefore, CryoSieve is highly effective in selecting the most informative particles.
+
+The results demonstrate that CryoSieve is proficient in discarding more than half of the particles, utilizing the CryoSieve score—a metric reflecting the discrepancy between the particle image and its reference projection. Crucially, this process does not compromise the quality of the final reconstruction.
+
+RC: (6-2) Besides, after removing the “unreliable high-frequency signals”, have the authors observed improved cryo-EM densities or structural features that were blurred or missing in the published maps?
+
+AR: Thank you for your suggestion. Detailed comparisons of improved cryo-EM densities or structural features were plotted in Supplementary Figure 8.
+
+We revised our manuscript as
+
+For some datasets, the density maps showed a certain degree of improvement, which was visualized by the restoration of some previously blurred or missing side chains in the density map (Supplementary Figure 8).
+
+RC: (7) Page 6, lines 168-171. The authors compared CryoSieve with two other sorting methods of NCC and AGC. The authors should discuss in more detail why their method is better than the other two.
+
+AR: Thank you for your suggestion. The main reason that CryoSieve outperforms both NCC and AGC may be due to the incorporation of the high-pass operator in the calculation of the CryoSieve score. Through both
+theoretical analysis and simulation verification, this high-pass operator exhibited superior results. We found that high-pass operator is essential in the determination of particle scores. NCC and cisTEM score calculates particle score using information across all frequencies, resulted in worse performance than CryoSieve. Without truncating high frequencies, the score is likely dominated by low-frequency components, complicating the distinction of non-contributory particles in cryo-EM.
+
+We have revised our manuscript as
+
+Therefore, CryoSieve significantly outperforms other particle sorting algorithms, demonstrating that the majority of particles are dispensable in the final stacks. A key factor in CryoSieve’s superiority over both NCC, AGC and non-alignment classification is the integration of the highpass operator when computing the CryoSieve score. Without the truncation of high frequencies, scores may be predominantly influenced by low-frequency components, making it challenging to differentiate non-contributory particles in cryo-EM.
+
+RC: (8) Did the authors observe the preferred orientation issue while sorting particles based on high-frequency signals?
+
+AR: Thank you for the suggestion. To explore the pose distribution before and after applying CryoSieve, we visualized the directional distribution reported by CryoSPARC for all particles, the finest subset, and those particles sieved out by CryoSieve (Supplementary Figure 6). The pose distributions of the removed particles were similar to those of all particles in the final stacks.
+
+RC: (9) In cryo-EM, nonalignment classification is routine and effective for the classification of heterogeneous data. The nonalignment classification can sort and remove particles, in the meanwhile can identify additional conformational states. The authors should compare the performance of CryoSieve with the nonalignment classification in terms of removing particles while maintaining the resolution.
+
+AR: We compared the performance of CryoSieve with nonalignment classification. We applied nonalignment classification to sieve particles in the final stack. The particles were divided into four classes, and only particles belonging to the class with the highest resolution were retained. Subsequently, we used CryoSPARC to perform ab initio refinement on this selected group of particles.
+
+For three of the eight evaluated datasets, non-alignment eliminated more than 10% of all particles, resulting in some improvement in resolution. However, this improvement is significantly less noticeable than what is observed with CryoSieve (Supplementary Material V).
+
+We have revised this manuscript as
+
+For the non-alignment classification applied to hemagglutinin, LAT1, and apoferritin, fewer than half of the particles were removed, resulting in some enhancement (Supplementary Material V). However, this enhancement still falls notably short of the results achieved by CryoSieve (Supplementary Material V). For the other five datasets, the retaining ratios using non-alignment classification exceeded 90%, which meant that the quality of maps reconstructed from the retained particles either remained unchanged or deteriorated (Supplementary Material V).
+
+RC: (10) Page 6, lines 181-183. The authors used “Einstein-from-noise” to justify the removal of the deposited Euler angles. This statement is not accurate because there is no evidence that the published reconstructions/Euler angles suffered from the “Einstein-from-noise” issue. The authors should revise the sentence to say “to remove bias in the published maps”.
+AR: Thanks for your suggestion. We agree with it and have accordingly revised the statement. The main motivation for re-estimating the Euler angles is eliminating any bias in the original final stack. In this stack, the estimation of Euler angles could potentially be influenced by the particles that were discarded using CryoSieve. We revised our manuscript as
+
+For all the aforementioned methods (CryoSieve, NCC, AGC, and random), we discarded the published refined Euler angles deposited on EMPIAR to avoid the Eisenstein-from-noise effect.
+
+For all the aforementioned methods (CryoSieve, NCC, AGC, non-alignment classification, and random), we discarded the refined Euler angles published and deposited on EMPIAR to prevent the inadvertent transfer of information from the removed particles to the retained particles.
+
+RC: (11) Page 6, lines 185-187. The authors should include all metrics, in addition to FSC and Q-score, that they have used to evaluate the maps before and after the CryoSieve procedure. As B factors are important for evaluating data quality, they should plot B factors with respect to the number of iterations in Figure 2 and Supplementary Figure 2.
+
+AR: Thank you for your suggestions. We have plotted the Rosenthal-Henderson curve against the number of iterations for all datasets, comparing CryoSieve, NCC, cisTEM, and random, as shown in Figure 2. The content previously displayed in Supplementary Figure 2 has been incorporated into the current version of Figure 2.
+
+RC: (12-1) Page 7, Figure 1 caption: Did the authors apply the same B-factor to sharpen the maps before and after CryoSieve?
+
+AR: The B-factor to sharpen the maps before and after the application of CryoSieve is the same.
+
+    · · · the same B-factor The density maps were first FSC-weighted (based on FSCs given by CryoSPARC), and then B-factor sharpened using equivalent B-factors for the same protein, before and after CryoSieve’s sieving: −90Å$^2$ for TRPA1, −180Å$^2$ for hemagglutinin, −100Å$^2$ for LAT1, −60Å$^2$ for pfCRT, −70Å$^2$ for TSHR-Gs, −80Å$^2$ for TRPM8, −65Å$^2$ for apoferritin, and −110Å$^2$ for streptavidin.
+
+RC: (12-2) In addition to the sharpened maps, the authors may compare the non-sharpened maps in a Supplementary Figure.
+
+AR: Thank you for the suggestion. We have added Supplementary Figure 1 to provide a comparative view of the unsharpened maps for all eight datasets, both before and after sieving.
+
+We have revised our manuscript as
+
+The raw density maps corresponding to these results, unsharpened by B-factor, are presented in Supplementary Figure 1.
+
+RC: (12-3) For a better comparison, they should include the contour levels that were used to draw maps before and after CryoSieve.
+
+AR: Thank you for your suggestion. We include the contour levels that were used to draw maps before and after CryoSieve in Figure 1 and Supplementary Figure 1.
+The equivalent contour level was applied for each protein respectively, as indicated at the base of each ratio bar.
+
+RC: (13) Page 7, lines 194-195. If cisTEM reports a per-particle score, the authors should explain why the score can’t be used as a particle sorting criterion.
+
+AR: We have moved the cisTEM results from the supplementary materials to the main body to facilitate a comparison with CryoSieve, NCC, AGC, and the non-alignment method, as shown in Figure 2. However, it is important to note that during the 3D refinement in cisTEM, the poses are either re-estimated or refined. This makes the comparison not strictly fair. We have highlighted this consideration in the caption of Figure 2.
+
+We have revised our manuscript as
+
+cisTEM is capable of reporting a score for each single particle image after 3D reconstruction, though it is not a particle sorting criterion. Due to differences in alignment and other image processing workflows between cisTEM and cryoSPARC, cisTEM cannot be strictly compared with CryoSieve.
+
+and
+
+CisTEM can report a score for each single particle image after 3D refinement. During the 3D refinement process of cisTEM, the pose parameters of particles are re-estimated or refined. Therefore, due to differences in alignment and other image processing workflows between cisTEM and CryoSPARC, cisTEM cannot be strictly compared with CryoSieve.
+
+RC: (14) Page 8, lines 211-214. Why did CryoSieve remove a substantial number of high-resolution 2D particles in TRPA1, but not in the other five data sets (Figure 2)? The authors should perform 2D class averaging on additional data sets.
+
+AR: Thank you for your comment. For TRPA1, the broad resolution range was not conducive for plotting, resulting in inadequate segmentation of the high-resolution range into various resolution categories in the histogram. We have added an additional histogram focusing on the partial resolution range for TRPA1, clearly indicating that particles with high resolution (7.4–7.1Å) were completely retained by CryoSieve.
+
+We have also included two additional datasets, apoferritin and streptavidin, in Figure 2 after processing them through 2D classification. For apoferritin, particles within the highest resolution range (5.5–5.3Å) were predominantly those retained by CryoSieve. In contrast, the streptavidin dataset, possibly due to using a phase plate during data collection, displayed unusually high resolutions during the 2D classification step. This anomaly made a direct comparison between the retained and discarded particles ineffective.
+
+We have revised our manuscript as
+
+In five out of the six datasets, particle images with the highest resolution, i.e., 8.5–9.6 Å in hemagglutinin, 6.6–8.2 Å in LAT1, 7.2–11.6 Å in pfCRT, 7.2–8.5 Å in TSGH-Gs, and 11.6–7.5 Å in TRPM8, were entirely retained by CryoSieve.
+
+In six out of the eight datasets, particle images with the highest resolution, i.e., 7.4–7.1 Å in TRPA1, 8.5–9.6 Å in hemagglutinin, 6.6–8.2 Å in LAT1, 7.2–11.6 Å in pfCRT, 7.2–8.5 Å in TSGH-Gs, and 11.6–7.5 Å in TRPM8, were entirely retained by CryoSieve. For apoferritin, the majority of
+particles within the highest resolution range (5.5-5.3 Å) were constituted by the particles retained by CryoSieve. However, for streptavidin, possibly due to the adoption of a phase plate during data collection, unusually high resolutions were reported in the 2D classification step, rendering such a comparison between retained and removed particles ineffective.
+
+For TRPA1 and apoferritin, the bar with the highest resolution range was further finely divided and then plotted in a histogram, which is displayed to the right of the global histogram.
+
+RC: (15) Page 8, section 2.4: Using simulated data, the authors claim that CryoSieve can effectively detect radiation-damaged particles better than NCC and cisTEM. The authors should also compare CryoSieve performance with the AGC method and nonalignment classification method. In addition, they should use experimental data, not just simulated data to show its effectiveness in the treatment of experimental radiation damage.
+
+AR: Thank you for your suggestion. We have incorporated both the ACG method and non-alignment classification in the analysis of the efficacy of removing radiation-damaged particles. We have added the experiments related to the experimental radiation damage data and compared CryoSieve performance with the AGC method and nonalignment classification method. The experimental setup is given as follows.
+
+To verify the possibility of this conjecture, we acquired micrograph movie stacks of the proteasome using a Titan Krios 300keV cryo-EM equipped with a K3 direct electron detection camera. The defocus range was set between 0.5μm and 1.5μm. Each stack comprised 32 frames with a total electron dose of 50e−1−2. The electron dose was uniformly distributed across all frames. Particles were picked from identical positions using averages from frames 5–14, 10–19, 15–24, and 20–29. Consequently, we constructed a dataset consisting of 183,464 particles that represented four different levels of absorbed electron doses.
+
+As shown in Figure 4, the results show CryoSieve’s proficiency in identifying particles with radiation damage compared to other particle sorting algorithms.
+
+RC: (16) Page 9, Table 2: For B-factor calculation, the authors should use the Rosenthal and Henderson’s B-factor method instead of values from the CryoSPARC auto-processing.
+
+AR: Thank you for your suggestion. We have included the Rosenthal and Henderson’s B-factors in Table 2. Additionally, the process of determining these B-factors is illustrated in Supplementary Figure 5. We also add an additional column in Figure 2, representing the Rosenthal-Henderson B-factors of CryoSieve, NCC, cisTEM, and random across iterations.
+
+We have revised our manuscript as
+
+The third column depicts Rosenthal-Henderson B-factors.
+
+and
+
+The process of fitting and solving for Rosenthal and Henderson’s B-factors is visualized in Supplementary Figure 5.
+
+along with adding the Supplementary Figure 5.
+RC: (17) Page 11, lines 330-345. The discussion on sample preparation is off-topic to the work and should be removed or revised in the context of CryoSieve.
+
+AR: Thank you for your suggestion. We have removed the paragraph detailing sample preparation. It is worth-noting that, in addition to progress towards theoretical limits, CryoSieve could potentially offer a quantitative metric for assessing sample quality. This could impact future technology in structural biology. Our initial collaboration with Professor Hongwei Wang supports this conjecture, and we plan to submit another report from this perspective.
+
+RC: (18) The authors should have shared their code as an attachment for a better evaluation of the work.
+
+AR: CryoSieve is now open-sourced and available on GitHub. A detailed tutorial can also be found on its homepage. Additionally, the datasets used in this manuscript, along with the expected outputs after running CryoSieve, have been deposited on GitHub and can be accessed via CryoSieve’s homepage.
+
+We have updated the ‘Code Availability’ section in this manuscript accordingly.
+
+CryoSieve will be open-source upon publication and is also available upon request during the review process.
+
+CryoSieve is now open-sourced and available on GitHub (https://github.com/mxhulab/cryosieve). A detailed tutorial can also be found on its homepage. Moreover, datasets used in this manuscript, along with the expected outputs after running CryoSieve, have been deposited on GitHub and can be accessed via CryoSieve’s homepage.
+Reviewers' Comments:
+
+Reviewer #1:
+Remarks to the Author:
+The revisions for the manuscript "Not final yet: a minority of final stacks yields superior amplitude in single-particle cryo-EM" from Zhu et al, which approaches the important question in cryo-EM of the number of images required to get high resolution structures, greatly improved the quality of the paper. The additional experiments comparing the results of the method for CryoSPARC and RELION are quite intriguing. The additional supplemental figures, description of the method, and release of the github code are an excellent addition.
+
+Reviewer #2:
+Remarks to the Author:
+The revised manuscript by Zhu, et. al has been much improved. All of my concerns were mostly addressed satisfactorily. Here are a few minor things for authors to consider to further improve the accuracy and readability of their paper.
+P. 3, lines 101-103. “We conclude that, for these datasets, the opportunity for further improvement lies in generating fewer futile particles during sample preparation rather than further improving the quality of the particle images that constitute the finest subset.”. This sentence may lead to the interpretation that the images in the finest subset is perfect which is true. Please revise the sentence.
+P. 4. Line 123. “It tends to deviate significantly from the true amplitude.” How did the authors obtain the true amplitude? Which module did the authors use in their cryoSPARC reconstructions, homologous refinement or reconstruction only? The authors may need to use the correct module in cryoSPARC as they did for Relion (relion_recontruct) to get comparable results.
+P.7. Lines 200-202. “For the non-alignment classification applied to hemagglutinin, LAT1, and apoferritin, less than half of the particles were removed, resulting in some enhancement (Supplementary Material V).” For non-alignment classification, one can remove more particles by varying the number of classes (K), the regularization parameter (tau2_fudge), and the number of iterations. In general, a higher tau2_fudge value (>4) may lead to the removal of more particles. How many interactions did they use nonalignment for their classifications? The authors might want to play with these parameters and include them in their Supplementary Material V.
+
+P.7. Lines 213-215. “Thus, the retained particles were used for ab initio reconstruction by CryoSPARC to obtain refreshed sets of Euler angles and density maps.” This sentence causes a confusion with their previous argument that “CryoSPARC is not the optimal choice for reconstruction.” (P. 4, line 122). The authors should clarify how they use cryoSPARC and Relion for each of the steps in order to perform CryoSieve. Adding external programs and modules to Supplemental Figure 3 would be helpful.
+
+P. 21. Fig. 2. The y-axis label of B factor plots in the right panels shouldn’t be reverted. B-factors can be plotted the same as the left panels for resolution, with a higher resolution, i.e. smaller value, at the top of the y-axis.
+
+P . 23. Fig. 4. It’s strange that panel a starts from the second raw. On the PDF version, I couldn’t tell any particles on the images. The authors should rearrange their panels and consider replacing panel a with a different plot or just a diagram to show how they distributed the total dose across four subsets. In addition, the standard format of dose is e<sup>-</sup>/Å<sup>2</sup>, not e<sup>-1</sup>/Å<sup>2</sup>.
+Authors’ Response to Reviews of
+A minority of final stacks yields superior amplitude in single-particle cryo-EM
+
+Jianying Zhu, Qi Zhang, Hui Zhang, Zuoqiang Shi, Mingxu Hu and Chenglong Bao
+Submitted to Nature Communications, NCOMMS-23-22170A
+
+RC: Reviewers’ Comment.    AR: Authors’ Response,   □ Manuscript Text
+
+We sincerely thank the valuable suggestions and comments from the reviewers. We list our point-to-point replies in the following context and hope that the revision can address the concerns.
+
+Response to Referee #1
+
+RC: The revisions for the manuscript 'Not final yet: a minority of final stacks yields superior amplitude in single-particle cryo-EM' from Zhu et al, which approaches the important question in cryo-EM of the number of images required to get high resolution structures, greatly improved the quality of the paper. The additional experiments comparing the results of the method for CryoSPARC and RELION are quite intriguing. The additional supplementary figures, description of the method, and release of the github code are an excellent addition.
+
+AR: Thanks for your support of this paper.
+Response to Referee #2
+
+RC: (1) The revised manuscript by Zhu, et. al has been much improved. All of my concerns were mostly addressed satisfactorily. Here are a few minor things for authors to consider to further improve the accuracy and readability of their paper.
+
+AR: Thanks for your support. We appreciate your suggestions for improving our paper.
+
+RC: (2) P. 3, lines 101-103. “We conclude that, for these datasets, the opportunity for further improvement lies in generating fewer futile particles during sample preparation rather than further improving the quality of the particle images that constitute the finest subset.”. This sentence may lead to the interpretation that the images in the finest subset is perfect which is true. Please revise the sentence.
+
+AR: Thank you for your comment. We have revised the statement as
+
+From our experiments, we suggest that advancements during the sample preparation process, aimed at increasing the proportion of the finest subset in the final stack, could potentially facilitate the development of cryoEM.
+
+We conclude that, for these datasets, there is greater room for further improvement in generating fewer futile particles during sample preparation than further improving the quality of the particle images that constitute the finest subset.
+
+RC: (3) P. 4. Line 123. “It tends to deviate significantly from the true amplitude.” How did the authors obtain the true amplitude? Which module did the authors use in their cryoSPARC reconstructions, homogeneous refinement or reconstruction only? The authors may need to use the correct module in cryoSPARC as they did for Relion (relion_reconstruct) to get comparable results.
+
+AR: In our experiment, we utilized the reconstruction-only module in CryoSPARC reconstructions, which is understood to perform a similar function to relion_reconstruct. Given that RELION is open-source software deploying Fourier central-slice-theorem-based methods for reconstruction, we can verify its output of true amplitudes. In contrast, our findings indicate that the output amplitudes from CryoSPARC differ from those from RELION in terms of magnitude by several orders. Furthermore, the exact reconstruction process of CryoSPARC remains unclear to us due to its closed-source nature. For these reasons, we have found that the density reconstructed by CryoSPARC significantly deviates from the true amplitude.
+
+RC: (4) P.7. Lines 200-202. “For the non-alignment classification applied to hemagglutinin, LAT1, and apoferritin, less than half of the particles were removed, resulting in some enhancement (Supplementary Material V).” For non-alignment classification, one can remove more particles by varying the number of classes (K), the regularization parameter (tau2_fudge), and the number of iterations. In general, a higher tau2_fudge value (>4) may lead to the removal of more particles. How many interactions did they use nonalignment for their classifications? The authors might want to play with these parameters and include them in their Supplementary Material V.
+
+AR: Thank you for your suggestion. We varied the number of classes (K), the regularization parameter (tau2_fudge) and the number of iterations for experimental datasets. The particles were divided into K classes, and only particles belonging to the class with the highest resolution were retained. The number of classes (K) is 20 or 40, the number of iterations (iter_num) is 40 or 80, and tau2_fudge is 6 or 8. We selected the particle with the highest FSC resolution of the reconstruction density map from these eight possible combinations and reported them. The box size of TSHR-Gs was 448, the particle number of LAT1 and apoferritin were 250,712
+and 382,391, our machine run out of memory for some parameters of these datasets, so the iteration number of TSHR-Gs, LAT1, and apoferritin were set to 25.
+
+The additional experimental results were given in Supplementary Material V (Supplementary Table 5 and Supplementary Table 6).
+
+In addition, we conducted variations in several parameters across all eight experimental datasets: the number of classes (K), the regularization parameter (tau2_fudge), and the number of iterations used in nonalignment classification. In this process, particles were categorized into various classes, with only those in the highest-resolution class being retained for further analysis. Specifically, we tested the configurations where the number of classes (K) were set to 20 and 40, the number of iterations (iter_num) were 40 and 80, and tau2_fudge values were chosen as 6 and 8. We selected the particle with the highest FSC resolution of the reconstruction density map from these eight possible combinations of paraemters and reported them. The box size of TSHR-Gs was 448, the particle number of LAT1 and apoferritin were 250,712 and 382,391, our machine run out of memory for some parameters of these datasets, so the iteration number of TSHR-Gs, LAT1, and apoferritin were set to 25. Subsequently, we used CryoSPARC to perform ab initio refinement on this selected group of particles. The results are shown in Supplementary Table 5 and Supplementary Table 6
+
+RC: (5) P.7. Lines 213-215. “Thus, the retained particles were used for ab initio reconstruction by CryoSPARC to obtain refreshed sets of Euler angles and density maps.” This sentence causes a confusion with their previous argument that “CryoSPARC is not the optimal choice for reconstruction.” (P. 4, line 122). The authors should clarify how they use CryoSPARC and Relion for each of the steps in order to perform CryoSieve. Adding external programs and modules to Supplementary Figure 3 would be helpful.
+
+AR: We used RELION to reconstruct density maps from final stacks in the particle selection step, and then used CryoSPARC to re-estimate poses of the retained particles in ab initio refinement step.
+
+Given that \( q_j \) relies on the accurate amplitude of the reconstructed-density map \( x^{(k)} \), CryoSPARC is not the optimal choice for reconstruction (Supplementary Figure 2).
+
+Given that \( q_j \) relies on the accurate amplitude of the reconstructed density map \( x^{(k)} \), CryoSPARC is not the optimal choice for reconstruction in particle selection step (Supplementary Figure 2).
+
+We have modified the flow chart in Supplementary Figure 3, including the modules and software we used.
+
+RC: (6) P. 21. Fig. 2. The y-axis label of B factor plots in the right panels shouldn’t be reverted. B-factors can be plotted the same as the left panels for resolution, with a higher resolution, i.e. smaller value, at the top of the y-axis.
+
+AR: We appreciate your suggestion, and Figure 2 has been revised accordingly.
+
+RC: (7) P . 23. Fig. 4. It’s strange that panel a starts from the second raw. On the PDF version, I couldn’t tell any particles on the images. The authors should rearrange their panels and consider replacing panel a with a different plot or just a diagram to show how they distributed the total dose across four subsets. In addition, the standard format of dose is e-/Å2, not e-/l/Å2.
+
+AR: Thank you for your suggestion. We have re-arranged the layout of panels in Figure 2, placing panel (a) in the 1st column of this figure. Additionally, we have corrected the unit from \( e^{-1} \text{Å}^{-2} \) to \( e^{-1} \text{Å}^{-2} \).
+Figure 1: **Flow chart scheme for CryoSieve**. CryoSieve operates through multiple iterations. Each iteration comprises both density map reconstruction and particle sieving.
+
+Given the extremely low signal-to-noise ratio (SNR) in single-particle cryo-EM images, directly observing particles within images is difficult. Panel (a) illustrates how variations in electron dose do not result in noticeable changes when observed with the naked eye, thereby highlighting the need for a specialized algorithm for analysis. In response to this requirement, we propose the utilization of CryoSieve.

0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/peer_review/peer_review.md

@@ -0,0 +1,398 @@
+Peer Review File
+
+Peptide nano-blanket impedes fibroblasts activation and subsequent formation of pre-metastatic niche
+
+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/.
+REVIEWER COMMENTS</B>
+
+Reviewer #1 (Remarks to the Author):
+
+This manuscript reported that an MMP-2 activatable peptide assembly FR17 impedes the establishment of primary tumor induced pre-metastatic niche (PMN) as a ‘flame-retarding blanket’. The peptide FR17 consists of two main parts, which is (1) the self-assembly peptide domain FFKY and (2) thymopentin with good hydrophilic property and immune modulation effect. Considerable amount of in vitro experiments were done to illustrate the successful inhibition of PMN formation from different aspects, including the interruption of the activation of fibroblast, the inhibition of the vascular leakage and angiogenesis, the impediment of MDSC recruitment and modulatory of the immune microenvironment. Besides, In vivo experiments were further verified the FR17 treatment inhibits melanoma lung metastasis. Although the MMP2 responsive behavior of peptides in vivo remains to be confirmed, and low efficiency of MMP2 implied in the limitation of the designed peptide, this work, indeed, illustrates the promises of using enzymatic self-assembly for treating cancer. Thus, I support the acceptance of this work if the authors were able to addressing the following issues.
+
+1. The key problem is the lack of the validation of the nanoblankets. The morphology changes before and after enzyme treatment should be compared to verify the in-situ formation of this blanket like structure. This experiment is imperative to support the claim of nanoblankets in vivo.
+2. MS data of peptide sequence after MMP2 treatment should be given to verify the enzymatic response of designed peptides.
+3. The concentrations of peptides used for TEM measurement can be listed in the caption of Figure 1.
+4. There are some outstanding works about the enzymatic responsive peptides to induce cancer cell death or inhibit tumor growth, for example 10.1021/ja510156v and j.chempr.2019.06.020. The authors should properly refer the closely related works.
+5. FFKY is not a part of b-amylod since the sequence of A-beta is “DAEFRHDSGYEVHHQKLVFFAEVDVGSNKGAIIGLMVGGVVIA”. Refs14,15 are not the first use of FFKY in self-assembly. The author should give a more proper reference of the FFKY in the literature.
+
+Reviewer #2 (Remarks to the Author):
+
+This article applied peptide nano-blanket to impede pre-metastatic niche formation in the lung, especially by suppressing fibroblast activation.
+The article is interesting, revealing great potential for newly developed technologies, and fascinating, showing possible treatment prior to cancer cell arrival for decreasing metastasis. This reviewer does agree that this story is important and is novel, however, this reviewer feels that there are several issues that need to be addressed:
+1) My biggest concern is that authors didn’t show where and how nano-blanket is attached. The authors need to show in some way how much nano-blanket resides within the lung and also in other organs. They claim that this is site specific, but liver pre-metastatic niche formation should also be occurring. Is it
+really lung specific?
+2) In the same note as above, how long does nano-blanket stay in the organ. Based on figure 6, it seems like metastasis in the lung is just simply delayed. Would it be possible to keep the treatment going and would that truly impede metastasis.
+3) Also, it would be important to show whether nano-blanket is degradable. When would it degrade after landing to the lung. Or would it stay without degrading.
+4) With the current treatment, does authors believe that they are flaming out all activated fibroblasts in the lung, or is it partial flame-out. It was not clear how they decided upon the injection amount.
+5) In figure 4, authors show that activated fibroblast CM induces angiogenesis and nano-blanket treated CM suppresses this effect. It is not clear whether there are any left-over nano-blanket in the CM, which could directly impede angiogenesis. Authors need to show that there are no nano-blanket left in the CM.
+6) Minor point: Please show in figure 2a an arrow pointing to what we should be looking at. It is hard to appreciate what we should be seeing. Also, FFKY image looks completely different from the others. Was this expected?
+7) Minor point: In figure 3f, invasion assay and contraction assay should be labeled separately.
+8) Minor point: In figure 4d, MCM free control also shows big gap in the current image. Quantitative analysis is needed and an image that represents the quantitative data needs to be replaced.
+9) Minor point: Although bone-marrow recruited cells, especially MDSCs, are analyzed in the data, not much is mentioned in the introduction. It would be helpful for the readers to include more information on why this analysis is necessary.
+
+Reviewer #3 (Remarks to the Author):
+
+In this manuscript, the authors constructed an enzyme-activatable peptide FR17 to release self-assembly monomer FG8 to form nano-blanket in PMN microenvironment for impeding fibroblasts activation and preventing metastatic cascades. The design is interesting, and the authors did comprehensive work to justify their hypothesis. The presentation of the work is also clear and neat. The following comments can be considered for improvement.
+
+1. In page 3 line 14, the authors stated that nano-blanket in PMN can impede fibroblasts activation to prevent metastatic. Why did the authors choose fibroblasts as the target of intervention? As primary tumor-derived cytokines and exosomes and myeloid-derived suppressor cells (MDSCs) all contribute to metastasis, it is necessary to provide the background about the keys factors influencing PMN microenvironment, and discuss the role of fibroblasts in PMN and the current studies made in this field based on fibroblasts.
+2. Is there any direct evidence about the production of FG8 in the presence of MMP2?
+3. In supplementary Figure 4, FG8 was shown to achieve self-assembly in water? Did the authors investigate self-assembly of FG8 under different pH conditions? The authors should consider the condition in PMN and investigate the self-assembly of FG8 under the corresponding condition.
+4. Please provide semi-quantitative data in Figure 3g, 4e, 5c, supplementary Figure 11 and 12.
+5. Why did the authors choose subcutaneous injection to administrate FR17 instead of intravenous injection? Please explain.
+** See Nature Research’s author and referees' website at www.nature.com/authors for information about policies, services and author benefits.
+Point-by-point Response to the Reviewers’ Comments
+
+Reviewer #1 (Remarks to the Author):
+
+This manuscript reported that an MMP-2 activatable peptide assembly FR17 impedes the establishment of primary tumor induced pre-metastatic niche (PMN) as a ‘flame-retarding blanket’. The peptide FR17 consists of two main parts, which is (1) the self-assembly peptide domain FFKY and (2) thymopentin with good hydrophilic property and immune modulation effect. Considerable amount of in vitro experiments were done to illustrate the successful inhibition of PMN formation from different aspects, including the interruption of the activation of fibroblast, the inhibition of the vascular leakage and angiogenesis, the impediment of MDSC recruitment and modulatory of the immune microenvironment. Besides, in vivo experiments were further verified the FR17 treatment inhibits melanoma lung metastasis. Although the MMP2 responsive behavior of peptides in vivo remains to be confirmed, and low efficiency of MMP2 implied in the limitation of the designed peptide, this work, indeed, illustrates the promises of using enzymatic self-assembly for treating cancer. Thus, I support the acceptance of this work if the authors were able to addressing the following issues.
+
+1. The key problem is the lack of the validation of the nanoblankets. The morphology changes before and after enzyme treatment should be compared to verify the in-situ formation of this blanket like structure. This experiment is imperative to support the claim of nanoblankets in vivo.
+
+Response: We appreciate your comments and suggestions and totally agree. The morphology changes before and after enzyme treatment of FR17 has been observed via TEM. With hydrophilic property, FR17 wouldn’t aggregate in aqueous solution, as indicated in Figure 1a below, which couldn’t be measured and analyzed by the particle size meter (Zetasizer Nano ZS, Malvern Instruments), while the peptide self-assemblies aggregated by enzymatic degradation with an average diameter of 500–800 nm was showcased in Figure 2b in the manuscript. The lamellar structure of the peptide nano-blanket was pseudo-colored in gold in TEM and Cryo-TEM images (Figure 1a, b). When added enzyme to FR17 and left to stand for 72 h, a visible semitransparent thin layer is woven in the aqueous system, which can be easily ripped up by gently shaking and split into smaller pieces through sharply shaking (Figure 1c). Besides, the specific microenvironment-responsive assembly of peptide nano-blanket was validated at the cellular level on fibroblasts. As indicated by Figure 1d, peptide nano-blanket (pseudo-colored in gold) was constructed on the cell surface of fibroblasts in the condition of melanoma-conditioned media (MCM) cultivation, for MCM contains and also increases the
+secretion of MMPs.
+
+![TEM images of FR17 (500 μM) before and after MMP2 treatment (1 μg/ml). Scale bar = 200 nm. Cryo-TEM image of the peptide nano-blanket assembled by FR17 (500 μM) treated with MMP2 (1 μg/ml). Scale bar = 200 nm. Macroscopic images of the thin layer formed by FR17 treated with enzyme and let stand for 72 h. The soft thin layer broke into pieces after gently shaking and dispersed into nanoscale fragment after sharply shaking. SEM images of the peptide nano-blanket assembled on the cell surface of fibroblasts in the condition of MCM inducement. The in-situ formed peptide nano-blanket on the surface of fibroblasts is pseudo-colored in gold. Scale bar = 1 μm in the macro images. Scale bar = 500 nm in the enlarged images circled with dash line.](page_184_186_1080_1092.png)
+
+Figure I. Peptide nano-blanket transformed from FR17. a, TEM images of FR17 (500 μM) before and after MMP2 treatment (1 μg/ml). Scale bar = 200 nm. b, Cryo-TEM image of the peptide nano-blanket assembled by FR17 (500 μM) treated with MMP2 (1 μg/ml). Scale bar = 200 nm. c, Macroscopic images of the thin layer formed by FR17 treated with enzyme and let stand for 72 h. The soft thin layer broke into pieces after gently shaking and dispersed into nanoscale fragment after sharply shaking. d, SEM images of the peptide nano-blanket assembled on the cell surface of fibroblasts in the condition of MCM inducement. The in-situ formed peptide nano-blanket on the surface of fibroblasts is pseudo-colored in gold. Scale bar = 1 μm in the macro images. Scale bar = 500 nm in the enlarged images circled with dash line.
+What’s more, in order to demonstrate the *in-situ* assembly of the peptide nano-blanket *in vivo*, TPE-FR17 was administrated to PMN mice. It has been demonstrated in Supplementary Figure 6 in the manuscript that FR17 modified with the aggregation-induced emission (AIE) luminogens TPE endows peptide with the fluorescence “turn-on” effect when TPE-FR17 was cleaved to release TPE-FG8 which would spontaneously aggregate. In short, TPE-FR17 was subcutaneously administrated to PMN mice (on Day 10) and healthy mice at the dose of 100 \( \mu \)M/kg. 12 h-post peptide administration, mice were euthanized for cardiac perfusion and the lung tissues were collected for frozen sections. The frozen sections were observed under high-resolution laser confocal microscopy excited by 405 nm laser. As indicated in Supplementary Figure 7 as follow, fluorescent dots were observed in PMN lung with higher expression of MMP2 (Supplementary Figure 8d) rather than in healthy lung, indicating the specific enzymatic cleavage of TPE-FR17 and the *in-situ* assembly of the peptide nano-blanket in PMN lung.
+Supplementary Figure 7. The aggregation-induced emission effect of the in-situ assembly of peptide nano-blanket at 12 h-post subcutaneous administration of TPE-FR17 in healthy or PMN lung. Peptide assemblies of the monomer TPE-FG8 released from TPE-FR17 in PMN were pseudo-colored in white for contrast on the upper panel and pseudo-colored in red in the merged images. Scale bar = 200 μm.
+
+2. MS data of peptide sequence after MMP2 treatment should be given to verify the enzymatic response of designed peptides.
+
+Response: Thank you for your suggestion, which helps to improve and perfect our manuscript. Though we had considered the MMP2-cleavable peptide linker PLGLAG as the confirmed and well-applied MMP2 recognition sequence, it’s essential to verify the enzymatic response of the designed peptides in this work as well. Therefore, LC-MS/MS was applied to verify the production of FG8 from FR17 or sFD17 after MMP2 treatment. Briefly, peptide FR17 or sFD17
+(200 μM) was cultivated with pre-activated hMMP2 (200 ng/ml) for 10 min. The reaction was terminated by adding 3 volume of methanol and centrifuged at 1,3000 rpm for 15 min to discard the protein precipitation. Samples were analyzed by LC-MS/MS. Reference FG8 (the MS and MS/MS of FG8 has been added in Supplementary Figure 1 & 2 in the revised manuscript), and FR17 or sFD17 without enzyme treatment were analyzed as standard control. As demonstrated by Supplementary Figure 3 as follow, the retention time of FG8 (tR = 4.9 min) is ahead of FR17 (tR = 5.5 min) or sFD17 (tR = 5.3 min) in chromatogram. After MMP2 cultivation for 10 min, the characteristic peak of FG8 aroused. What’s more, the peptide sequence produced by MMP2 degradation was able to be identified under the auto-optimized condition of the ion pair 464.8/379.6, in which 464.8 represents [M+2H]^{2+} of FG8 (Exact mass 927.49) and 379.6 represents [M+2H]^{2+} of FFK(GP)Y (Exact mass 757.38). By calculation, the responsive rate of MMP2 to FR17 is 34.28% in 10 min, the responsive rate of MMP2 to sFD17 is 35.82% in 10 min.
+
+![Schematic diagrams and chromatograms showing enzyme cleavage and mass spectrometry analysis of FR17, sFD17, and FG8](page_186_624_1077_627.png)
+
+Supplementary Figure 3. Enzyme cleavage of FR17 and sFD17 to release the self-assembled monomer FG8. a, Schematic of the enzyme cleavage of FR17 or sFD17. b, Liquid chromatography-tandem mass spectrometry (LC-MS/MS) of FR17 and sFD17 (200 μM) before and after MMP2 (200 ng/ml) treatment. The characteristic peak of FG8 was acquired under the auto-optimized condition of the ion pair 464.8/379.6, in which 464.8 represents [M+2H]^{2+} of FG8 (Exact mass 927.49) and 379.6 represents [M+2H]^{2+} of FFK(GP)Y (Exact mass 757.38).
+3. The concentrations of peptides used for TEM measurement can be listed in the caption of Figure 1.
+
+Response: Thank you for your reminder. The concentrations of peptides have been listed in the figure caption in the revised manuscript.
+
+![TEM images and size distribution plots of FR17, sFD17, FG8, and FFKY peptides](page_246_355_1017_181.png)
+
+Figure 2. Enzyme-activated self-assembly of FR17 and all-atom molecular dynamics (MD) simulation of the self-assembly of FG8. **a**, TEM images of FR17 or sFD17 (**500 μM**) treated with MMP2 (**1 μg/ml**) (scale bar = 200 nm), FG8 (**500 μM**) and FFKY (**500 μM**) (scale bar = 100 nm). **b**, Size distribution of FR17 and sFD17 (**500 μM**) after MMP2 (**1 μg/ml**) cleavage.
+
+4. There are some outstanding works about the enzymatic responsive peptides to induce cancer cell death or inhibit tumor growth, for example 10.1021/ja510156v and j.chempr.2019.06.020. The authors should properly refer the closely related works.
+
+Response: Thank you very much for your suggestions. Some great works which are closely related to our research have been cited in the introduction as follow, including 10.1021/ja510156v and j.chempr.2019.06.020 as you suggested.
+
+“Inspired by the self-assembled peptides found in many natural life processes, researchers have modified, designed and synthesized diverse self-assembled peptides applied as functional biomaterials for various applications¹,²,³. In the application on anti-tumor and anti-metastasis therapy, enzyme-responsive self-assembly or ligand-receptor interactions-triggered morphology transforming of the peptide nanofibrils and hydrogel have been developed to induce cell death⁴,⁵,⁶; to restrict tumor cell invasion⁷, to serve as biocompatible drug delivery platforms⁸, to achieve specific targeting of tumor⁹ and imaging¹⁰ as well.”
+
+5. FFKY is not a part of b-amyloid since the sequence of A-beta is “DAEFRHDSGYEVHHQKLVFFAEDVGSNKGAIIGLMVGGVVIA”. Refs14,15 are not the first use of FFKY in self-assembly. The author should give a more proper reference of the FFKY in the
+literature.
+
+Response: Thank you for your suggestion. The reason why the former sentence was organized as “…the backbone of a self-assembly peptide domain Phe-Phe-Lys-Tyr (FFKY), which is derived from β-amyloid (Aβ) peptide.” is to trace back to the fountainhead of FFKY, dipeptide FF, which is extracted from the main fragment of Aβ with self-assembly property11. To avoid misreading, this sentence has been revised as follow:
+
+“…the backbone of a self-assembly peptide domain Phe-Phe-Lys-Tyr (FFKY), a variant of Phe-Phe (FF), which is derived from the Alzheimer’s β-amyloid (Aβ).”
+
+In addition, as the original fragment of Aβ16-20, KLVFF has been applied as therapeutic agents7, 12, imaging agents13 or delivery platform14 in the field of Alzheimer’s disease15 and cancer16. Further researches cut the peptide fragment down to dipeptide FF to obtain discrete nanotubes through self-assembly in aqueous solution1. Afterwards, felicitous modification and re-designment on FF with prefect biocompatibility has been reported with a wide range of applications17. For example, FFKF was developed to construct drug delivery system which can be complete degraded by cathepsin proteases18. Yang and his team employed FFFK to develop molecular hydrogel for co-delivery of anti-cancer drugs19. In another work, the application of FFYK was explored in organelles targeting and cancer cell killing20. Some introduced naphthyl group on N terminal of Phe to favor intercellular hydrophobic interactions21.
+
+Here in our manuscript, taking advantages of both the self-assembly feature of diphenylalanine structural motif FF with the assistance of Y and the editable site provided by K to combine with hydrophilic fragment via enzyme-cleavable linker, FFKY was employed as the backbone of the self-assembled monomer FG8, which would spontaneously fold into lamellar structure, constructing the peptide nano-blanket.
+Reviewer #2 (Remarks to the Author):
+
+This article applied peptide nano-blanket to impede pre-metastatic niche formation in the lung, especially by suppressing fibroblast activation.
+
+The article is interesting, revealing great potential for newly developed technologies, and fascinating, showing possible treatment prior to cancer cell arrival for decreasing metastasis. This reviewer does agree that this story is important and is novel, however, this reviewer feels that there are several issues that need to be addressed:
+
+1) My biggest concern is that authors didn’t show where and how nano-blanket is attached. The authors need to show in some way how much nano-blanket resides within the lung and also in other organs. They claim that this is site specific, but liver pre-metastatic niche formation should also be occurring. Is it really lung specific?
+
+Response: Thank you for your suggestions. With the MMP2-cleavable PLGLAG peptide linkage, FR17 can be degraded to release self-assembly monomer FG8 at the site highly-expressed with MMP2, which is secreted by the stromal cells to the intercellular substance or localize to the cell surface by binding to integrins\(^{22,23}\). Therefore, the dissociated FG8 aggregate to construct peptide nano-blanket on the cell surface or in the intercellular substance *in-situ*. As directly viewed by SEM of the fibroblasts induced by MCM with peptide treatment and the STEM images of the PMN lung collected from the model mice received peptide administration, Figure II exhibits the construction of nano-blanket on cell surface under certain conditions *in vitro* (Figure IIa) and its suspected appearance in the intercellular substance in PMN lung *in vivo* (Figure IIb), and the morphology and the size of which ranges from hundreds to thousands nanometers depending on the intercellular space specifically at PMN sites.
+Figure II. The in-situ assembled peptide nano-blanket in vitro and in vivo. a, SEM images of the peptide nano-blanket assembled on the cell surface of fibroblasts in the condition of MCM inducement. The in-situ formed peptide nano-blanket on the surface of fibroblasts is pseudo-colored in gold. Scale bar = 1 μm in the macro images. Scale bar = 500 nm in the enlarged images circled with dash line. b, STEM images of the peptide nano-blanket in the intercellular substance in PMN lung, which was collected from PMN mouse at 12 h-post subcutaneous administration of FR17. Scale bar = 1 μm.
+
+To reveal the in-situ assembly and distribution of the peptide nano-blanket in vivo, TPE-FR17 was administrated to PMN mice. It has been demonstrated in Supplementary Figure 6 in the manuscript that FR17 modified with the aggregation-induced emission (AIE) luminogens TPE endows peptide with the fluorescence “turn-on” effect when TPE-FR17 was cleaved to release TPE-FG8 which would spontaneously aggregate. In short, TPE-FR17 was subcutaneously administrated to PMN mice (on Day 10) and healthy mice at the dose of 100 μM/kg. At the time point of 1, 2, 4, 8, 12, 24, 48 h-post peptide administration, mice were euthanized for cardiac perfusion and the lung tissues were collected for frozen sections. The frozen sections were observed under high-resolution laser confocal microscopy excited by 405 nm laser. As indicated in Figure III as follow, fluorescent dots were observed in PMN lung with higher expression of MMP2 (Figure IV) rather than in healthy lung, indicating the specific enzymatic cleavage of TPE-FR17 and the in-situ assembly of the peptide nano-
+blanket in PMN lung. What’s more, according to the density of the fluorescent dots presented in the lung sections, the peptide nano-blanket assembled in 1 h after peptide administration, and the process of construction and increased accumulation of the peptide nano-blanket last for 12 h, which might be contributed by the sustained release of TPE-FR17 from the subcutaneous drug storage naturally formed by subcutaneous injection.
+Figure III. The AIE effect of the in-situ assembly of peptide nano-blanket from 1 h to 48 h-post subcutaneous administration of TPE-FR17 in PMN lung. The lung collected at 12 h-post subcutaneous administration of TPE-FR17 to Healthy mouse was set as control.
+Peptide assemblies of the monomer TPE-FG8 released from TPE-FR17 in PMN were pseudo-colored in white for contrast on the upper panel and pseudo-colored in red in the merged images. Scale bar = 200 μm. The semi-quantification was calculated from 6 visual fields per time-point via ImageJ.
+
+We do acknowledge the wide-reported formation of liver PMN in diverse tumors24, 25, 26. However, in the PMN model we’ve referenced27, both the pioneers and our team focused on the pulmonary pre-metastatic and metastatic niches. Though the MCM inducement (which contains tumor-derived secreted factors) might arise broad alterations on the cellular and molecular level as well as the local microenvironment changes, especially in some certain target organs including lung, liver, etc.28, 29 In this model, B16F10 melanoma cells were intravenously administrated as the simulation of circulative tumor cells in the spontaneous process in other metastatic models, who would wander through blood vessels and some of which would successfully colonize in the prepared PMN in lung. Given the circumstances of how this PMN model was established and the fact that no hepatic metastases were observed during the entire pathological process on our PMN model mice (Supplementary Figure 21), our research focused on the pulmonary pre-metastatic niche and the further developed pulmonary metastases.
+
+As we demonstrated, the in-situ assembly of peptide nano-blanket is enzyme-activated at the site with higher expression of MMP2, which is considered as a site-specific hall-marker of PMN30, 31. Therefore, the site-specific assembly of the peptide nano-blanket might not be lung specific in the patients with multiple pre-metastatic sites. To determine whether the responsive assembly of the peptide nano-blanket would occur in liver in the PMN model adopted in the research, MMP2 expression in lung and in liver tissues collected from healthy or PMN mice has been compared. As suggested by Figure IV, the highest expression of MMP2 was detected in PMN lung. Meanwhile, there’s a slight enhancement of MMP2 expression in PMN liver when compared to healthy liver, which remains in the lower level than that of PMN lung.
+
+![Western blot showing MMP2 and GAPDH expression in H-lung, H-liver, PMN-lung, PMN-liver samples](page_1012_1342_388_180.png)
+
+Figure IV. MMP2 expression in healthy or PMN lung and liver on Day 10. H represents for healthy. PMN model was established as described in Method, ie. mice were intraperitoneally injected with melanoma-conditioned medium (300 μl per mice) for 10 consecutive days, and a tail vein injection of B16F10 cells (1 × 10^5 per mice) was given to the mice on Day 7.
+Furthermore, liver tissues as well as other major organs including heart, spleen and kidney were also collected for frozen sections at the time point of 1, 2, 4, 8, 12, 24, 48 h-post peptide administration. The frozen sections were looked over under high-resolution laser confocal microscopy excited by 405 nm laser to examine the AIE effect of peptide assembly. As showcased in Figure V, several fluorescent dots were observed in PMN liver sections with the same accumulation trend as it did in PMN lung in time dimension but with lesser peptide aggregates. For comparison, images of lung and liver sections at 12-h post peptide administration were arranged together in Figure VI. And no fluorescent dot of peptide aggregation has been observed in other organ sections as presented in Figure VII.
+Figure V. The AIE effect of the in-situ assembly of peptide nano-blanket from 1 h to 48 h-post subcutaneous administration of TPE-FR17 in PMN liver. The liver collected at 12 h-post subcutaneous administration of TPE-FR17 to Healthy mouse was set as control.
+Peptide assemblies of the monomer TPE-FG8 released from TPE-FR17 in PMN were pseudo-colored in white for contrast on the upper panel and pseudo-colored in red in the merged images. Scale bar = 200 μm. The semi-quantification was calculated from 6 visual fields per time-point via ImageJ.
+
+![Fluorescent channel and merged channel images of PMN lung and liver showing peptide assemblies](page_354_393_1042_627.png)
+
+Figure VI. The AIE effect of the in-situ assembly of peptide nano-blanket at 12 h-post subcutaneous administration of TPE-FR17 in PMN lung and liver. Peptide assemblies in PMN were pseudo-colored in white for contrast on the upper panel and pseudo-colored in red in the merged images. Scale bar = 200 μm.
+Figure VII. Representative images of the heart, spleen and kidney after TPE-FR17 administration. Scale bar = 200 μm.
+
+2) In the same note as above, how long does nano-blanket stay in the organ. Based on figure 6, it seems like metastasis in the lung is just simply delayed. Would it be possible to keep the treatment going and would that truly impede metastasis?
+
+Response: According to the in-situ peptide assembly reflected by AIE effect as illustrated in Figure III & V above, the peptide nano-blanket formed and further accumulated for 12 hours after subcutaneous administration of the peptide. Almost no fluorescent signal was observed on all organ sections at 24 h-post peptide administration, suggesting that the nano-blanket would constantly construct and stay for hours but would soon be degraded within 24 h.
+
+The peptide FR17 was designed to response to construct peptide nano-blanket in PMN stromal microenvironment, exerting inhibition effect on fibroblasts activation so as to prevent metastatic cascades. In this work, we would like to more concentrate on the process of PMN development, emphasizing on the early intervention on metastasis of FR17. Therefore, we explored further on the subsequent impact of inhibition on fibroblasts activation induced by FR17 intervene, revealing
+the underlining mechanism on cellular interactions among fibroblasts, vascular endothelial cells and extracellular components, and intervention on PMN recruited MDSCs. As for whether if it’s possible to achieve complete suppression on metastasis by tailoring the treatment plan, a preliminary agreement has been made after group discussion. As commonly acknowledged as a complex multistep process, tumor metastasis is promoted and led by multiple factors, including not only the preparation of PMN but also other vital contributors such as the specific phenotypes and invasion ability of tumor cells, the developing stages at the exact time point when take measures for drug intervention or surgical operation. In the animal models we’ve established for the investigation on PMN, a half million highly invasive malignant cells in rapid proliferation were administrated in one single injection to the mice directly through the blood circulation. This simulation of the tens and thousands of circulative tumor cells wondering in the circulation would be diagnosed as the terminal stage of tumor with high aggressiveness in clinic and would be fatal to the patients. Though FR17 administration has been verified to serve as a “flame-retarding blanket” at PMN site specifically to extinguish the “fire” of tumor-supportive microenvironment adaption, FR17 exhibits only suppression on tumor cells migration but no significant anti-tumor effect (Supplementary Figure 19) because the peptide carrying no anti-tumor therapeutics at all. Accordingly, to completely remove all of the metastatic lesions in rapid growth, the co-administration of FR17 and other anti-tumor therapeutics with tumor-killing effect might work, which would be appealing to us in our further researches in the future.
+
+3) Also, it would be important to show whether nano-blanket is degradable. When would it degrade after landing to the lung? Or would it stay without degrading.
+
+Response: Yes, most of the peptide assemblies were biodegradable in vivo with high biocompatibility\(^{18}, ^{32}\), which is one of their superiorities valued in biomedical applications. As illustrated in Figure III & V above, the peptide nano-blanket was constructed and further accumulated for 12 h after subcutaneous administration of the peptide, all of which would be degraded within 24 h with no fluorescent signal being observed on all organ sections at 24 h-post peptide administration. Furthermore, the preliminary safety evaluation of peptide administration on lung metastasis model with MCM-induced PMN suggested the bio-safety of the peptide nano-blanket. Long-term administration of FR17 or sFD17 has no significant hepatic toxicity on PMN model mice (Supplementary Figure 20b). No organic lesions in H & E organ sections was observed (Supplementary Figure 21).
+
+4) With the current treatment, does authors believe that they are flaming out all activated fibroblasts in the lung, or is it partial flame-out. It was not clear how they decided upon the injection amount.
+Response: After the administration of FR17or sFD17 for one week, most of the fibroblasts presented Vimentin+αSMA+ resting state in the lung of PMN model mice as illustrated in the immunofluorescence-stained lung sections in Figure VIII below. Yet it’s inaccurate to draw the conclusion that all of the Vimentin+αSMA+ activated fibroblasts in the lung have been beaten out. Different subsets of fibroblasts, including resting and activated fibroblasts, are both considered to be essential to maintain lung architecture and function, playing a role in regulating air and blood relationships by modulating alveolar geometry. It’s reported that there’s a small population of activated fibroblasts resident even in the healthy lung. The conversion between its resting and activated phenotype maintains a delicate balance. Back to 1970s, it has been suggested that activated lung fibroblasts contribute to immune defense and protect the lung against proteolytic damage. Yet in PMN development, the over production of extracellular matrix and chemotactic factors by the over-aroused fibroblasts go out-of-balance, recruiting immune suppressor cells, heading to the direction of pro-metastasis lesion. Therefore, the wise way is to contain the over-arousement of activated fibroblasts in advance, which has been indicated by the data provided as follow (Supplementary Figure VIII) and in the manuscript from several aspects of both cell markers and bio-functions, production of extracellular matrix (ECM), ECM-remodeling enzymes and cytokines (Figure 3, Supplementary Figure 13).
+
+![Representative images of Vimentin+αSMA+ activated fibroblasts in the lung harvested from mice administrated with different peptides on Day 10. Scale bar = 50 μm.](page_246_1012_1012_384.png)
+
+Figure VIII. Representative images of Vimentin+αSMA+ activated fibroblasts in the lung harvested from mice administrated with different peptides on Day 10. Scale bar = 50 μm.
+
+The drug dose was determined according to former applications and the preliminary experiments. Referred studies on both peptide assemblies6, 7, 33 and the clinical evaluation of hydrophilic drug TP534, 35, mice in different groups were treated with a low (20 μM/kg), medium (40 μM/kg) or high (80 μM/kg) dose of peptide in the preliminary experiment. The therapeutic outcome of
+medium dose is close to the high dose groups, which is better than that of low dose group (Data not shown). On the other hand, according to the determined enzyme-responsive assembly of FR17 (200 \( \mu \)M in response to 200 ng/ml MMP2), the suitable dose of FR17 administration was calculated and further applied as 40 \( \mu \)M/kg, with the peptide concentration reaching to higher than 500 \( \mu \)M *in vivo*. Though the current data illustrated the success in flaming out fibroblast activation induced by tumor-derived secretion factors both *in vitro* (Figure 3a-i, Supplementary Figure 12) and *in vivo* (Figure 3j-k, Supplementary Figure 13, Figure VIII). Still, further exploration of the possible dose-dependent manner of the peptide nano-blanket or FR17 on the suppression extent of pre-metastasis associated fibroblast and the mechanism underneath is required for its clinical applications in the future. Thank you very much for your question.
+
+5) In figure 4, authors show that activated fibroblast CM induces angiogenesis and nano-blanket treated CM suppresses this effect. It is not clear whether there are any left-over nano-blanket in the CM, which could directly impede angiogenesis. Authors need to show that there are no nano-blanket left in the CM.
+
+**Response:** We apologize for the unclear statement in the method that led misunderstanding. There was no nano-blanket nor peptide being left to directly influence angiogenesis. The description of the experimental procedure on this section has been revised in detail to help with comprehension. As illustrated in Supplementary Figure 14a below, MLF-conditioned medium (FCM) was obtained as follows: MLFs were stimulated by MCM (supplemented with 10% (v/v) FBS) with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 \( \mu \)M) for 48 h, the original medium was discard, then replaced with fresh complete medium after PBS rinsing to eliminate the direct influence of MCM and peptide drugs on the later experiment subject. After incubated for another 24 h, the cell supernatants were collected, centrifuged at 2000 rpm for 10 min to discard the cell debris. Thus, the replacement of the culture medium ensured to get rid of the direct effect of MCM and peptide drugs on the endothelial cells.
+![Experimental procedure flowchart showing B16F10 cells cultured for 24h, MCM treatment, MLF, drug treatment (TPS, sFD17, FR17), culturing for 48h, fresh media change, culturing for another 24h, conditional FCM, ECs, and following experiments: proliferation, tube forming, cell layer permeability](page_312_120_823_563.png)
+
+Supplementary Figure 14a, The experimental procedure to obtain the conditional FCM after MCM stimulation and the peptide treatment on MLF in vitro for further experiments on endothelial cells.
+
+6) Minor point: Please show in figure 2a an arrow pointing to what we should be looking at. It is hard to appreciate what we should be seeing. Also, FFKY image looks completely different from the others. Was this expected?
+
+Response: Sorry for the misreading caused by the presented TEM images. We’ve pseudo-colored the lamellar structure of the peptide nano-blanket in gold in the following images, hoping this would help to make out these overlapped curly thin layers. In addition, it’s exactly right that the morphology of FFKY assemblies is completely different from the peptide nano-blanket, which is assembled by FG8. With the high hydrophobicity offered by multi-aromatic groups, FFKY would assemble into nanofibers/nanotubes via intermolecular \( \pi-\pi \) effects, which could be similar to the morphology of FFKF assemblies as formerly reported\(^{19}\).
+Figure 2a. TEM images of FR17 or sFD17 (500 μM) treated with MMP2 (scale bar = 200 nm), FG8 (500 μM) and FFKY (500 μM) (scale bar = 100 nm).
+
+7) Minor point: In figure 3f, invasion assay and contraction assay should be labeled separately.
+
+Response: Thank you very much for your thoughtful suggestion. We’ve labeled two assays separately in Figure 3 in the revised manuscript.
+
+Figure 3. FR17 interrupted the activation of fibroblast induced by tumor derived factors. a, Schematic drawing illustrates the in-situ assembled peptide nano-blanket interrupts the activation of fibroblast. When activated by tumor derived factors during PMN development, the expression of proangiogenic factors and ECM remodeling factors would be up-regulated in fibroblast, as well as the ECM components production. While the peptide nano-blanket could calm down fibroblast activation, down-regulating the above factors. b, Schematic illustration to show the protocol of MCM stimulation and peptide treatment on mice lung fibroblast (MLF) in vitro. c, Cell proliferation of MLF after MCM stimulation and peptide treatment. n = 4. d, Secretion of MMP9 (n =4), VEGF (n = 4 for peptide treated groups and 3 for the control groups) and fibronectin (n =4)
+in the culture media of MLF after stimulated by MCM, treated with or without peptide. e, qPCR analysis of Acta2 (i.e., αSma), Mmp9, Vegfa, Fibronectin1 (Fn1) expression in MLF after MCM stimulation and peptide treatment. f & g, Migration assay to evaluate MLF migration ability after MCM stimulation and peptide treatment. n = 4. Scale bar = 100 μm. h & i, Collagen gel contract assay to evaluate contracting function of MLF after MCM stimulation and peptide treatment. n = 3. j, Expression level of fibronectin in the lung harvested from the PMN model mice administrated with different peptides on Day 10. k, Representative images and semi-quantification of αSMA+ fibroblasts in the lung harvested from mice administrated with different peptides on Day 10. Data is presented as mean ± SD. n = 6. One-way ANOVA followed by Tukey’s multiple comparisons test was employed for statistical evaluation. Scale bar = 50 μm.
+
+8) Minor point: In figure 4d, MCM free control also shows big gap in the current image. Quantitative analysis is needed and an image that represents the quantitative data needs to be replaced.
+
+Response: As presented in Figure 4d and Figure IX with enhanced brightness, endothelial cells tightly attached to each other with the distribution of strong VE-cadherin signals along the contact membrane in MCM free control group. It is probable that the cell that sprawled out in the middle of the enlarged image might has been confused with a big gap with low brightness. On the contrary, VE-cadherin signals along the boundary of the intercellular spaces left out by neighboring endothelial cells were lost in MCM and TP5 groups. To reflect the re-distribution of VE-cadherin on cell membrane, fluorescent signals alongside the diagonal line (as indicated in semitransparent yellow oblique line on the lower panel) were analyzed via ImageJ. The no-signal zones labeled in light golden represent for the intercellular gaps contributed by the dismissal of VE-cadherin, indicating the breakage in the integrality of endothelial. Quantitatively, the disruption on endothelial cell-cell connection led to dye leakage in endothelial transwell permeability assay (Figure X). In short, MLFs pre-treated with MCM and peptides were seeded at the lower well. The endothelial cells were seeded on a 0.4 μm Transwell insert above the top of the well until grown to confluence. Rhodamine B-dextran was added to the upper insert on the endothelial cell layer. After 1 h incubation, the translocation of Rhodamine B-dextran from the insert to the lower well leaking through the endothelial cell layer was measured to reveal the relative permeability of the cell layer.
+Figure IX, Integrality of the endothelial cell monolayer after cultivated with conditional FCM collected from MCM and peptide pre-stimulated MLF, indicated by VE-cadherin on the membrane. a, The white box and the white arrows in the enlarged images indicate the tight junction between the endothelial cells. While the yellow box and yellow arrows indicate the disruption of cell-cell connection. Scale bar = 50 μm. b, The distribution of VE-cadherin on cell membrane, fluorescent signals alongside the diagonal line (as indicated in semitransparent yellow oblique line) were analyzed via ImageJ.
+
+Figure X, Schematic illustration of the transwell permeability assay. Permeability of the endothelial cell layer in vitro when co-cultured with MLF pre-treated with MCM and peptide. The relative permeability was normalized by diving the fluorescence signals of the treatment groups by the control group. Data is presented as mean ± SD. n = 3. One-way ANOVA followed by Tukey’s multiple comparisons test was employed for data analysis.
+9) Minor point: Although bone-marrow recruited cells, especially MDSCs, are analyzed in the data, not much is mentioned in the introduction. It would be helpful for the readers to include more information on why this analysis is necessary.
+
+Response: Thank you for your suggestion. The introduction of how MDSCs take part in PMN and metastasis has been added as follow:
+
+“In 2005, Lyden et al. brought to light the recruitment of VEGFR+ myeloid progenitor cells to PMN by localized FN deposition36, which would be over-produced by activated resident fibroblasts. This specific cell population and its subtypes were then unified and classified as MDSCs with its potent capability to suppress immune responses37, who make major contributions in developing immunosuppressive microenvironment via activation of nitric oxide (NO) signaling or reactive oxygen species (ROS) pathway38.”
+
+Besides, necessity of the researches on MDSC in our study has been brought up in the relevant sections in Page 16 as follow:
+
+“Given the fact that MDSC occupies a vital position in PMN construction and metastasis formation, MDSC has attracted wide concerns in recent years…The recruitment of MDSC was found to be related to the enhancement of ECM production and remodeling in PMN, including cross-linking collagen by LOX secreted by tumor cells39, FN deposition33, periostin (POSTN)-enrichment40. As MDSC plays a crucial role in developing immunosuppressive and inflammatory microenvironment, some pioneers have demonstrated that blocking the recruitment of MDSC could be a promising strategy to suppress early metastasis by preventing the development of breeding ground for tumor41,42,43,44.”
+Reviewer #3 (Remarks to the Author):
+
+In this manuscript, the authors constructed an enzyme-activatable peptide FR17 to release self-assembly monomer FG8 to form nano-blanket in PMN microenvironment for impeding fibroblasts activation and preventing metastatic cascades. The design is interesting, and the authors did comprehensive work to justify their hypothesis. The presentation of the work is also clear and neat. The following comments can be considered for improvement.
+
+1. In page 3 line 14, the authors stated that nano-blanket in PMN can impede fibroblasts activation to prevent metastatic. Why did the authors choose fibroblasts as the target of intervention? As primary tumor-derived cytokines and exosomes and myeloid-derived suppressor cells (MDSCs) all contribute to metastasis, it is necessary to provide the background about the key factors influencing PMN microenvironment, and discuss the role of fibroblasts in PMN and the current studies made in this field based on fibroblasts.
+
+Response: Thank you for your suggestion. As we mentioned in the introduction, there’re several main participators contribute the construction of PMN, including 1) primary tumor-derived cytokines and exosomes, 2) myeloid-derived suppressor cells (MDSCs), and 3) the tumor re-educated stromal environment, which consists of pre-metastasis associated fibroblasts, destabilized vasculature and extracellular matrix (ECM). Though few researches have focused on PMN-associated fibroblasts, cancer-associated fibroblasts have received considerable attention in the past decades as accomplices of cancer cells to promote tumor development and to establish the metastatic niche\(^{45, 46, 47}\). To organize the above defining factors in a clearer way, we’ve revised the introduction as follow:
+
+“Relevant studies revealed that complex interactions between multiple participators and alteration in regulative pathways energized the construction of PMN, such as primary tumor-derived cytokines and exosomes, myeloid-derived suppressor cells (MDSCs), and the tumor re-educated stromal environment including pre-metastasis associated fibroblasts, destabilized vasculature and extracellular matrix (ECM) \(^{12, 13}\) (Fig. 1a). In 2005, Lyden et al. brought to light the recruitment of VEGFR\(^+\) myeloid progenitor cells to PMN by localized FN deposition\(^{33}\), which would be over-produced by activated resident fibroblasts. This specific cell population and its subtypes were then unified and classified as MDSCs with its potent capability to suppress immune responses\(^{48}\), who make major contributions in developing immunosuppressive microenvironment via activation of nitric oxide (NO) signaling or reactive oxygen species (ROS) pathway\(^{49}\). What’s more, recently studies revealed the irritation of stromal cells especially fibroblasts in primary tumor and in distal site induced by tumor-derived secreted factors via STAT3 signaling\(^{35}\) and JNK signaling\(^{34}\)
+pathways. It’s reported that tumor-educated fibroblasts serve to reconstruct ECM, induce angiogenic and pro-inflammatory response of endothelial cells, preparing a tumor supportive host stromal13, which gives a clue to the tipping point of PMN initializing as fibroblast activation.”
+
+2. Is there any direct evidence about the production of FG8 in the presence of MMP2?
+
+Response: Thank you for your advice. LC-MS/MS was applied to verify the production of FG8 from FR17 or sFD17 after MMP2 treatment. Briefly, peptide FR17 or sFD17 (200 \( \mu \)M) was cultivated with pre-activated hMMP2 (200 ng/ml) for 10 min. The reaction was terminated by adding 3 volume of methanol and centrifuged at 1,3000 rpm for 15 min to discard the protein precipitation. Samples were analyzed by LC-MS/MS. Reference FG8 (the MS and MS/MS of FG8 has been added in Supplementary Figure 1 & 2 in the revised manuscript), and FR17 or sFD17 without enzyme treatment were analyzed as standard control. As demonstrated by Supplementary Figure 3 as follow, the retention time of FG8 (tR = 4.9 min) is ahead of FR17 (tR = 5.5 min) or sFD17 (tR = 5.3 min) in chromatogram. After MMP2 cultivation for 10 min, the characteristic peak of FG8 aroused. What’s more, the peptide sequence produced by MMP2 degradation was able to be identified under the auto-optimized condition of the ion pair 464.8/379.6, in which 464.8 represents [M+2H]^{2+} of FG8 (Exact mass 927.49) and 379.6 represents [M+2H]^{2+} of FFK(GP)Y (Exact mass 757.38). By calculation, the responsive rate of MMP2 to FR17 is 34.28% in 10 min, the responsive rate of MMP2 to sFD17 is 35.82% in 10 min.
+Supplementary Figure 3. Enzyme cleavage of FR17 and sFD17 to release the self-assembled monomer FG8. a, Schematic of the enzyme cleavage of FR17 or sFD17. b, Liquid chromatography-tandem mass spectrometry (LC-MS/MS) of FR17 and sFD17 (100 μM) before and after MMP2 (200 ng/ml) treatment. The characteristic peak of FG8 was acquired under the auto-optimized condition of the ion pair 464.8/379.6, in which 464.8 represents [M+2H]^{2+} of FG8 (Exact mass 927.49) and 379.6 represents [M+2H]^{2+} of FFK(GP)Y (Exact mass 757.38).
+
+3. In supplementary Figure 4, FG8 was shown to achieve self-assembly in water. Did the authors investigate self-assembly of FG8 under different pH conditions? The authors should consider the condition in PMN and investigate the self-assembly of FG8 under the corresponding condition.
+
+Response: We appreciate you for arising such a good question. In the revision, we’ve compared the self-assembly property of FG8 in weak acidic condition to neutral pH condition. Since the biochemical condition in PMN has not been revealed or widely accepted, we’ve decided to adjust the pH value to 6.8 as a simulation condition of PMN *in vitro*, because a myriad of studies show that localized interstitial acidosis (pH 6.5–6.8) is a biochemical hallmark in inflammatory tissues^{50}, ^51, ischemic organs^{52}, and solid tumors^{53}. FG8 assembled in water, PBS (pH 7.4), PBS (pH 6.8) or
+dissolved in methanol at the concentration of 100 \( \mu \)M gave similar performance on size distribution (100-200 nm, which is only for reference for the Zetasizer size meter is more applied to regular nanoparticles) and TEM morphology (the thin layer/membrane-like structure), indicating that FG8 is able to achieve assembly in both neutral and weak acidic conditions. Besides, TPE labeled FG8 with aggregation-induced emission effect was also employed to reflect the spontaneous aggregation of FG8 in different pH conditions. As illustrated in Supplementary Figure 5b, the aggregation of FG8 presented slightly stronger fluorescent signal in pH 6.8 than in neutral condition at the same concentration. Therefore, the following data provided evidence for the adaptability of the self-assembly of FG8 in different conditions.
+
+![Size distribution and fluorescence spectra of FG8 assemblies in different conditions](page_340_670_1067_393.png)
+
+Supplementary Figure 5. Self-assembly of FG8 in different pH conditions. a, Size distribution of the peptide assemblies of FG8 formed in water or PBS (pH 6.8 or pH 7.4). b, Fluorescence spectra of the TPE-FG8 (100 \( \mu \)M) in water, PBS (pH 6.8 or pH 7.4) or methanol (dissolved) excited by 405 nm. c, TEM images of FG8 assemblies in water, PBS (pH 6.8 or pH 7.4) or methanol (dissolved). Scale bar = 200 nm.
+
+4. Please provide semi-quantitative data in Figure 3g, 4e, 5c, supplementary Figure 11 and 12.
+
+Response: Thank you for your suggestion to improve the manuscript. The semi-quantitative data of Figure 3g, 4e, 5c, supplementary Figure 11 and 12 has been provided in the revised manuscript as follow.
+Figure 3g (Labeled as Figure 3j in the revised manuscript)
+
+![Western blot analysis of fibronectin and GAPDH](page_246_120_563_180.png)
+
+Figure 4e
+
+![Western blot analysis of VEGFa, ANG2, MMP9, MMP2, and GAPDH](page_246_420_563_180.png)
+
+Figure 5c
+
+![Western blot analysis of TGF-β1 and GAPDH](page_246_720_563_180.png)
+Supplementary Figure 11 (Labeled as Supplementary Figure 15 in the revised manuscript). FR17 administration down-regulated matrix metalloproteinase in pulmonary PMN. Representative images of MMP2 and MMP9 in the lung harvested from the PMN model mice administrated with different peptides. Semi-quantification was calculated from six random fields via ImageJ. Scale bar = 50 μm.
+
+Supplementary Figure 12 (Labeled as Supplementary Figure 16 in the revised manuscript). FR17 administration prevented MDSC recruitment to pulmonary PMN by influencing the extracellular matrix remodeling. a, Images under low magnification ratio of the serial sections of the lungs harvested from the model mice treated with different peptides were taken. Serial sections show the co-location and distribution of periostin (POSTN, red), lysyl oxidase (LOX, green) and Fibronectin (FN, red), CD11b+Gr1+ MDSC (yellow merged from green and red). Scale bar = 200 μm. b, The co-location analysis of CD11b+Gr1+ MDSC and POSTN, LOX, FN
+alongside the yellow arrow marked on the above panel via ImageJ.
+
+5. Why did the authors choose subcutaneous injection to administrate FR17 instead of intravenous injection? Please explain.
+
+Response: As the most common administration route for peptides in clinic, subcutaneous injection is cost-effective and suitable for self-administration and repeated injections\(^{54}\). More importantly, the short half-life in the bloodstream is the hallmark of peptide pharmacokinetics contributed by proteolytic cleavage by proteases and peptidases, renal clearance, liver metabolism and immunogenicity, especially for the peptides with high hydrophily\(^{55}\). Compared to intravenous administration, drug could be sustained-released from the temporary formed drug storage under the skin after subcutaneous administration, maintaining serum concentrations and extending retention time\(^{56, 57, 58}\). Moreover, the subcutaneous administration is bioequivalent in effect to intravenous injection, albeit with less between-patient variability\(^{59}\). Overall, as a hydrophilic short chain peptide, FR17 was applied subcutaneously in this study to prolong its physiological effect *in vivo*.
+
+References
+
+1 Ekiz MS, Cinar G, Khalily MA, Guler MO. Self-assembled peptide nanostructures for functional materials. *Nanotechnology* **27**, 402002 (2016).
+
+2 Abbas M, Zou Q, Li S, Yan X. Self-Assembled Peptide- and Protein-Based Nanomaterials for Antitumor Photodynamic and Photothermal Therapy. *Adv Mater* **29**, 1-16 (2017).
+
+3 Loo Y, et al. Self-Assembled Proteins and Peptides as Scaffolds for Tissue Regeneration. *Adv Healthc Mater* **4**, 2557-86 (2015).
+
+4 Tanaka A, et al. Cancer cell death induced by the intracellular self-assembly of an enzyme-responsive supramolecular gelator. *J Am Chem Soc* **137**, 770-5 (2015).
+
+5 Feng Z, Han X, Wang H, Tang T, Xu B. Enzyme-Instructed Peptide Assemblies Selectively Inhibit Bone Tumors. *Chem* **5**, 2442-2449 (2019).
+
+6 Guo, WW, et al. Intracellular restructured reduced glutathione-responsive peptide nanofibers for synergetic tumor chemotherapy. *Biomacromolecules* **21**, 444-453 (2020).
+
+7 Hu XX, et al. Transformable Nanomaterials as an Artificial Extracellular Matrix for Inhibiting Tumor Invasion and Metastasis. *ACS Nano* **11**, 4086-4096 (2017).
+
+8 Ben-Nun Y, et al. Cathepsin nanofiber substrates as potential agents for targeted drug delivery. *J Control Release* **257**, 60-67 (2017).
+
+9 Chen W, et al. Combined Tumor Environment Triggered Self-Assembling Peptide Nanofibers and Inducible Multivalent Ligand Display for Cancer Cell Targeting with Enhanced Sensitivity and Specificity. *Small* **16**, e2002780 (2020).
+10 Wang, TT, et al. AIE/FRET-based versatile PEG-Pep-TPE/DOX nanoparticles for cancer therapy and real-time drug release monitoring. Biomater. Sci. 8, 118-124 (2020).
+
+11 Reches M, Gazit E. Casting metal nanowires within discrete self-assembled peptide nanotubes. Science 300, 625-7 (2003).
+
+12 Tjernberg LO, et al. Arrest of beta-amyloid fibril formation by a pentapeptide ligand. J Biol Chem 271, 8545-8 (1996).
+
+13 Chandra Saha P, Das RS, Chatterjee T, Bhattacharyya M, Guha S. Supramolecular β-Sheet Forming Peptide Conjugated with Near-Infrared Chromophore for Selective Targeting, Imaging, and Dysfunction of Mitochondria. Bioconjug Chem 31, 1301-1306 (2020).
+
+14 Cheng DB, et al. Endogenous Reactive Oxygen Species-Triggered Morphology Transformation for Enhanced Cooperative Interaction with Mitochondria. J Am Chem Soc. 141, 7235-7239 (2019).
+
+15 Pederzoli F, et al. Nanomedicine Against Aβ Aggregation by β-Sheet Breaker Peptide Delivery: In Vitro Evidence. Pharmaceutics 11, 572 (2019).
+
+16 Luo S, et al. Targeting self-assembly peptide for inhibiting breast tumor progression and metastasis. Biomaterials 249, 120055 (2020).
+
+17 Versluis F, van Esch JH, Eelkema R. Synthetic Self-Assembled Materials in Biological Environments. Adv Mater. 28, 4576-92 (2016).
+
+18 Ben-Nun Y, et al. Cathepsin nanofiber substrates as potential agents for targeted drug delivery. J Control Release 257, 60-67 (2017).
+
+19 Mao L, et al. Conjugation of two complementary anti-cancer drugs confers molecular hydrogels as a co-delivery system. Chem Commun (Camb) 48, 395-7 (2012).
+
+20 Wang H, et al. Integrating Enzymatic Self-Assembly and Mitochondria Targeting for Selectively Killing Cancer Cells without Acquired Drug Resistance. J Am Chem Soc 138,16046-16055 (2016).
+
+21 Gao Y, Shi J, Yuan D, Xu B. Imaging enzyme-triggered self-assembly of small molecules inside live cells. Nat Commun. 3, 1033 (2012).
+
+22 Brooks PC, et al. Localization of matrix metalloproteinase MMP-2 to the surface of invasive cells by interaction with integrin αvβ3. Cell 85, 683–693 (1996).
+
+23 Egeblad M, Werb Z. New functions for the matrix metalloproteinases in cancer progression. Nat Rev Cancer 2,161-74 (2002).
+
+24 Costa-Silva B, et al. Pancreatic cancer exosomes initiate pre-metastatic niche formation in the liver. Nat Cell Biol. 17, 816-26 (2015).
+
+25 Houg DS, Bijlsma MF. The hepatic pre-metastatic niche in pancreatic ductal adenocarcinoma. Mol Cancer 17, 95 (2018).
+
+26 Zhao S, et al. Highly-metastatic colorectal cancer cell released miR-181a-5p-rich extracellular vesicles promote liver metastasis by activating hepatic stellate cells and remodelling the tumour microenvironment. J Extracell Vesicles 11, e12186 (2022).
+
+27 Kaplan RN, et al. VEGFR1-positive haematopoietic bone marrow progenitors initiate the pre-metastatic niche. Nature 438, 820-827 (2005).
+28 Peinado H, et al. Pre-metastatic niches: organ-specific homes for metastases. Nat Rev Cancer **17**, 302-317 (2017).
+
+29 Aguado BA, Bushnell GG, Rao SS, Jeruss JS, Shea LD. Engineering the pre-metastatic niche. Nat Biomed Eng **1**, 0077 (2017).
+
+30 Erler, JT, et al. Hypoxia-induced lysyl oxidase is a critical mediator of bone marrow cell recruitment to form the pre-metastatic niche. Cancer Cell **6**, 35–44 (2009).
+
+31 Peinado H, et al. Pre-metastatic niches: organ-specific homes for metastases. Nat Rev Cancer **17**, 302-317 (2017).
+
+32 Zhang Z, Ai S, Yang Z, Li X. Peptide-based supramolecular hydrogels for local drug delivery. Adv Drug Deliv Rev. **174**, 482-503 (2021).
+
+33 Zhao XX, et al. In Situ Self-Assembled Nanofibers Precisely Target Cancer-Associated Fibroblasts for Improved Tumor Imaging. Angew Chem Int Ed Engl. **58**, 15287-15294 (2019).
+
+34 Cascinelli N, Belli F, Mascheroni L, Lenisa L, Clemente C. Evaluation of clinical efficacy and tolerability of intravenous high dose thymopentin in advanced melanoma patients. Melanoma Res. **8**, 83-9 (1998).
+
+35 Zeng FL, et al. Clinical efficacy and safety of synthetic thymic peptides with chemotherapy for non-small cell lung cancer in China: A systematic review and meta-analysis of 27 randomized controlled trials following the PRISMA guidelines. Int Immunopharmacol. **75**, 105747 (2019).
+
+36 Kaplan RN, et al. VEGFR1-positive haematopoietic bone marrow progenitors initiate the pre-metastatic niche. Nature **438**, 820-827 (2005).
+
+37 Gabrilovich DI, et al. The terminology issue for myeloid-derived suppressor cells. Cancer Res. **67**, 425 author reply 6 (2007).
+
+38 Gabrilovich, DI. Myeloid-derived suppressor cells. Cancer Immunol. Res. **5**, 3-8 (2017).
+
+39 Erler, J. T. et al. Hypoxia-induced lysyl oxidase is a critical mediator of bone marrow cell recruitment to form the premetastatic niche. Cancer Cell **15**, 35–44 (2009).
+
+40 Wang, Z. et al. Periostin promotes immunosuppressive premetastatic niche formation to facilitate breast tumour metastasis. J. Pathol. **239**, 484-495 (2016).
+
+41 Long, Y. et al. Self-delivery micellar nanoparticles prevent premetastatic niche formation by interfering with the early recruitment and vascular destruction of granulocytic myeloid-derived suppressor cells. Nano Lett. **20**, 2219-2229 (2020).
+
+42 Jiang, T. et al. Metformin and Docosahexaenoic Acid Hybrid Micelles for Premetastatic Niche Modulation and Tumor Metastasis Suppression. Nano lett. **19**, 3548-3562 (2019).
+
+43 Lu, Z et al. Epigenetic therapy inhibits metastases by disrupting premetastatic niches. Nature **579**, 284-290 (2020).
+
+44 Kaczanowska, S. et al. Genetically engineered myeloid cells rebalance the core immune suppression program in metastasis. Cell **184**, 2033-2052 (2021).
+
+45 Kong J, et al. Extracellular vesicles of carcinoma-associated fibroblasts creates a pre-metastatic niche in the lung through activating fibroblasts. Molecular Cancer **18**, 175 (2019).
+46 Pein M, et al. Metastasis-initiating cells induce and exploit a fibroblast niche to fuel malignant colonization of the lungs. Nat Commun. **11**, 1494 (2020).
+
+47 Zhou X, et al. Melanoma cell-secreted exosomal miR-155-5p induce proangiogenic switch of cancer-associated fibroblasts via SOCS1/JAK2/STAT3 signaling pathway. J. Exp. Clin. Cancer Res. **37**, 242 (2018).
+
+48 Gabrilovich DI, et al. The terminology issue for myeloid-derived suppressor cells. Cancer Res. **67**, 425 author reply 6 (2007).
+
+49 Gabrilovich, DI. Myeloid-derived suppressor cells. Cancer Immunol. Res. **5**, 3-8 (2017).
+
+50 Lardner A. The effects of extracellular pH on immune function. J Leukoc Biol. **69**, 522-30 (2001).
+
+51 Riemann A, et al. Acidic environment activates inflammatory programs in fibroblasts via a cAMP-MAPK pathway. Biochim Biophys Acta **1853**, 299-307 (2015).
+
+52 Xiong ZG, et al. Neuroprotection in ischemia: blocking calcium-permeable acid-sensing ion channels. Cell **118**, 687-98 (2004).
+
+53 Gatenby RA, Gillies RJ. Why do cancers have high aerobic glycolysis? Nat Rev Cancer. **4**, 891-9 (2004).
+
+54 Kovalainen M, et al. Novel delivery systems for improving the clinical use of peptides. Pharmacol Rev. **67**, 541-61 (2015).
+
+55 Diao L, Meibohm B. Pharmacokinetics and pharmacokinetic-pharmacodynamic correlations of therapeutic peptides. Clin Pharmacokinet. **52**, 855-68 (2013).
+
+56 Sennello LT, et al. Single-dose pharmacokinetics of leuprolide in humans following intravenous and subcutaneous administration. J Pharm Sci. **75**, 158-60 (1986).
+
+57 Wills RJ, Dennis S, Spiegel HE, Gibson DM, Nadler PI. Interferon kinetics and adverse reactions after intravenous, intramuscular, and subcutaneous injection. Clin Pharmacol Ther. **35**, 722-7 (1984).
+
+58 Petros WP. Pharmacokinetics and administration of colony-stimulating factors. Pharmacotherapy **12**, 32S-38S (1992).
+
+59 Mannucci PM, et al. Intravenous and subcutaneous administration of desmopressin (DDAVP) to hemophiliacs: pharmacokinetics and factor VIII responses. Thromb Haemost. **58**, 1037-9 (1987).
+REVIEWERS’ COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+This manuscript has improved considerably, although the evidence of nano-blanket could be sounder. In principle, I would support the acceptance of this work. It, however, baffles me that the authors appear to be evasive on referencing the prior works done on FFKY. The authors wrote the following paragraph in the rebuttal to acknowledge the prior use of FFKY, but still refused to reference those works properly. I think this could be ethically problematic. I would suggest the authors properly reference the use of FFKY in the first place when FFKY appears in main text.
+
+“In addition, as the original fragment of Aβ16-20, KLVFF has been applied as therapeutic agents7, 12, imaging agents13 or delivery platform14 in the field of Alzheimer's disease15 and cancer16. Further researches cut the peptide fragment down to dipeptide FF to obtain discrete nanotubes through self-assembly in aqueous solution1. Afterwards, felicitous modification and re-designment on FF with prefect biocompatibility has been reported with a wide range of applications17. For example, FFKF was developed to construct drug delivery system which can be complete degraded by cathepsin proteases18. Yang and his team employed FFFK to develop molecular hydrogel for codelivery of anti-cancer drugs19. In another work, the application of FFYK was explored in organelles targeting and cancer cell killing20. Some introduced naphthyl group on N terminal of Phe to favor intercellular hydrophobic interactions21. Here in our manuscript, taking advantages of both the self-assembly feature of diphenylalanine structural motif FF with the assistance of Y and the editable site provided by K to combine with hydrophilic fragment via enzyme-cleavable linker, FFKY was employed as the backbone of the self-assembled monomer FG8, which would spontaneously fold into lamellar structure, constructing the peptide nano-blanket.”
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have overall answered all of my concerns.
+This reviewer would think that Figure II. In the rebuttal letter as well as organ distribution in the lung, liver, and other organs should be placed in the paper.
+Also, pseudo coloring helped the visualization of the nanoblanket for researchers outside of the field. It would be nice to include an image with pseudo coloring as well on the side.
+
+** See Nature Portfolio's author and referees' website at www.nature.com/authors for information about policies, services and author benefits
+Reviewer #1 (Remarks to the Author):
+
+This manuscript has improved considerably, although the evidence of nano-blanket could be sounder. In principle, I would support the acceptance of this work. It, however, baffles me that the authors appear to be evasive on referencing the prior works done on FFKY. The authors wrote the following paragraph in the rebuttal to acknowledge the prior use of FFKY, but still refused to reference those works properly. I think this could be ethically problematic. I would suggest the authors properly reference the use of FFKY in the first place when FFKY appears in main text.
+
+“In addition, as the original fragment of Aβ_{16-20}, KLVFF has been applied as therapeutic agents^{7, 12}, imaging agents^{13} or delivery platform^{14} in the field of Alzheimer's disease^{15} and cancer^{16}. Further researches cut the peptide fragment down to dipeptide FF to obtain discrete nanotubes through self-assembly in aqueous solution^{1}. Afterwards, felicitous modification and re-designment on FF with prefect biocompatibility has been reported with a wide range of applications^{17}. For example, FFKF was developed to construct drug delivery system which can be complete degraded by cathepsin proteases^{18}. Yang and his team employed FFFK to develop molecular hydrogel for codelivery of anti-cancer drugs^{19}. In another work, the application of FFYK was explored in organelles targeting and cancer cell killing^{20}. Some introduced naphthyl group on N terminal of Phe to favor intercellular hydrophobic interactions^{21}. Here in our manuscript, taking advantages of both the self-assembly feature of diphenylalanine structural motif FF with the assistance of Y and the editable site provided by K to combine with hydrophilic fragment via enzyme-cleavable linker, FFKY was employed as the backbone of the self-assembled monomer FG8, which would spontaneously fold into lamellar structure, constructing the peptide nano-blanket.”
+
+Response: Thank you for your suggestion. To fully present the entire process of how the idea of FR17 was form based on FFKY to our readership, we’re introduced the origin and development of the backbone of the assembly peptide, FFKY, in the Introduction section as follow:
+
+“One of the research branches, which has been widely applied, was based on the β-sheet regions of Amyloid-β (Aβ)^{30}. As the original fragment of Aβ_{16-20}, KLVFF has been applied as therapeutic agents^{26, 31}, imaging agents^{32} or delivery platform^{33} in the field of Alzheimer's disease^{34} and cancer^{35}. Further researches cut the peptide fragment down to dipeptide FF to obtain discrete nanotubes through self-assembly in aqueous solution^{36}. Afterwards, felicitous modification and re-designment on FF with prefect biocompatibility has been reported with a wide range of applications^{37}. For example, FFKF was developed to construct drug delivery system which can be complete degraded by cathepsin proteases^{37}. Yang and his team employed FFFK to develop molecular hydrogel for codelivery of anti-cancer drugs^{38}. In another work, the application of FFYK was explored in organelles targeting and cancer cell killing^{39}. Some introduced naphthyl group on N terminal of Phe to favor intercellular hydrophobic interactions of FFKY^{40}.”
+
+“The self-assembly feature of FG8 is provided by its backbone, FFKY, taking advantages of both
+the self-assembly feature of diphenylalanine structural motif FF with the assistance of Y and the editable site provided by K to combine with hydrophilic fragment via enzyme-cleavable linker.”
+
+Reviewer #3 (Remarks to the Author):
+
+The authors have overall answered all of my concerns.
+
+This reviewer would think that Figure II in the rebuttal letter as well as organ distribution in the lung, liver, and other organs should be placed in the paper.
+
+Also, pseudo coloring helped the visualization of the nano-blanket for researchers outside of the field. It would be nice to include an image with pseudo coloring as well on the side.
+
+Response: We appreciate for your acknowledgement on our work. According to your suggestion, Figure II in the last response letter and inserted in the paper as Supplementary Figure 5. The pseudo coloring images have been inserted into Figure 2 to give a direct impression of the morphology of peptide nano-blanket on the readership as follow.
+
+![Cryo-TEM and macroscopic images of FR17 and sFD17 nano-blankets](page_349_682_1047_377.png)
+
+Supplementary Figure 5. Enzyme-activated assembly of FR17 and sFD17. a, Cryo-TEM image of the peptide nano-blanket assembled by FR17 (500 μM) treated with MMP2 (1 μg/ml) for 24 h. Scale bar = 200 nm. b, Macroscopic images of the thin layer formed by FR17 treated with enzyme and let stand for 72 h. The soft thin layer broke into pieces after gently shaking and
+dispersed into nanoscale fragments after sharply shaking. c, STEM images of the peptide nano-blanket in the intercellular substance in PMN lung, which was collected from PMN mouse at 12 h-post subcutaneous administration of FR17. Scale bar = 1 \( \mu \)m. The lamellar structure of the peptide nano-blanket was pseudo-colored in gold.
+
+![TEM images and molecular structures of peptide nano-blankets and assemblies](page_184_355_1080_1016.png)
+
+Figure 2. Enzyme-activated self-assembly of FR17 and all-atom molecular dynamics (MD) simulation of the self-assembly of FG8. a, TEM images of FR17 or sFD17 (500 \( \mu \)M) treated with MMP2 (scale bar = 200 nm), FG8 (500 \( \mu \)M, scale bar = 100 nm) and FFKY (500 \( \mu \)M, scale bar = 100 nm). The lamellar structure of the peptide nano-blanket was pseudo-colored in gold. b, Size distribution of FR17 and sFD17 (500 \( \mu \)M) after MMP2 cleavage. c, Molecular structure of FFKY. d, The FFKY assemblies generated at t = 200 ns of MD simulation of 16 FFKY molecules in water (containing NaCl for charge neutralization). e, Typical molecular cluster and the non-bonded interactions involved in FFKY assemblies in detail. f, Molecular structure of FG8. g, The FG8 assemblies generated at t = 200 ns of MD simulation of 16 FG8 molecules in water (containing NaCl for charge neutralization). h, Typical molecular cluster and the non-bonded interactions involved in FG8 assemblies in detail. The dashed purple lines denote the \( \pi \)-\( \pi \) stacking. The dashed blue lines denote the hydrogen bonds. The nitrogen atoms are labeled in blue. And the oxygen atoms are labeled in red. Source data are provided as a Source Data file.
+
+In addition, organ distribution of peptide nano-blanket indicated by the aggregation-induced
+emission (AIE) effect observed in different organs’ sections at 12 h-post TPE-FR17 administration has been presented in Supplementary Fig. 7-8 as follow. Besides, the further discussion on the peptide assemblies observed both in the PMN lung and liver, as well as the time-dependent aggregation and degradation of peptide nano-blanket is achievable in the last response letter, which will be revealed to the readership accompanied the paper as Peer Review File.
+
+“Fluorescent dots observed in the organs’ sections revealed the responsive-assembly of the peptide nano-blanket in PMN lung in vivo (Supplementary Fig. 7) with few aggregations in PMN liver and no observation of AIE effect in other organs (Supplementary Fig. 8).”
+
+![Fluorescent and merged channel images of healthy lung and PMN lung sections, showing peptide nano-blanket assembly and aggregation indicated by red arrows.](page_354_629_1042_627.png)
+
+Supplementary Figure 7. The aggregation-induced emission effect of the in-situ assembly of peptide nano-blanket in the lung in vivo. Lung secessions at 12 h-post subcutaneous administration of TPE-FR17 (100 μM/kg) in healthy or PMN lung. Peptide assemblies of the monomer TPE-FG8 released from TPE-FR17 in PMN were pseudo-colored in white on the upper panel and pseudo-colored in red in the merged images, and indicated by red arrows for contrast. Scale bar = 200 μm.
+Supplementary Figure 8. The aggregation-induced emission effect of the in-situ assembly of peptide nano-blanket in heart, spleen, kidney and liver in vivo. Organs were collected at 12 h-post subcutaneous administration of TPE-FR17 (100 μM/kg) to healthy or PMN mice. Peptide assemblies of the monomer TPE-FG8 released from TPE-FR17 in PMN were pseudo-colored in white on the upper panel and pseudo-colored in red in the merged images, and indicated by red arrows for contrast. Scale bar = 200 μm.

0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/preprint/preprint.md

@@ -0,0 +1,410 @@
+Peptide nano-blanket impedes fibroblasts activation and subsequent formation of pre-metastatic niche
+
+Yi Zhou
+Zhejiang University
+Peng Ke
+Shengli Clinical Medical College of Fujian Medical University
+Yiyi Xia
+Zhejiang University
+Honghui Wu
+Zhejiang University
+Zhentao Zhang
+Zhejiang University
+Haiqing Zhong
+Zhejiang University
+Qi Dai
+Zhejiang University
+Tiantian Wang
+Zhejiang University
+Mengting Lin
+Zhejiang University
+Yaosheng Li
+Zhejiang University
+Xinchi Jiang
+Zhejiang University
+Qiyao Yang
+Zhejiang University
+Yiying Lu
+Zhejiang University
+Xincheng Zhong
+Zhejiang University
+Min Han
+Zhejiang University   https://orcid.org/0000-0001-9373-8466
+Jianqing Gao (gaojianqing@zju.edu.cn)
+Zhejiang University
+Article
+
+Keywords: pre-metastatic niche, self-assembled peptide, pre-metastasis associated fibroblasts, 2 vascular endothelial cells, myeloid-derived suppressor cells, tumor metastasis
+
+Posted Date: August 10th, 2021
+
+DOI: https://doi.org/10.21203/rs.3.rs-699014/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+
+Version of Record: A version of this preprint was published at Nature Communications on May 25th, 2022. See the published version at https://doi.org/10.1038/s41467-022-30634-8.
+Peptide nano-blanket impedes fibroblasts activation and subsequent formation of pre-metastatic niche
+
+Yi Zhou1, Peng Ke1 4, Yiyi Xia1, Honghui Wu1, Zhentao Zhang1, Haiqing Zhong1, Qi Dai1 5, Tiantian Wang1, Mengting Lin1, Yaosheng Li1, Xinchi Jiang1, Qiyao Yang1 5, Yiying Lu1, Xincheng Zhong1, Min Han1 2 3*, Jianqing Gao1 2 3*
+
+1 Institute of Pharmaceutics, Zhejiang Province Key Laboratory of Anti-Cancer Drug Research, College of Pharmaceutical Sciences, Zhejiang University, Hangzhou 310058, P.R. China
+2 Cancer Center of Zhejiang University, Zhejiang University, Hangzhou 310058, P.R. China.
+3 Hangzhou Institute of Innovative Medicine, Zhejiang University, Hangzhou 310058, P.R. China.
+4 Shengli Clinical Medical College of Fujian Medical University, Fuzhou, 350001, China.
+5 Department of Radiation Oncology, Key Laboratory of Cancer Prevention and Intervention, The Second Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou 310058, China.
+
+* Corresponding author: hanmin@zju.edu.cn (M. H.); gaojianqing@zju.edu.cn (J. G.)
+Keywords: pre-metastatic niche, self-assembled peptide, pre-metastasis associated fibroblasts, vascular endothelial cells, myeloid-derived suppressor cells, tumor metastasis
+
+Abstract
+
+In various types of malignant tumors, metastasis is responsible for most of the tumor-induced death. Though emerging technologies provided early detection of tumor metastasis or even warning of high metastatic risk before the actual occurrence of metastasis, clinical treatment on metastasis prevention lags far behind. Evidences have illustrated that primary tumor induced pre-metastatic niche (PMN) formation in distal organs by producing pro-metastasis factors, spreading the spark to ignite the distal microenvironment. Given the fundamental role of PMN in the development of metastases, interruption of PMN formation would be a promising strategy to take early actions against tumor metastasis. Here we report an enzyme-activatable assembled peptide FR17 that can serve as a “flame-retarding blanket” at PMN site specifically to extinguish the “fire” of tumor-supportive microenvironment adaption. Our experiment demonstrated that the assembled peptide successfully reversed extracellular matrix deposition, vascular leakage and angiogenesis through inhibition on fibroblasts activation in PMN, which suppressed the remodeling of metastasis-supportive host stromal, and further prevented the recruitment of myeloid cells to PMN and then recovered the immunosuppressive microenvironment. Cell transcriptomic analysis of the pulmonary recruited MDSC suggested that FR17 intervention could regulate immune response activation, immune cells chemotaxis and migration pathways. Consequently, FR17 administration effectively inhibited pulmonary PMN formation and postoperative metastasis of melanoma, with only 30% lung-metastasis occurrence was observed for FR17 treated group at the time point when 100% occurrence was observed for the control
+group and 80% occurrence for anti-PD1 treated group, offering a robust therapeutic strategy against PMN establishing to prevent metastasis.
+
+Introduction
+
+Though therapeutic outcomes and survival rate for patients with various cancers have been greatly improved in last decades, effective treatments for patients with metastatic cancer are still limited¹.
+
+Though emerging technologies provided early detection of malignant transformation or even warning of high metastatic risk by biomarker screening before the actual occurrence of metastasis in clinic²⁻⁶, clinical treatment on metastasis prevention lags far behind. The contemporary therapeutic strategies against metastasis in clinic, including systemic chemotherapy, radiotherapy and immunotherapy, mainly focus on the later time period of metastasis development, or at least after the arrival and colonization of disseminated tumor cells to the distal organs, which have gained unsatisfied clinical outcomes⁷⁻⁸. Observations of the adjuvant chemotherapy resistance⁶ and treatment-related toxicities⁹ revealed the shortcoming of the currently available strategies¹⁰.
+
+Growing evidence illustrated that an inflammatory, neovascularized, immunosuppressive, tumor supportive microenvironment has emerged before the arrival and colonization of disseminated tumor cells, which is termed as pre-metastatic niche (PMN) formation¹¹. Relevant studies revealed that complex interactions between multiple participators and alteration in regulative pathways energized the construction of PMN, such as primary tumor-derived cytokines and exosomes, myeloid-derived suppressor cells (MDSCs), and the tumor re-educated stromal environment including pre-metastasis associated fibroblasts, destabilized vasculature and extracellular matrix (ECM)¹²,¹³ (Fig. 1a).
+Since PMN was considered as the foundation laid for circulating tumor cells colonization and one of the vital premises to develop metastasis in distant organs, we wonder if the early process of tumor metastasis can be terminated or even totally prevented by interrupting the formation of PMN. To view the entire process of PMN establishment and metastasis development as a progress of the occurrence and spread of a huge forest fire, it would be more efficient to contain and beat out the local flame by preventing the formation of PMN. Here we report an enzyme-activatable assembled peptide FR17 that can serve as a “flame-retarding blanket” at PMN site specifically, containing and suppressing the “fire blaze” of PMN formation to further develop into overt metastasis (Fig. 1b). As a substrate peptide of matrix metalloproteinase 2 (MMP2), FR17 can release self-assembly monomer FG8 to construct peptide nano-blanket in PMN stromal microenvironment, impeding fibroblasts activation so as to prevent metastatic cascades. We further explored the subsequent impact of inhibition on fibroblasts activation induced by FR17 intervene, revealing the underlining mechanism on cellular interactions among fibroblasts, vascular endothelial cells and extracellular components, and intervention on PMN recruited MDSCs via in vitro and in vivo experiments.
+Figure 1. Schematic illustration of peptide nano-blanket impedes fibroblasts activation and subsequent formation of pre-metastatic niche. a, Illustration of the pathological process of pre-metastatic niche (PMN) formation. The primary tumor produces pro-metastasis factors, such as tumor--derived secreted factors (TDSF), to induce fibroblast activation in metastatic destination organs. The tumor-educated activated fibroblasts serve to construct a metastasis-supportive host stromal, including extra cellular matrix (ECM) deposition and reconstruction, angiogenesis and vascular leakage, as well as myeloid-derived suppressor cells (MDSC) recruitment. b, After FR17 subcutaneous administration, the in-situ assembled peptide nano-blanket in PMN stromal microenvironment impedes fibroblasts activation so as to retard stromal and vessel pro-metastatic reconstruction, inhibiting MDSC recruitment and metastatic cascades.
+Results and Discussion
+
+Peptide nano-blanket transformed from FR17.
+
+The matrix metalloproteinase 2 (MMP2)-activatable self-assembled branched peptide (FR17, FFK/GPLGLAGG-YVDKR)Y consists of (1) the backbone of a self-assembly peptide domain Phe-Phe-Lys-Tyr (FFKY), which is derived from β-amyloid (Aβ) peptide\(^{14, 15}\); (2) thymopentin (TP5, Arg-Lys-Asp-Val-Tyr, RKDVY), the pentapeptide with perfect hydrophilic property and immune modulation effect, which is applied as the adjunctive therapeutic agent on cancer treatment in clinic to prevent postoperative infection and to activate immune response\(^{16, 17}\). TP5 was conjugated to the side-chain of Lys (K) of the main chain FFKY with the MMP2-cleavable peptide linker (Gly-Pro-Leu-Gly-Leu-Ala-Gly-Gly, GPLGLAGG)\(^{18}\), increasing the hydrophilic property of the entire peptide molecule. Furthermore, the sequence of TP5 in peptide FR17 was replaced by the scrambled pentapeptide of TP5 without immunomodulatory bioactivity, *i.e.* DVYKR, to form the scrambled group (sFD17, FFK/GPLGLAGG-RKYVD)Y) as peptide assembly control.
+
+Peptides were synthesized *via* Fmoc solid-phase peptide synthesis technology and the peptide sequences were verified by mass spectra (Supplementary Fig. 1-2). When specifically cleaved by MMP2, both FR17 and sFD17 are able to release self-assembled monomer FG8 (FFK/GPLG)Y), constructing peptide self-assemblies, the peptide nano-blanket. The transmission electron microscopy (TEM) images (Fig. 2a) and the Cryo-TEM image (Supplementary Fig. 3a) showed the lamellar structure of the peptide self-assemblies formed by enzymatic degradation (Supplementary Fig. 3b) with an average diameter of ~500 nm (Fig. 2b). In addition, circular
+dichroism (CD) analysis revealed alterations in the secondary structure of peptide by the presence of enzyme (Supplementary Fig. 3c). The peptide nano-blanket is woven by the self-assembled monomer FG8 (FFK/GPLGY), which would spontaneously fold into lamellar structure rather than the fiber clusters aggregated by FFKY as we previously reported14. The peptide assembly relies on the non-bonded interactions, typically hydrogen bonds and \( \pi-\pi \) stacking, between adjacent peptide molecules. The self-assembling pattern of FG8 changed due to the branch modification with GPLG side-chain. The all-atom molecular dynamics (MD) simulation of FFKY-assemblies and FG8-assemblies gives a microscopic account of the impacts of branch modification on peptide interactions. The MD simulation revealed that there are more intermolecular hydrogen bonds involved in FG8 cluster, which are more ordered and mostly formed between Phe of adjacent FG8 molecules (Supplementary Fig. 4a-b). While the intermolecular hydrogen bonds involved in FFKY assembly are less-formed and disordered (Fig. 2c-h). Besides, the aggregation-induced luminescence effect (AIE) was also employed to monitor the spontaneous aggregation of FG8, the self-assembled monomer, in aqueous system (Supplementary Fig. 4c-e). Taken together, these results indicated the self-assembly property of FG8 monomer, and MMP2-cleaved release of FG8 from FR17 or sFD17 to form the peptide nano-blanket.
+Figure 2. Enzyme-activated self-assembly of FR17 and all-atom molecular dynamics (MD) simulation of the self-assembly of FG8. a, TEM images of FR17 or sFD17 treated with MMP2 (scale bar = 200 nm), FG8 and FFKY (scale bar = 100 nm). b, Size distribution of FR17 and sFD17 after MMP2 cleavage. c, Molecular structure of FFKY. d, The FFKY assemblies generated at t = 200 ns of MD simulation of 16 FFKY molecules in water (containing NaCl for charge neutralization). e, Typical molecular cluster and the non-bonded interactions involved in FFKY assemblies in detail. f, Molecular structure of FG8. g, The FG8 assemblies generated at t = 200 ns of MD simulation of 16 FG8 molecules in water. h, Typical molecular cluster and the non-bonded interactions involved in FG8 assemblies in detail. The dashed purple lines denote the \( \pi \)-\( \pi \) stacking. The dashed blue lines denote the hydrogen bonds. The nitrogen atoms are labeled in blue. And the oxygen atoms are labeled in red.
+The pathological process in the MCM-induced PMN model in vivo
+
+In order to study the effect of FR17 administration on the process of PMN development, an in vivo PMN model induced by melanoma-conditioned media (MCM) has been established according to the previous research on PMN\(^{19}\). Briefly, tumor-derived factors secreted from primary tumor were replaced by MCM intraperitoneal injection to the mice for 10 consecutive days from Day 1 to 10. On Day 7, when the tumor supportive microenvironment was successfully established in the lung, B16F10 melanoma cells were intravenously administrated as the simulation of circulative tumor cells (CTC) wandering through blood vessels and some would successfully colonize in the prepared “fire scene” in lung (Supplementary Fig. 5b).
+
+The pathological process in the lung was assessed from various aspects during MCM-induced PMN establishment from Day 3 to Day 13 (Supplementary Fig. 5-6). Important cellular derived molecular components and cytokines were closely monitored during the process. The higher expression of the extracellular matrix (ECM) component fibronectin (FN) was generated by MCM inducement. Moreover, matrix metalloproteinase 9 (MMP9), vascular endothelial growth factor a (VEGFa) were up-regulated along with the pathological progress of lung PMN from Day 0 to Day 10 and reached a plateau on Day 10 (Supplementary Fig. 5a). An accordant trend was found on MMP2 level as well (Supplementary Fig. 5d). Meanwhile, the immune cell population analysis during the pathological process in the lung of MCM-induced PMN model was conducted by flow cytometry, which drew our attention to the crime culprit-cell induced PMN, MDSC, for the recruitment of MDSC increased through the timeline (Supplementary Fig. 6-7). It’s reported that MDSC contributes a lot in developing the immunosuppression microenvironment in PMN, preparing more suitable and fertile land for tumor cells to take root in\(^{13,20}\). Expression level of the
+major inflammatory mediator TGF-β1, possibly produced by MDSCs, increased gradually but sharply. Typical biomarkers produced by MDSC to exert immuno-suppressive effect, in other words, the incriminating tools for the microenvironment re-education, were also detected, *i.e.* reactive oxygen species (ROS), iNOS and arginase-1 (Arg-1) (Supplementary Fig. 5a, c) \(^{21}\). Major characteristics of PMN also emerged in lung as time went by, such as activation of fibroblasts (Supplementary Fig. 5e-f), angiogenesis (Supplementary Fig. 5g) and increase of vascular permeability (Supplementary Fig. 5h) \(^{12}\). All these data indicated that the *in vivo* PMN model had been well-established, reflecting the pathological process of PMN development. With the overall support provided by PMN, it would consequently accelerate and aggravate metastasis *in vivo* (Supplementary Fig. 8).
+
+**FR17 administration interrupts the activation of fibroblast induced by tumor derived factors**
+
+According to previous researches on tumor metastasis\(^{22}\) and our assessment on MCM-induced PMN mice model, the activation of the resident fibroblasts in distal tissues could be regarded as the tipping point of the beginning of PMN establishing, raising the alarm about laying the foundations of the potential metastasis. It’s reported that tumor-educated fibroblasts would serve to construct a tumor supportive host stromal *via* TGF-β signaling\(^{23}\), promoting ECM degradation and reconstruction, inducing angiogenic and pro-inflammatory response of endothelial cells, recruiting VLA-4\(^+\) bone marrow-derived cells (BMDCs) by localized FN deposition for niche formation\(^{19}\). Indeed, lysyl oxidase (LOX) cross-linked collagen surrounding pre-metastasis associated fibroblasts attracts CD11b\(^+\) myeloid cells invasion in destination organs, which corresponds to metastatic efficiency\(^{24}\).
+Figure 3. FR17 interrupted the activation of fibroblast induced by tumor derived factors. a,
+Schematic drawing illustrates the in-situ assembled peptide nano-blanket interrupts the activation of fibroblast. When activated by tumor derived factors during PMN development, the expression of proangiogenic factors and ECM remodeling factors would be up-regulated in fibroblast, as well as the ECM components production. While the peptide nano-blanket could calm down fibroblast activation, down-regulating the above factors. b, Schematic illustration to show the protocol of MCM stimulation and peptide treatment on mice lung fibroblast (MLF) in vitro. c, Cell proliferation of MLF after MCM stimulation and peptide treatment (n = 4). d, Secretion of MMP9, VEGF and fibronectin in the culture media of MLF after stimulated by MCM, treated with or without peptide. n = 4 for treated groups and 3 for the control group. e, qPCR analysis of Acta2 (i.e., αSma), Mmp9, Vegfa, Fibronectin1 (Fn1) expression in MLF after MCM stimulation
+and peptide treatment. f, Migration assay and collagen gel contract assay to evaluate MLF cellular functions after MCM stimulation and peptide treatment (n = 3). Scale bar = 100 μm. g, Expression level of fibronectin in the lung harvested from the PMN model mice administrated with different peptides on Day 10. h, Representative images and semi-quantification of αSMA+ fibroblasts in the lung harvested from mice administrated with different peptides on Day 10. Data is presented as mean ± SD. One-way ANOVA followed by Tukey’s multiple comparisons test was employed for statistical evaluation. Scale bar = 50 μm.
+
+Here in a cell model simulating the impact of secreted factors derived from primary tumor on lung fibroblasts in vitro (Fig. 3b), FR17 cultivation along with MCM stimulation showed an interruption effect on tumor derived factors irritating lung fibroblasts (Fig. 3a). The treatment with FR17 or sFD17, which would form peptide nano-blanket assembled with the monomer FG8 that was released by MMP2-induced cleavage, successfully prevented the activation of fibroblasts by MCM. The inhibition on the expression of pathologic fibroblast biomarker alpha smooth muscle actin (αSMA) as well as the fibroblasts proliferation and cellular bio-functions (Fig. 3c-f) also substantiated the blocking of lung fibroblasts activity by FR17 or the scrambled control sFD17. The reverse of the increase in cytokines secretion, like MMP9, VEGF, FN, was determined by ELISA kit (Fig. 3d). The qPCR results further suggested that the peptide assemblies might exert biological regulatory effect on cells while not just acting like a physical shield to wrap on the surface of cells (Fig. 3e). Besides, when irritated by tumor derived factors to present the activated αSMA+ phenotype, the migration ability as well as the collagen gel
+contracting function of fibroblasts got promoted. By contrast, cultivation with FR17 minified these functional changes to a large extent (Fig. 3f). Administration of FR17 to the in vivo PMN model also gave great relief to the activation of the fibroblasts in lung (Fig. 3h). What’s more, the above data also suggested that the “flame-retarding” effect on fibroblast activation was almost completely contributed by peptide nano-blanket on account of the similar outcomes of the scrambled control sFD17 to that of FR17 treatment.
+
+The stromal microenvironment, apart from fibroblasts, consisting of ECM and vasculature, could also be re-educated by tumor derived factors\(^{25}\). For ECM would have gone through remodeling to rebuild a supportive niche for tumor colonization in PMN model, Western blot analysis, Masson staining and Sirius Red staining revealed the down-regulation in the expression of FN (Fig. 3g) and collagen (Supplementary Fig. 9a-c) in the lungs of FR17 group and sFD17 group, which would have likely been over-expressed by the activated fibroblasts otherwise. Besides, versican expression in the lung was also lower in the FR17 or sFD17 intervened groups than that of PMN control group and TP5 treated group (Supplementary Fig. 9d), and the elevated expression of which has been reported to contribute to angiogenesis in tumor\(^{26}\).
+
+FR17 protects fibroblasts from activation to inhibit vascular leakage and angiogenesis.
+
+The activated fibroblasts in primary tumor, also known as cancer-associated fibroblasts (CAFs), have been reported to produce proangiogenic factors, such as VEGF, so as to promote tumor angiogenesis\(^{27}\). What’s more, the secretion of MMPs by CAFs induced tumor vascular leakage, exacerbating MDSCs and tumor cells extravasation into the vascular system. Therefore, we
+assumed that, as an important participator and vanguard in the construction of PMN, the activated fibroblast may also contribute to angiogenesis and the increasing vascular permeability in PMN (Fig. 4a). Further exploration was carried out on mice vascular endothelial cells to find out whether fibroblasts promote angiogenesis and vascular permeability during the arousalment induced by MCM (as shown in Supplementary Fig. 10a). The tube forming assay demonstrated the proangiogenic capability of fibroblasts activated by MCM (Fig. 4b). Meanwhile, the transwell permeability assay (as illustrated in Supplementary Fig. 10b) to mimic the inner layer of blood vessel *in vitro* verified that the fibroblast-conditioned medium (FCM) collected from activated mice lung fibroblasts (MLF) increased vascular permeability (Fig. 4c), indicated directly by the dismissal of endothelial adherence junctions mediated by vascular endothelial cadherin, VE-cadherin (Fig. 4d).
+
+For all experiments conducted on endothelial cells, the different conditional FCM was obtained from fibroblasts pre-treated with FR17, sFD17 or TP5 on interrupting MCM stimulation separately. The conditional FCM pre-treated with FR17 or sFD17, rather than TP5, offset the increase in bEnd3 cell proliferation induced by the indirect stimulation of MCM (Supplementary Fig. 10c). The acceleration in neovascularization and the disruption on endothelial cell-cell connection to cause vascular leakage were made up by indirect FR17 or sFD17 treatment on MLF as indicated in Fig. 4b-c.
+
+To examine the influence of FR17 intervention on the expression level of the proangiogenic factors and vascular remodeling enzymes in pulmonary PMN in general, Western blot analysis was carried out. When compared to PMN control group, the increased expression of VEGFa,
+ANG2, MMP9, MMP2 was reversed by FR17 treatment almost back to normal levels (Fig. 4e, S11). In addition, the scrambled control sFD17, which doesn't possess the drug-bioactivity of TP5, exhibited similar inhibition result as FR17 while TP5 didn't, suggesting the protective effect was contributed by the peptide assemblies formed by the enzyme-activatable self-assemble monomer. Vascular leakage assay on PMN model in vivo verified that FR17 or sFD17 administration attenuated the enhancement of pulmonary vascular permeability (Fig. 4f-g).
+Figure 4. FR17 administration protected fibroblasts from activation to inhibit vascular leakage and angiogenesis. a, Illustration of the inhibition of vascular leakage and angiogenesis via the peptide nano-blanket protection on fibroblasts from tumor re-education. b, Tube forming assay was performed on bEnd3 cells which were treated with conditional FCM collected from MLF pre-stimulated with MCM and peptide. Scale bar = 100 μm. c, Permeability of the endothelial cell layer in vitro when co-cultured with MLF pre-treated with MCM and peptide. Data is presented as mean ± SD. n = 3. d, Integrality of the endothelial cell monolayer after cultivated with conditional FCM collected from MCM and peptide pre-stimulated MLF, indicated by VE-cadherin on the membrane (n = 3). The white box and the white arrows in the enlarged images indicate the tight junction between the endothelial cells. While the yellow box and yellow arrows indicate the disruption of cell-cell connection. Scale bar = 50 μm in the upper panel. Scale bar = 30 μm in the lower panel. e, Expression level of VEGFa, ANG2, MMP9 and
+MMP2 in the pulmonary PMN harvested from mice administrated with different peptides on Day 10. f & g, Vascular permeability of the pulmonary PMN after peptide administration. The semi-quantification was calculated from 5 random visual fields for each section taken from n = 3 biologically independent mice. Data is presented as mean ± SD. Scale bar = 100 μm. One-way ANOVA followed by Tukey’s multiple comparisons test was employed for statistical evaluation.
+
+Above data confirmed that FR17 can be employed as an *in-situ* spontaneously-assembled “flame-retarding blanket” to beat out the “flames” on fibroblasts caused by tumor derived factors in PMN, preventing pro-metastatic angiogenesis and vascular destabilization.
+
+**FR17 administration impedes MDSC recruitment to pulmonary PMN and modulates the immuno-microenvironment.**
+
+Given the fact that MDSC occupies a vital position in PMN construction and metastasis formation, MDSC has attracted wide concerns in recent years. It’s firstly reported by David Lyden’s team that VEGFR+ myeloid progenitor cells are recruited and formed cell clusters to the sites highly expressing FN, which is most likely produced by resident fibroblasts, in the distal tissues prior to the arrival of tumor cells\(^{19}\). The recruitment of MDSC was found to be related to the enhancement of ECM production and remodeling in PMN, including cross-linking collagen by LOX secreted by tumor cells\(^{24}\), FN deposition\(^{28}\), periostin (POSTN)-enrichment\(^{29}\). As MDSC plays a crucial role in developing immunosuppressive and inflammatory microenvironment,
+some pioneers have demonstrated that blocking the recruitment of MDSC could be a promising strategy to suppress early metastasis by preventing the development of breeding ground for tumor^{30-33}.
+
+In our PMN model, the accumulation of MDSC in lung was decreased in FR17 and sFD17 groups, revealed by immunofluorescent staining (Fig. 5a) and flow cytometry analysis on Day 10 (Fig. 5b). Moreover, focal enrichment areas of FN and the co-localization of LOX attracted an increasing number of CD11b^+Gr^+ MDSC in serial sections of the PMN lung comparing to healthy lung (Fig. 5a, Supplementary Fig. 12), consistent with the previous reports^{24}. The immunofluorescent stains indicated that FR17 intervention successfully down-regulated the expression of LOX, FN and POSTN in the lung induced by tumor-derived factors to impede the construction of PMN, which probably resulted in the cut in MDSC recruitment (Supplementary Fig. 12). To evaluate prevention of peptide administration on developing PMN immunosuppressive environment, protein level of TGF-\(\beta\) in PMN (Fig. 5c), the well-known immuno-modulator produced by MDSC to regulate the establishment of immunosuppressive tumor supportive niche, as well as the pro-inflammatory cytokine IL-6 was detected (Fig. 5d). FR17 treatment successfully down-regulated TGF-\(\beta\) and IL-6 expression in lung, so as sFD17. In the meantime, TP5 administration exhibited partial abatement on the expression of these cytokines^{34}, suggesting both the enzyme-responsive assembled peptide nano-blanket and the immunoregulatory agent TP5 facilitated the normalization of immunosuppressive and inflammatory environment of PMN.
+Figure 5. FR17 administration prevented MDSC recruitment, pausing the development of the immunosuppressive microenvironment in PMN. a, Serial sections of the lungs harvested from the model mice treated with different peptides. Serial sections show the distribution of lysyl oxidase (LOX) and Fibronectin (FN) and the co-location of CD11b+Gr1+ myeloid derived suppressor cells (MDSC). b, Recruitment of CD11b+Ly6g+ MDSC to the lungs of the model mice treated with different peptides on Day 10. Data is presented as mean ± SD. n = 4 biologically independent mice for treatment groups, n = 3 for the control group. c, Expression of TGF-β1 in the lung harvested from PMN model mice administrated with different peptides. d,
+IL-6 expression in the lung tissue fluid collected from PMN model mice administrated with different peptides. Data is presented as mean ± SD. n = 3 biologically independent mice.
+
+One-way ANOVA followed by Tukey’s multiple comparisons test was employed for statistical evaluation. e, GO enrichment analysis of CD11b+Ly6g+ MDSC sorted from different treatment groups. RNA preparations were extracted from CD11b+Ly6g+ MDSCs sorted from lungs pooled from 10-12 mice per sample. The size of the dots corresponds to the number of genes per pathway, and the color indicates \( p \)-value. Statistical significance was considered at least at \( p < 0.05 \).
+
+To find out whether the impact on MDSC-induced immunosuppressive microenvironment in PMN was mainly contributed by TP5 or by the *in-situ* assembled peptide nano-blanket, cell transcriptomic analysis of MDSCs recruited to the lung was carried out on different treatment groups. Given the fact that CD11b+Ly6g+ MDSC takes the majority (over 90%) of MDSC population in the lung of MCM-induced PMN model (Supplementary Fig. 6a-b), CD11b+Ly6g+ MDSC were sorted on Day 10 from pulmonary PMN after different treatments as the representative subset for further transcription analysis (Supplementary Fig. 13). GO enrichment analysis indicated that the regulation effect of FR17 might relate to the activation of immune response pathway, cytokine production involved in immune response, regulation of leukocyte and lymphocyte activation, leukocyte chemotaxis and migration (Fig. 5e). Relative enriched pathways provided possible comments on the underlined mechanisms, including: regulation on cytokine biosynthetic process, adaptive immune response based on somatic recombination
+immune receptors built from immunoglobulin super-family domains, interferon-γ-mediated signaling pathway and the impact on chemokine receptor binding as well as CXCR chemokine receptor binding (Supplementary Fig. 14). From above, when compared with the differential gene enrichment pathway of sFD17 and TP5, we found that the regulation on immune cells chemotaxis and migration pathways was contributed by the in-situ peptide assemblies, for the peptide nano-blanket would only form in FR17 or sFD17 treated lung while not in free TP5 treated group. And the Venn diagram and further enrichment analysis to put TP5 aside suggested these would correspond to the regulation on cell surface receptor signaling pathway by peptide assemblies (Supplementary Fig. 14b-c). In addition, the regulation on leukocyte differentiation and T cell activation pathway was mainly contributed by TP5 (Supplementary Fig. 14d), which has been commonly accepted as one of the mechanisms for TP5 to exert immune regulation effect35-37.
+
+Moreover, the in-situ assembled peptide nano-blanket formed by FR17 inhibited tumor cells migration as well, with hardly any effect on cell viability (Supplementary Fig. 15).
+
+FR17 administration inhibits melanoma lung metastasis in vivo.
+
+First, we evaluated the anti-metastasis efficacy of FR17 on MCM-induced lung metastasis model. C57/BL6 mice were pre-induced by MCM administration for 10 days to set up the tumor supportive pre-metastatic niche in lung. Tumor cells were then injected through the tail vein on Day 7 to simulate the circulative tumor cells (CTC) wandering through blood vessels and eventually settled down in pulmonary PMN. As illustrated in Fig. 6a, peptide subcutaneous
+administration started from Day 3 and ended at 2 weeks-post lung tumor inoculation. The metastasis inhibition efficacy was evaluated in terms of the tumor module number on Day 28.
+
+FR17 effectively suppressed tumorigenesis and the development of metastasis in lung, compared to both the control and TP5 groups (Fig. 6b-d). Moreover, the good outcomes of sFD17 treatment which was close to that of FR17 group emphasized the main role of nano-blanket assembled by the enzyme-activated released monomer in intervening PMN construction. Therefore, the *in-situ* assembled nano-blanket also played a part in inhibiting metastasis development and growth.
+
+Preliminary safety evaluation on these model mice at the end of the experiment revealed no severe safety concern of FR17 administration (Supplementary Fig. 16-17). The systemic immune modulation effect of TP5 was reflected on the recovery of thymus shrinking as well (Supplementary Fig. 16d, 17). What’s more, peptide administration prevented tumor cells infiltration into the spleen (Supplementary Fig. 17).
+
+The anti-metastasis activity of FR17 was further verified on a post-surgery metastasis model\(^{30}\) (illustrated by Fig. 6e), which is closer to the clinical treatment of resectable melanoma followed-up with adjuvant therapy for patients at high risk of recurrence. Here we compared FR17 therapy with the rising-star of checkpoint inhibitors, PD1 antibody\(^{38-39}\), which was approved by FDA as adjuvant therapy on primary tumor excised with lymph node management or metastatic melanoma\(^{40}\). As demonstrated in Fig. 6f, the recurrence of lung metastasis was put off and the overall survival was greatly improved by FR17 therapy when compared with control (Fig. 6g-i). The median overall survival prolonged for 23.3% (from 30 days to 37 days), catching up with the outcomes of anti-PD1 treatment (from 30 days to 36 days). And the median
+lung-metastasis-free survival prolonged from less than 19 days of both control group and anti-PD1 treated group to 23 days of FR17 treated group. At the first time-point for lung metastasis monitoring via bioluminescence imaging, lung metastasis has been observed in all of the 10 mice from the control group, 8 mice from anti-PD1 treated group while only 3 mice from FR17 treated group on Day 19. The retard of the occurrence of lung metastasis after FR17 administration emphasized the importance of PMN intervention. What’s more, the relapse rate of the primary tumor post-resection was reduced from 10 / 10 in the control group to 4 / 10 in FR17 treated group (Supplementary Fig. 18).
+Figure 6. FR17 administration inhibited lung metastasis in vivo by retarding PMN formation. a, Schedule of MCM-induced lung metastasis model with peptide treatment. Peptide administration started from Day 3 for metastasis prevention. b, Number of the metastatic nodules in the lung of different treatment groups (n = 7). Data is presented as mean ± SD. One-way ANOVA followed by Tukey’s multiple comparisons test was employed for statistical evaluation.
+
+c, Images of the lung metastasis harvested on Day 28 from different treatment groups. d, Representative images of Hematoxylin & Eosin staining of lung sections from different treatment groups. The lung metastasis is circled by yellow dotted lines. Scale bar = 200 μm. e, Schedule of post-surgery metastasis model. Peptide administration started from Day 7 to 25. f, Representative in vivo bioluminescent images of mice without treatment or treated with FR17 or anti-PD1 (n = 10). The grey patches represent dead mice in the control group. g, Survival curves of the mice treated with FR17 or anti-PD1 or without treatment (n = 10). h, The cumulative
+incidence of new pulmonary metastases in the mice treated with FR17 or anti-PD1 or without treatment (n = 10). i, Semi-quantification of the *in vivo* bioluminescent signals in the lungs of the mice treated with FR17 or anti-PD1 or without treatment (n = 5). Data is presented as mean ± SD.
+
+**Conclusions**
+
+In summary, we have explored the magical retarding effect of the PMN microenvironment responsive-assembled peptide nano-blanket on fibroblasts activation, impeding PMN development. The enzyme-activatable assembled peptide FR17 can be enzyme-cleaved to release the self-assembly monomer FG8 to construct a lamellar structure, which is named as peptide nano-blanket. Experiments demonstrated that FR17 administration not only beat out the “flame” set up on resident fibroblasts induced by tumor derived factors, but also interrupted the subsequent PMN formation, including preventing pro-metastatic angiogenesis and vascular destabilization, and then intervening MDSCs’ recruitment as well as their bio-functions. Astonishingly, when treated with sFD17, tumor-induced fibroblast activation was also impeded by peptide-assemblies without the assistance of TP5 so as to arrest the following pro-metastatic pathological process. This finding illustrated that the “flame-retarding” effect on fibroblast activation could be contributed by the drug-free peptide nano-blanket alone, presenting a broad application prospect of drug-free peptide assemblies to regulate PMN microenvironment and to prevent tumor distant metastasis. This work also elucidated the important role of pre-metastasis associated fibroblast in the complex interactions between major participators in PMN formation
+and metastasis development, suggesting that reprogramming or intervention on the key juncture could make a big difference on fighting against tumor metastasis.
+
+References
+
+1. Gdowski, A. S., Ranjan, A. & Vishwanatha, J. K. Current concepts in bone metastasis, contemporary therapeutic strategies and ongoing clinical trials. J. Exp. Clin. Cancer Res. **36**, 108-121 (2017).
+
+2. Pashayan, N. & Paul, P. D. P. The challenge of early detection in cancer. Science (*New York, N.Y.*) **368**, 589-590 (2020).
+
+3. Rodriguez-Ruiz, A. et al. Stand-alone artificial intelligence for breast cancer detection in mammography: Comparison with 101 radiologists. *J. Natl. Cancer Inst.* **111**, 916-922 (2019).
+
+4. Poudineh, M., Sargent, E. H., Pantel, K., Kelley, S. O. Profiling circulating tumour cells and other biomarkers of invasive cancers. *Nat. Biomed. Eng.* **2**, 72-84 (2018).
+
+5. Christensen, E. et al. Early detection of metastatic relapse and monitoring of therapeutic efficacy by ultra-deep sequencing of plasma cell-free DNA in patients with urothelial bladder carcinoma. *J. Clin. Oncol.* **37**, 1547-1557 (2019).
+
+6. Abbosh, C. et al. Phylogenetic ctDNA analysis depicts early-stage lung cancer evolution. *Nature* **545**, 446-451 (2017).
+
+7. Curti, B. D. & Faries, M. B. Recent advances in the treatment of melanoma. *N. Engl. J. Med.* **384**, 2229-2240 (2021).
+8. Oliver Sartor, M. D. & Johann S. de Bono, M. B. Metastatic prostate cancer. N. Engl. J. Med. **378**, 645-657 (2018).
+
+9. Zhang, C. et al. Radiotherapy and cytokine storm: Risk and mechanism. Front. Onco. **11**, 670464 (2021).
+
+10. Cardoso, F. et al. 70-Gene signature as an aid to treatment decisions in early-stage breast cancer. N. Engl. J. Med. **375**, 717-29 (2016).
+
+11. Peinado, H. et al. Pre-metastatic niches: organ-specific homes for metastases. Nat. Rev. Cancer **17**, 302-317 (2017).
+
+12. Liu, Y. & Cao, X. Characteristics and significance of the pre-metastatic niche. Cancer Cell **30**, 668-681 (2016).
+
+13. Zhou, Y., Han, M., & Gao, J. Q. Prognosis and targeting of pre-metastatic niche. J. Control. Release **325**, 223-234 (2020).
+
+14. Wang, T. T. et al. AIE/FRET-based versatile PEG-Pep-TPE/DOX nanoparticles for cancer therapy and real-time drug release monitoring. Biomater. Sci. **8**, 118-124 (2020).
+
+15. Guo, W. W. et al. Intracellular restructured reduced glutathione-responsive peptide nanofibers for synergetic tumor chemotherapy. Biomacromolecules **21**, 444-453 (2020).
+
+16. Zeng, F. L. et al. Clinical efficacy and safety of synthetic thymic peptides with chemotherapy for non-small cell lung cancer in China: A systematic review and meta-analysis of 27 randomized controlled trials following the PRISMA guidelines. Int. Immunopharmacol. **75**, 105747 (2019).
+
+17. Li, S. et al. Supramolecular nanofibrils formed by coassembly of clinically approved drugs for tumor photothermal immunotherapy. Adv. Mater. **33**, e2100595 (2021).
+18. Qin, S. Y. et al. Theranostic GO-based nanohybrid for tumor induced imaging and potential combinational tumor therapy. Small **10**, 599-608 (2014).
+
+19. Kaplan, R. N. et al. VEGFR1-positive haematopoietic bone marrow progenitors initiate the pre-metastatic niche. *Nature* **438**, 820-827 (2005).
+
+20. Engblom, C., Pfirschke C. & Pittet M. J. The role of myeloid cells in cancer therapies. *Nat. Rev. Cancer* **16**, 447-462 (2016).
+
+21. Gabrilovich, D. I. Myeloid-derived suppressor cells. *Cancer Immunol. Res.* **5**, 3-8 (2017).
+
+22. Pein, M. et al. Metastasis-initiating cells induce and exploit a fibroblast niche to fuel malignant colonization of the lungs. *Nat. Commun.* **11**, 1494 (2020).
+
+23. Kong, J. et al. Extracellular vesicles of carcinoma-associated fibroblasts creates a pre-metastatic niche in the lung through activating fibroblasts. *Molecular Cancer* **18**, 175 (2019).
+
+24. Erler, J. T. et al. Hypoxia-induced lysyl oxidase is a critical mediator of bone marrow cell recruitment to form the premetastatic niche. *Cancer Cell* **15**, 35-44 (2009).
+
+25. Paolillo, M. & Schinelli, S. Extracellular matrix alterations in metastatic processes. *Int. J. Mol. Sci.* **20**, 4947 (2019).
+
+26. Asano, K. et al. Stromal versican regulates tumor growth by promoting angiogenesis. *Sci. Rep.* **7**, 17225 (2017).
+
+27. Zhou, X. et al. Melanoma cell-secreted exosomal miR-155-5p induce proangiogenic switch of cancer-associated fibroblasts *via* SOCS1/JAK2/STAT3 signaling pathway. *J. Exp. Clin. Cancer Res.* **37**, 242 (2018).
+
+28. Murgai, M. et al. KLF4-dependent perivascular cell plasticity mediates pre-metastatic niche
+formation and metastasis. Nat. Med. **23**, 1176-1190 (2017).
+
+29. Wang, Z. et al. Periostin promotes immunosuppressive premetastatic niche formation to facilitate breast tumour metastasis. *J. Pathol.* **239**, 484-495 (2016).
+
+30. Long, Y. et al. Self-delivery micellar nanoparticles prevent premetastatic niche formation by interfering with the early recruitment and vascular destruction of granulocytic myeloid-derived suppressor cells. *Nano Lett.* **20**, 2219-2229 (2020).
+
+31. Jiang, T. et al. Metformin and Docosahexaenoic Acid Hybrid Micelles for Premetastatic Niche Modulation and Tumor Metastasis Suppression. *Nano lett.* **19**, 3548-3562 (2019).
+
+32. Lu, Z et al. Epigenetic therapy inhibits metastases by disrupting premetastatic niches. *Nature* **579**, 284-290 (2020).
+
+33. Kaczanowska, S. et al. Genetically engineered myeloid cells rebalance the core immune suppression program in metastasis. *Cell* **184**, 2033-2052 (2021).
+
+34. Lunin, S. M. et al. Thymic peptides restrain the inflammatory response in mice with experimental autoimmune encephalomyelitis. *Immunobiology* **218**, 402-7 (2013).
+
+35. Cascinelli, N. et al. Evaluation of clinical efficacy and tolerability of intravenous high dose thymopentin in advanced melanoma patients. *Melanoma Res.* **8**, 83-9 (1998).
+
+36. Wang, Y. et al. The novel role of thymopentin in induction of maturation of bone marrow dendritic cells (BMDCs). *Int. Immunopharmacol.* **21**, 255-60 (2014).
+
+37. Lunin, S. M. et al. Thymus peptides regulate activity of RAW 264.7 macrophage cells: inhibitory analysis and a role of signal cascades. *Expert Opin. Ther. Targets* **15**, 1337-46 (2011).
+
+38. Rizvi, N. A. et al. Activity and safety of nivolumab, an anti-PD-1 immune checkpoint
+inhibitor, for patients with advanced, refractory squamous non-small-cell lung cancer (CheckMate 063): a phase 2, single-arm trial. *Lancet Oncol.* **16**, 257-65 (2015).
+
+39. Gong, N. et al. Proton-driven transformable nanovaccine for cancer immunotherapy. *Nat. Nanotechnol.* **15**, 1053-1064 (2020).
+
+40. NIH. National Cancer Institute, Melanoma Treatment (PDQ#)-Health Professional Version. Available from: https://www.cancer.gov/types/skin/hp/melanoma-treatment-pdq#_402.
+Acknowledgements
+
+This work was supported by the National Natural Science Foundation of China (Nos. 81673022).
+
+We thank Professor Yu Kang at College of Pharmaceutical Sciences, Zhejiang University for guidance on molecular dynamics simulation. We thank Qichun Wei’s lab for providing the luciferase transfected B16F10. We thank Qin Han, Chenyu Yang, and Dandan Song at the Center of Cryo-Electron Microscopy (CCEM), Zhejiang University for their technical assistance on Confocal laser scanning microscopy and transmission electron microscopy. We thank Yanwei Li at the Core Facilities of Zhejiang University School of Medicine for technical assistance in flow cytometry analysis.
+
+Author Contributions
+
+M.H. and Y.Z. conceived the project and designed the experiments. M.H. and J.G. supervised the project, discussed and commented on the manuscript. Y.Z. performed the majority of the experiments and data analysis. P.K. assisted with cell culture of B16F10, preparing MCM and animal experiments. Y.X. synthesized and characterized TPE-FR17 and TPE-FFKY. H.W. assisted with the establishment of animal models. Y.Z. performed the flow cytometry experiments with assistance from Z.Z., T.W., M.L., P.K., Y.S.L., Y.X., H.Z., X.Z. and Q.Y.; Y.Y.L. supported the operation of confocal microscope. Y.Z., Q.D., P.K., H.Z., Y.X., Y.S.L., X.J. and H.W. operated the tumor resection surgery. Y.Z. wrote the manuscript. All authors discussed the results and commented on the manuscript.
+Competing Interests statement
+
+All the authors declare no conflicting interests.
+Methods
+
+Characterization of FR17 and sFD17. Peptide FR17 and sFD17 were synthesized via Fmoc solid-phase peptide synthesis technology by APeptide Shanghai. The molecule structure and amino acid sequence of peptide were confirmed by mass spectra.
+
+Enzyme induced assembly of FR17 and sFD17. Peptide FR17 or sFD17 was dissolved in HBS buffer (pH 7.4, containing 50 mM HEPES, 150 mM NaCl, 10 mM CaCl2) at 500 μM. Activated hMMP2 was added to the peptide solution at the working concentration of 1 μg/mL. The solutions were then incubated in 37 °C air-bath for 24 h. The size of the enzyme treated sample, which has formed the peptide nano-blanket, was measured by Zetasizer Nano ZS (Malvern Instruments). The assembly morphology of the assembled nano-blanket was observed with transmission electron microscope (FEI Tecnai G2 spirit) for TEM image and 200kv transmission electron microscope (FEI Talos F200C) for Cryo-TEM image.
+
+All-atom molecular dynamics simulation. The molecular structures of peptide FFKY and FG8 monomer were first constructed via AMBERTOOL based on the AMBER14SB force field. Dynamics simulation runs were performed utilizing Gromacs 2018.4 package1. System configurations were visualized using VMD software2, and generated into images mainly employing GRACE software. The simulation was performed in water boxes containing 16 FFKY or FG8 molecules. FFKY system was simulated containing NaCl to neutralize electric charge of
+the amidogen on the side-chain if Lys. Energy was minimized according to the steepest-descent method. Bond lengths were constrained by the LINCS algorithms. The nonbonded LJ interactions were cut off at 1.2 nm. Electrostatics was treated utilizing the Particle Mesh Ewald (PME) scheme. All production runs were simulated in the NPT ensemble using V-rescale coupling scheme with the temperature maintained at 298.15 K and parrinello-rahman coupling scheme with pressure kept at 1.0 bar and isotropic coupling type. The time constants for the pressure and temperature couplings were respectively set to 2.0 and 0.2 ps. Besides, the compressibility value was \(4.5 \times 10^{-5}\) bar\(^{-1}\). Periodic boundary conditions with a time step of 0.002 ps were adopted. Simulations were carried out for 200 ns and the structural coordinate information was recorded per 50 ps.
+
+Cell lines and cell culture. B16F10 cells were purchased from Cell Bank of Chinese Academy of sciences (Shanghai, China) and cultured in Dulbecco's Modified Eagle Medium (DMEM, Cienry, China) containing 10% (v/v) fetal bovine serum (FBS, Gibco, Grand Island, USA) and penicillin-streptomycin Solution (100×, TBD, Tianjin, China) at 1% (v/v). And the luciferase transfected B16F10 (Luc-B16F10) was kindly gifted by Qichun Wei’s lab, the Second Affiliated Hospital, Zhejiang University. The mouse lung fibroblasts (MLF) were purchased from iCell Bioscience Inc. (Shanghai, China) and cultured in Dulbecco's Modified Eagle Medium/Nutrient Mixture F-12, 1:1 mixture (DMEM/F12 medium, Multicell, Wisent Int., Canada) containing 10% (v/v) fetal bovine serum (FBS, Gibco, Grand Island, USA) and penicillin-streptomycin Solution (100×, TBD, Tianjin, China) at 1% (v/v). The mouse endothelial cells bEnd3 was kindly gifted
+by Fuqiang Hu’s lab, Institution of Pharmaceutics, College of Pharmaceutical Sciences, Zhejiang University. The bEnd3 cells were cultured in the same medium as B16F10. Cells were cultured in a cell incubator containing 5% CO_2 at 37 °C and passaged when reached 80%-90% confluence.
+
+The melanoma-conditioned medium (MCM) was obtained as follows: when B16F10 reached 70-80% confluence, washed with PBS and changed the medium to serum-free medium and incubated for 24 h. The cell supernatants were collected, centrifuged at 2000 rpm for 10 min to discard the cell debris. The MLF-conditioned medium (FCM) was obtained as follows: MLFs were stimulated by MCM (supplemented with 10% (v/v) FBS) with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 μM) for 48 h, then replaced with fresh complete medium and incubate for another 24 h. The cell supernatants were collected, centrifuged at 2000 rpm for 10 min to discard the cell debris.
+
+Cell proliferation assays. MLFs in rapid proliferation were plated in 96-well plates at the density of \( 4 \times 10^3 \) per well and cultured overnight. Former media were removed and the cells were cultivated with 100 μL MCM supplemented with 10% (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 μM). Cells cultivated with fresh complete medium was set up as control. After incubation for 48 h, cell proliferation was measured using a Cell Counting Kit-8 assay (Cat. 13E02A60, Boster Biotech., China) according to the manufacturer’s instructions. The optical density at 450 nm (OD_{450} nm) was measured using a multiwell plate reader (ELX800, BioTek, USA). Each group was repeated at least 4 times, and cell proliferation was presented as mean ± SD.
+Cytokines secretion and gene expression of MLFs. An equal number of MLFs in rapid proliferation were seeded in a 24-well plate per well, cultivated with MCM supplemented with 10% (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 μM) for 48 h in at least triplicate. Cells cultivated with fresh complete medium was set up as control. The cell supernatants were collected, centrifuged at 2000 rpm for 5 min to discard cell debris. MMP9 ELISA kit (EK0466, Boster Biotech., China) and VEGF ELISA kit (EK0541, Boster Biotech., China) were employed to measure the secretion level of MMP9 and VEGF in the cell supernatant. An equal number of MLFs in rapid proliferation were seeded in 24-well plate cultivated with (serum-free) MCM, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 μM) for 48 h in quadruplicate. Cells cultivated with fresh DMEM/F12 medium was set up as control. The cell supernatants were collected, centrifuged at 2000 rpm for 5 min to discard cell debris. The secretion of FN was measured by FN ELISA kit (EK0351, Boster Biotech., China) according to the manufacturer’s instructions. Cells with different treatments were harvested for RT-qPCR analysis to determine the gene expression level of *M-Acta2* (Mouse αSMA), *M-Mmp9*, *M-Vegfa* and *M-Fn1*. The primer sequences are provided as follows:
+
+<table>
+  <tr>
+    <th>Primer Name</th>
+    <th>Sequence</th>
+    <th>Length (bp)</th>
+    <th></th>
+  </tr>
+  <tr>
+    <td>*M-Gapdh*-F:</td>
+    <td>GGTTGTCTCCTGCGACTTCA</td>
+    <td>58.2</td>
+    <td></td>
+  </tr>
+  <tr>
+    <td>*M-Gapdh*-R:</td>
+    <td>TGGTCCAGGGTTTCTTACTCC</td>
+    <td>58.2</td>
+    <td>183bp</td>
+  </tr>
+  <tr>
+    <td>*M-Vegfa*-F:</td>
+    <td>GCTACTGCCGTCCGATTGAG</td>
+    <td>60.87</td>
+    <td></td>
+  </tr>
+  <tr>
+    <td>*M-Vegfa*-R:</td>
+    <td>ACTCCAGGGCTTCATCGTTACAG</td>
+    <td>62.25</td>
+    <td>132bp</td>
+  </tr>
+  <tr>
+    <td>*M-Mmp9*-F:</td>
+    <td>CACAGCCAATATGACCAGGAT</td>
+    <td>59.1</td>
+    <td></td>
+  </tr>
+</table>
+<table>
+  <tr>
+    <th>Primers</th>
+    <th>Sequence</th>
+    <th>Position</th>
+    <th>Length (bp)</th>
+  </tr>
+  <tr>
+    <td><i>M-Mmp9-R:</i> 5' CAGGAAGACGAAGGGGAAGA 3'</td>
+    <td></td>
+    <td>59.1</td>
+    <td>115bp</td>
+  </tr>
+  <tr>
+    <td><i>M-Fn1-F:</i> 5' CTATTTACCAACCGCAGACTCAC 3'</td>
+    <td></td>
+    <td>58.7</td>
+    <td></td>
+  </tr>
+  <tr>
+    <td><i>M-Fn1-R:</i> 5' TGCTTGTTTCCTTGCAGACTT 3'</td>
+    <td></td>
+    <td>58.5</td>
+    <td>115bp</td>
+  </tr>
+  <tr>
+    <td><i>M-Acta2-F:</i> 5' CAACTGGTATTGTGCTGGACTC 3'</td>
+    <td></td>
+    <td>57.3</td>
+    <td></td>
+  </tr>
+  <tr>
+    <td><i>M-Acta2-R:</i> 5' ATCTCACGCTCGGCAGTAGT 3'</td>
+    <td></td>
+    <td>57.3</td>
+    <td>181bp</td>
+  </tr>
+</table>
+
+Cell Migration assays. MLFs in rapid proliferation were plated in Culture-Inserts (2 Well, Ibidi, Germany) at the density of \( 1 \times 10^5 \) in 70 \( \mu \)L per well and grew to confluence overnight. Culture-Inserts as well as the former medium were gently removed. Then the MLFs were cultivated with MCM supplemented with 10% (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 \( \mu \)M) in triplicate. Cells cultivated with fresh complete medium was set up as control. Pictures of the cell scratches were taken under the microscope at 0 h. After incubated for 24 h, MLFs were fixed and stained with crystal violet. Pictures were taken and analyzed with ImageJ.
+
+Collagen gel contraction assay. An equal number of MLFs seeded in 6 cm dishes were cultivated with MCM supplemented with 10% (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 \( \mu \)M) for 48 h. Cells cultivated with fresh complete medium was set up as control. The pre-treated MLFs were digested and re-suspended at the density of \( 2 \times 10^6 \) per mL, kept on ice for later use. The neutral collagen solution was prepared as follows: 224 \( \mu \)L type I collagen gel (3 mg/ml, Cat. C8062, Solarbio, China) was
+quickly mixed with 100 μL conditional medium and 8 μL NaOH (0.1 N) on ice. Then 300 μL of the pre-treated MLFs suspension was added to the collagen solution on ice immediately. And the neutral cell-collagen mixture was added to 48-well plates 200 μL per well in triplicate and allowed to solidify for 45 min at room temperature. After incubated at 37 °C for 12 h, the gels were photographed. ImageJ software was used to measure gel area and evaluate contraction. Gel contraction was assessed as the ratio of the gel area to the area of the well.
+
+Tube forming assay. Mouse endothelial bEnd3 cells seeded in 6 cm dishes were cultivated with conditional FCM (from MLF previously stimulated by MCM with or without TP5, sFD17 or FR17 peptide as illustrated above) for 24 h. The cells were harvested and seeded in 48-well plates pre-coated with Matrigel (Cat. 356230, BD, USA) in triplicate for each group. After 6 h incubation, five visual fields were randomly chosen from each well and photographed by microscope (CKX53, OLYMPUS, Japan). The tube forming results were analyzed by ImageJ.
+
+Transwell permeability assay. An equal number of MLFs seeded in 6 cm dishes were cultivated with MCM supplemented with 10% (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 μM) for 48 h. Cells cultivated with fresh complete medium was set up as control. The pre-treated MLFs were digested and seeded at the lower well of a 24-well transwell plate at 2.5×10^5 cells per well. Each group was triplicate. Then, 7,000 Mouse endothelial bEnd3 cells were seeded on a 0.4 μm Transwell insert (Cat. 3413, Corning Costar, USA) above the top of the well until grown to confluence. Rhodamine B-dextran (70 kDa,
+Cat. R9379, Sigma-Aldrich, USA) was added to the upper insert on the endothelial cell layer.
+
+After 1 h incubation, the translocation of Rhodamine B-dextran from the insert to the lower well passing through the endothelial cell layer was measured by a microplate reader (SPARK, TECAN, Switzerland) at an excitation/emission wavelength of 540 / 625 nm. The relative permeability of the cell layer was normalized by diving the fluorescence signals of the treatment groups by the control group.
+
+Integrity of the endothelial cell monolayer. Mouse endothelial bEnd3 cells grown in 35-mm confocal dishes to 100% confluence were treated with conditional FCM, which was obtained from MLFs stimulated by MCM with or without the treatment of different peptides (TP5, sFD17, FR17 at the concentration of 100 \( \mu \)M) as illustrated above. After incubated with conditional FCM for 24 h, the single endothelial cell layer was gently washed with PBS and fixed with 4% paraformaldehyde for 15min, permeabilized with 0.2% Triton X-100 for 15min and blocked with 2% bovine serum albumin (BSA) for another 15 min. Cells were incubated with anti-VE-cadherin antibody (1:1000, Cat. Ab205336, Abcam, UK) containing 0.2% BSA and 0.1% Triton X-100 in PBS at 4 °C overnight. Cells were washed with PBS for three times and incubated for 1 h with AF647 labeled goat anti-rabbit IgG (H+L) (1:100, Cat. 33113ES60, Yeasen, China). The nuclei were labeled by DAPI solution (ready-to-use) (Solarbio, China). Images were taken by confocal microscope (Leica, German).
+
+Mice and animal models. C57BL/6 mice (male, 5-week-old) purchased from Slaccas (Shanghai,
+China) were adaptive fed for more than one week for subsequent experiments. The animals were maintained under standard laboratory housing conditions where foods and water can be reached freely. All the animal experiments were conducted following the guidelines which have been approved by the Ethics Committee of Zhejiang University.
+
+For MCM-induced lung metastasis model, MCM (300 \( \mu \)L per mice) was intraperitoneally injected to the mice for 10 consecutive days from Day 1 to 10. On Day 7, a tail vein injection of B16F10 or Luc-B16F10 cells (\( 1 \times 10^5 \) per mice) was given to the mice. Lung metastasis was monitored twice a week by bioluminescence imaging (IVIS® Spectrum In Vivo Imaging System, PerkinElmer, USA) if viable. For bioluminescence imaging, mice were intraperitoneally injected with D-luciferin potassium salt (150 mg/kg, Gold Biotechnology, USA). The bioluminescence of pulmonary metastases was detected 10 min later.
+
+For post-surgery metastasis model, \( 1 \times 10^6 \) B16F10 cells were subcutaneously inoculated in the back of the C57BL/6 mice (male) aged 6 weeks above the right hindlimb on Day 1. On Day 12, tumor resection surgery was conducted on mice to remove the entire tumor tissues as well as the skin cover the tumor under anesthesia. On the next day, \( 1 \times 10^5 \) luciferase-expressing B16F10 cells were injected into mice through tail vein. The tumor recurrence and lung metastasis were monitored twice a week. The tumor volume and body weight were recorded twice a week. Tumor volume was calculated as (width\(^2\) \(\times\) length) / 2.
+
+Pre-metastatic niche study. MCM (300 \( \mu \)L per mice) was intraperitoneally injected to the mice for 10 consecutive days. On Day 7, a tail vein injection of B16F10 or Luc-B16F10 cells (\( 1 \times 10^5 \))
+per mice) was given to the mice. On Day 3, 6, 10, 13, mice were euthanized for cardiac perfusion and the lung tissues were collected for further analysis.
+
+For Western-blot analysis, total proteins from the lung tissues were extracted using Tissue Protein Extraction Reagent (T-PER™, Cat. 78510, Thermo Pierce, Thermo Scientific, USA) and quantified with a Bradford Protein Assay Kit (Cat. P0010, Beyotime, Beijing, China). Samples (60 µg) were separated on 10% or 8% SDS-PAGE gels, then transferred to PVDF nitrocellulose membrane (Cat. IPVH00010, Merck Millipore). Membranes were incubated with the appropriate primary antibodies in 3% BSA, including Fibronectin (1:500, Cat. ab2413, Abcam, UK), MMP9 (1:1000, Cat. ab38898, Abcam, UK), VEGFa (1:500, Cat. ab119, Abcam, UK), TGF-β1 (1:1000, Cat. ab179695, Abcam, UK), iNOS (1:1000, Cat. ab204017, Abcam, UK), Arginase 1 (1:1000, Cat. ab124917, Abcam, UK). Antibody against GAPDH (1:10000, Cat. ab181602, Abcam) was used as control. After incubation with appropriate secondary antibody Goat anti-Mouse IgG (H+L) (1:5000, Cat. 31160, Thermo Pierce) or Goat anti-Rabbit IgG (H+L) (1:5000, Cat. 31210, Thermo Pierce), the intensity of the immunoreactive proteins was stabilized by SuperSignal® West Dura Extended Duration Substrate (Cat. 34075, Thermo Pierce) and visualized on X-ray film.
+
+For ELISA analysis, the lung tissues were ground and centrifuged to gain supernatant to measure the MMP2 (Cat. OM457413, Omnimabs, USA) and ROS (Cat. OM641674, Omnimabs, USA) level.
+
+For immunofluorescence staining, the left lung was fixed with 4% paraformaldehyde and 30% sucrose solution overnight, and embedded into paraffin and sliced into sections. The paraffin
+lung sections were deparaffinized and rehydrated, then stained with primary antibodies: αSMA (1:500, Cat. Ab7817, Abcam, UK), or CD34 (1:500, Cat. Ab81289, Abcam, UK). Secondary antibody Cy3 conjugated goat anti-rabbit IgG (1:500, Cat. 111-165-003, Jackson, USA) was utilized in 1:500 dilution and stained with DAPI before observation.
+
+For flow cytometry analysis, lung tissues harvested from mice were mechanically minced into 1-2 mm pieces using scissors and then dissociated into single cell suspension at 37 °C on a shaker for 30 min by enzymes. The digesting solution contains 2 mg/mL collagenase I (Cat. BS163, BioSharp, Germany), 2 mg/mL collagenase II (Cat. BS164, BioSharp, Germany) and DNase I (Cat. KGF008, KeyGEN BioTech., China). Digestion was stopped by adding 2 volumes PBS and filtered through a 70 μM cell strainer (Cat. CSS013070, Jet BIOFIL®, China). The cell suspension was centrifuged at 400 g for 5 min to discard the supernatant. Cell precipitations were then resuspended in 5 mL RBC lysis buffer (Cat. R1010, Solarbio, China) and centrifuged again to discard the supernatant. The single-cell-suspensions washed with PBS and resuspended were incubated with FITC-antimouse-CD45 (Cat. 553079, BD, USA), PE-antimouse-NK1.1 (Cat. 108708, Biolegend, USA), PE-antimouse-CD3 (Cat. 100205, Biolegend, USA), PE-antimouse-TER119 (Cat. 116207, Biolegend, USA), PE-antimouse-CD19 (Cat. 152407, Biolegend, USA), APC-antimouse-CD11b (Cat. 101211, Biolegend, USA), BV605-antimouse-MHC II (Cat. 107639, Biolegend, USA), BB700-CD11c (Cat. 566505, BD, USA), BV421-antimouse-Ly6c (Cat. 562727, BD, USA) and PE/CF594-antimouse-Ly6g (Cat. 562700, BD, USA), BV711-antimouse-F4/80 (Cat. 123147, Biolegend, USA), PE/Cy7-antimouse-CD103 (Cat. 121426, Biolegend, USA) antibodies in 100 μL 1% BSA
+containing 50 μL BD Brilliant Stain Buffer for 30 min at 4 °C in dark. After centrifuged and washed with PBS, cell pellets were fixed and membrane were perforated with Fix/Perm Buffer (Cat. 562574, BD, USA). The cell pellets were then stained with BV650-antimouse-CD206 (Cat. 141723, Biolegend, USA) for 40 min in the dark at room temperature. After centrifuged and washed with PBS, the stained cell pellets were analyzed by BD Fortessa flow cytometry. The data were analyzed using FlowJo software.
+
+To investigate the impact of PMN formation induced by MCM injection, metastasis development and mice survival were monitored on MCM-induced PMN model and on the mice that didn’t receive MCM injection but were inoculated directly with Luc-B16F10 cells (1 × 10^5 per mice) on Day 7. Lung metastasis was monitored twice a week by bioluminescence imaging (IVIS Spectrum, USA).
+
+Influence of peptide interference on mice PMN
+
+For mice PMN model, MCM (300 μL per mice) was intraperitoneally injected to the mice for 10 consecutive days. On Day 7, a tail vein injection of B16F10 or Luc-B16F10 cells (1 × 10^5 per mice) was given to the mice. Mice were randomly divided into 4 groups, namely control, TP5, sFD17 and FR17. Peptides, including TP5, sFD17 and FR17, were administrated separately to the mice subcutaneously from Day 3 at 40 μM/kg/day. On Day 10, mice were euthanized for cardiac perfusion and the lung tissues were collected for further analysis.
+
+The lung tissues harvested from different groups were fixed, embedded into paraffin and sliced
+into sections. To visualize the activation of lung fibroblasts and angiogenesis in pulmonary PMN, the paraffin lung sections after deparaffinization and rehydration, were stained with primary antibodies: αSMA (1:500, Cat. ab7817, Abcam, UK), or CD34 (1:500, Cat. ab81289, Abcam, UK). Secondary antibody Cy3 conjugated goat anti-rabbit IgG (Cat. 111-165-003, Jackson, USA) was utilized in 1:500 dilution and stained with DAPI before observation. To visualize the extracellular matrix environment alteration in pulmonary PMN, the lung sections were stained with appropriate primary antibodies: MMP2 (1:200, Cat. 10373-2-ap, PTG, USA), or MMP9 (1:1000, Cat. ab228402, Abcam, UK), Secondary antibody Cy3 conjugated goat anti-rabbit IgG (1:500, Cat. 111-165-003, Jackson, USA) and DAPI. To visualize the collagen deposition in pulmonary PMN, Masson's trichrome staining of the lung sections were imaged and analyzed by ImageJ to calculate the collagen volume fraction (CVF) by dividing the blue collagen area by total tissue area. And the Sirius Red Staining was also carried out and the sections were visualized under the polarizing microscope (Nikon Eclipse Ci). To investigate the recruitment of MDSC in PMN, serial sections of lung tissues were stained with periostin (1:200, Cat. 19899-1-AP, PTG, USA), or co-stained with LOX (1:200, Cat. ab174316, Abcam, UK) and Fibronectin (1:200, Cat. ab92572, Abcam, UK) antibody, or CD11b (1:2000, Cat. ab133357, Abcam, UK) and Gr-1 (1:200, Cat. ab25377, Abcam, UK) antibody separately. Secondary antibody Cy3 conjugated goat anti-rabbit IgG (1:500, Cat. 111-165-003, Jackson, USA), goat anti-rabbit IgG conjugated to HRP (1:2000, Cat. ab6721, Abcam, UK) and fluorescent TSA-488 (1:200, Wuhan Pinuofei, China) were applied according to Tyramide Signal Amplification technology. The stained sections were observed and imaged under the confocal microscope. Images were analyzed by ImageJ if necessary.
+To investigate the alteration of protein expression level in PMN, Western-blot or ELISA experiments were carried out. For Western-blot assay, protein samples were extracted separately from three independent mice from each group. Total proteins from the lung tissues were extracted using Tissue Protein Extraction Reagent (T-PERTM, Cat. 78510, Thermo Pierce, Thermo Scientific, USA) and quantified with a Bradford Protein Assay Kit (Cat. P0010, Beyotime, Beijing, China). Samples (60 µg) were separated on 10% or 8% SDS-PAGE gels, then transferred to PVDF nitrocellulose membrane (Cat. IPVH00010, Merck Millipore). Membranes were incubated with the appropriate primary antibodies in 3% BSA, including Fibronectin (1:500, Cat. ab2413, Abcam, UK), Versican (1:1000, Cat. ab270445, Abcam, UK), VEGFa (1:500, Cat. ab119, Abcam, UK), ANG2 (1:500, Cat. ab155106, Abcam, UK), MMP9 (1:1000, Cat. ab38898, Abcam, UK), MMP2 (1:500, Cat. ab97779, Abcam, UK), TGF-β1 (1:1000, Cat. ab179695, Abcam, UK). Antibody against GAPDH (1:10000, Cat. ab181602, Abcam) was used as control. After incubation with secondary antibody Goat anti-Mouse IgG (H+L) (1:5000, Cat. 31160, Thermo Pierce) or Goat anti-Rabbit IgG (H+L) (1:5000, Cat. 31210, Thermo Pierce), the intensity of the immunoreactive proteins was stabilized by SuperSignal® West Dura Extended Duration Substrate (Cat. 34075, Thermo Pierce) and visualized on X-ray film. For ELISA assay, lung tissues were ground and centrifuged to gain supernatant to measure the IL-6 (Cat. EK0411, Boster, China) level.
+
+In vivo vascular permeability assay. MCM (300 µL per mice) was intraperitoneally injected to the mice for 7 consecutive days. And the mice were randomly divided into 4 groups, namely
+control, TP5, sFD17 and FR17. Peptides, including TP5, sFD17 and FR17, were administrated separately to the mice subcutaneously from Day 3 at 40 \( \mu \)M/kg/day. On day 7, 100 mg/kg Rhodamine B-dextran (70 kDa, Cat. R9379, Sigma-Aldrich, USA) was intravenously injected to the mice. After 3 h, mice were injected with FITC-lectin (Cat. L0770, Sigma-Aldrich, USA) at 10 mg/kg through the tail-vein. Ten minutes later, each mouse was anesthetized and transcardiac perfused with 20 mL saline to remove the excess dye and followed by 5 mL of 4% formaldehyde. The lung tissues were formaldehyde-fixed and cryo-sectioned. Slices were observed and imaged by fluorescence microscopy for vascular leakage. There were 3 mice in each group and 5 visual fields were randomly chosen for each section. The relative vascular permeability was analyzed by dividing the dye leakage of each group by healthy control.
+
+Recruitment of MDSC to PMN. On Day 10 of PMN mice model, lung tissues were harvested from different treatment groups and mechanically minced and digested to obtain the single-cell-suspensions as described above. The single-cell-suspensions washed with PBS and resuspended were incubated with APC-antimouse-CD11b (Cat. 101211, Biolegend, USA) and PE/Cy7-antimouse-Ly6g (Cat. 127617, Biolegend, USA) antibodies in 100 \( \mu \)L 1% BSA for 30 min at 4 \( ^\circ \)C in dark. After centrifuged and washed with PBS, cell pellets were analyzed by BD Fortessa flow cytometry. The data were analyzed using FlowJo software.
+
+mRNA sequencing of CD11b+Ly6g+ MDSC recruited to PMN. On Day 10 of PMN mice model, lung tissues were harvested from different treatment groups and digested into single cells
+as introduced as above. The single-cell-suspensions after washing with PBS and re-suspension were incubated with APC-antimouse-CD11b (Cat. 101211, Biolegend, USA) and PE/Cy7-antimouse-Ly6g (Cat. 127617, Biolegend, USA) antibodies in 100 \( \mu \)L 1% BSA for 30 min at 4 \( ^\circ \)C in the dark. After centrifuged and washed with PBS, CD11b^+Ly6g^+ MDSCs were sorted from the PMN lungs of 10-12 individual mice from each group per sample by FACS (Beckman moflo Astrios EQ). Total RNA was extracted by TRIzol for cDNA preamplification using the NEBNext® Ultra™ Directional RNA Library Prep Kit for Illumina®, then analyzed using Qubit2.0 Fluorometer, Agilent 2100 bioanalyzer and qRT-PCR. Significantly enriched gene sets were defined as \( P \) values < 0.05 comparing to the control group.
+
+Inhibition of tumor metastasis on MCM-induced PMN lung metastasis model. The PMN models were established as introduced above, mice were randomly divided into 4 groups (n = 7), namely control, TP5, sFD17 and FR17. From day 3 to 21, mice from different groups were subcutaneously administrated with saline, TP5 (40 \( \mu \)M/kg per day), sFD17 (40 \( \mu \)M/kg per day) and FR17 (40 \( \mu \)M/kg per day) separately. Body weight was recorded every 3 days. On day 20, blood was collected from the submarginal ocular venous plexus under anesthesia for blood tests including complete blood count, alanine aminotransferase (ALT), aspartate aminotransferase (AST), blood urea nitrogen (BUN) and serum creatinine (CREA). On day 28, mice were euthanized and major organs, such as heart, liver, spleen, lung, kidney and thymus were collected after cardiac perfusion. Lung tumor nodules were then counted under the stereo microscope. The major organs were fixed, embedded into paraffin for Hematoxylin & Eosin staining. The thymus
+coefficient was calculated as the thymus weight divided by the body weight and the spleen coefficient was calculated as the spleen weight divided by the body weight.
+
+Inhibition of tumor metastasis post-surgery. The post-surgery metastasis model was established as illustrated above. Mice were randomly divided into 3 groups, namely control, anti-PD1 and FR17. For FR17 treatment, peptide was subcutaneously administrated to the mice from Day 7 to Day 25 at the dose of 40 \( \mu \)M/kg per day. For anti-PD1 treatment, 100 mg anti-PD1 (Cat. BE0146, Bio X Cell, USA) was given by *i.p.* injection twice per week starting from day 3 post tumor resection and given two times per week from Day 15 to Day 25 for a total of 4 times. Lung metastasis was monitored twice a week by bioluminescence imaging (IVIS Spectrum, USA) until death. The recurrence of the excised subcutaneous tumor was closely monitored and tumor volume was measured and calculated following the ellipsoid volume formula: (width\(^2\) \(\times\) length) / 2.
+
+Statistics
+
+Statistical analysis was performed using GraphPad Prism 8.0.1 (GraphPad Software, CA, USA). Data were presented as means \( \pm \) SD. Statistical evaluation of differences between experimental groups was performed by one-way ANOVA followed by Tukey’s multiple comparisons test. Statistical significance was considered at least at \( p < 0.05 \).
+1 Abraham, M. J. et al. GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX, **1**, 19-25 (2015).
+
+2 Humphrey, W., Dalke A. & Schulten, K. VMD: visual molecular dynamics. *J. molecular graphics*, **14**, 33-38 (1996).
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SIV9.pdf
+• SIV9withoutauthorsinformation.pdf
+• SIV9withoutauthorsinformation.pdf
+• nrreportingsummary20210809T100118.224.pdf

0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/preprint/preprint.md

@@ -0,0 +1,274 @@
+Inter-species metabolic interactions in cheese flavour formation
+
+Chrats Melkonian (chrats.melkonian@gmail.com)
+Utrecht University & Wageningen University & Research https://orcid.org/0000-0002-1448-5542
+Francisco Zorrilla
+Medical Research Council Toxicology Unit, University of Cambridge
+Inge Kjærbølling
+Novozymes
+Sonja Blasche
+University of Cambridge
+Daniel Machado
+Norwegian University of Science and Technology
+Mette Junge
+Chr. Hansen
+Kim Soerensen
+Chr. Hansen
+Lene Andersen
+Chr. Hansen
+Kiran Patil
+University of Cambridge https://orcid.org/0000-0002-6166-8640
+Ahmad Zeidan
+Chr. Hansen
+
+Article
+
+Keywords: Cheddar cheese, Microbial interactions, multi-omics, metabolic modeling, Streptococcus thermophilus, Lactococcus lactis, Lactococcus cremoris
+
+Posted Date: March 16th, 2023
+
+DOI: https://doi.org/10.21203/rs.3.rs-2574132/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+Additional Declarations: There is NO Competing Interest.
+Version of Record: A version of this preprint was published at Nature Communications on December 21st, 2023. See the published version at https://doi.org/10.1038/s41467-023-41059-2.
+Inter-species metabolic interactions in cheese flavour formation
+
+Chrats Melkonian1,6,7,*, Francisco Zorrilla2, Inge Kjærbølling1, Sonja Blasche2, Daniel Machado3, Mette Junge4, Kim Ib Soerensen4, Lene Tranberg Andersen5, Kiran R Patil2, and Ahmad A. Zeidan1,*
+
+1Systems Biology, R&D Discovery, Chr. Hansen A/S, 2970 Hørsholm, Denmark
+2Medical Research Council Toxicology Unit, University of Cambridge, Cambridge, UK
+3Department of Biotechnology and Food Science,Norwegian University of Science and Technology, Trondheim, 7491, Norway
+4Global Application, Food Cultures & Enzymes, Chr. Hansen A/S, 2970 Hørsholm, Denmark
+5Strain Development, R&D Discovery, Chr. Hansen A/S, 2970 Hørsholm, Denmark
+6Theoretical Biology and Bioinformatics, Science for Life, Utrecht University, Utrecht, the Netherlands
+7Bioinformatics Group, Wageningen University and Research, Wageningen, Netherlands
+*Authors to whom correspondence should be addressed: chrats.melkonian@gmail.com, dkahze@chr-hansen.com
+
+ABSTRACT
+Cheese fermentation and flavour formation are governed by complex biochemical reactions driven by polymicrobial activity. While the compositional dynamics of cheese microbiomes is relatively well mapped, the mechanistic role of microbial interactions in flavour formation is yet unknown. We microbially and metabolically characterised a year-long Cheddar cheese process using a commonly used starter culture containing Streptococcus thermophilus and Lactococcus strains. By using an experimental strategy whereby certain strains were left out from the starting mixture, we identified the critical role of S. thermophilus in boosting Lactococcus growth and in shaping flavour compound profile. Controlled milk fermentations with systematic exclusion of single Lactococcus strains, combined with genomics, genome-scale metabolic modelling, and metatranscriptomics, indicated that proteolytic activity of S. thermophilus relieves nitrogen limitation for Lactococcus and boosts de novo nucleotide biosynthesis. While S. thermophilus had large contribution to the flavour profile, L. cremoris also played a role by limiting diacetyl and acetoin formation which leads to off-flavour when in excess. This off-flavour control could be attributed to different metabolic re-routing of citrate between L. cremoris and other L. lactis strains. Further, closely related L. lactis strains exhibited different interaction patterns with S. thermophilus highlighting the importance of strain-specificity in cheese-making. Overall, our results bring forward the critical role of competitive and cooperative microbial interactions shaping cheese flavour profile.
+
+Keywords:    Cheddar cheese, Microbial interactions, multi-omics, metabolic modeling, Streptococcus thermophilus, Lactococcus lactis, Lactococcus cremoris
+Introduction
+
+Fermented foods based on microbial consortia (e.g., cheese, kefir and kombucha) constitute a large part of the modern diet with many reported health benefits [1, 2]. In cheese, well defined Starter Lactic Acid Bacteria (SLAB) cultures are used to ensure consistent and flavorful products. During cheese making, microbes encounter dynamic conditions characterized by different nutrient availability and different stresses, starting with an initial feast during milk fermentation, followed by a long period of famine during cheese ripening. Cheese-making thus provides a controlled system with well characterized dynamics to understand the role of microbial interactions in shaping the characteristics and quality of fermented foods [3].
+
+In industrial cheese making, SLAB cultures are responsible for milk acidification via lactose fermentation. They are predominantly composed of mesophilic strains of *Lactococcus*, including the species *L. lactis* and *L. cremoris* as well as thermophilic strains of *Streptococcus thermophilus*. A rich literature exists on the physiology of both species in mono-cultures [4, 5, 6], and genome-scale metabolic models have been used to model their metabolism [7, 8]. In contrast, the interactions between the two taxa is still largely unknown from a mechanistic perspective despite indications of strong inter-dependencies [9]. Further, the current knowledge on cheese microbial interactions is limited to the scope of simplified artificial milk media, concerns pairwise associations, or focus only on the short time interval during the milk acidification. Less is known about the interactions involving more than two SLAB strains and how interaction networks evolve during the cheese ripening steps. Metabarcoding and metagenomics approaches have also been used to study the cheese microbiome [10, 11, 12]. While these studies provide a good overview on the community dynamics during cheese-making using higher-order taxonomy, they are limited in uncovering inter-species interactions and strain-level diversity.
+
+To uncover the role of microbial interactions and interacting agents, we used a strain dropout strategy whereby one single strain or a strain group was left out from the starter culture, and the cheese-making process was characterized in its entirety. We used an industrially relevant SLAB culture, containing: one *Streptococcus thermophilus* (ST), two major *L. lactis* (LLm1 & LLm2), one major *L. cremoris* (LC),
+and a mixture of 21 L. cremoris and L. lactis strains in smaller fractions (hereafter, *Lactoccocus* blend (LB)). We investigated the effect of leaving out *S. thermophilus* and *Lactoccocus* blend in a year-long cheddar-making experiment (Fig. 1a); followed by controlled milk experiments investigating hypotheses generated from the year-long experiments. For the latter, we used an integrative systems biology approach that combined different layers of biological information (Fig. 1b).
+Results
+
+1 S. thermophilus supports lactoccoci growth and shapes the metabolic profile during cheese ripening
+
+To study the interactions between the members of the SLAB culture, we started by quantifying the population dynamics during a year-long ripening cheese making experiment (see Methods). We used for variations of starter cultures: i) containing all member species of an industrial starter culture, viz., S. thermophilus (ST), L. cremoris (LC), L. lactis (LLm1 and LLm2), and Lactococcus blend (LB); ii) the same as (i) but prepared independently (HP); iii) excluding LB; and iv) excluding ST (Fig. 1a). In the three conditions that included S. thermophilus, we observed a trend of slow decline in the population of lactoccoci and S. thermophilus, starting with 9.25 and ending at 8 log10 CFU/g (Fig. 2a-c). In the single condition that excluded S. thermophilus, we observed a much steeper decline in lactoccoci population, ending at 6.5 log10 CFU/g (Fig. 2a-b). Also, the declining lactoccoci population exhibited different trajectories between the two batches, albeit both converging at 9 and 12 months (Fig. 2a-b). The non-SLAB count increased from 0 to 7.5 log10 CFU/g, in the highest case (Fig. 2d). Based on 95% confidence interval estimates, we observed high variability among the non-SLAB populations, but with no appreciable overall difference between the conditions (Fig. 2d). Leaving out Lactococcus blend or changing the way of packing the different strains in the culture did not result in any visible response to the community dynamics. Thus, inter-species interactions, especially involving S. thermophilus, appear to be driving the overall population dynamics.
+
+To measure the effect of the microbial interactions in the resulting phenotype of the cheese, we used targeted metabolomics analyses. Across conditions, the biggest change in the metabolic profiles of cheese was observed between the two-week mark and three months (Fig. 2e). This can be attributed to changes in peptide composition followed by the general increase in amino acids concentrations (Fig. 2f). Furthermore, cheese metabolomes continue to change at a slower pace between the three- to twelve-month marks, resulting in different end profiles. A notable time-depended accumulation of compounds after the third month was observed on acetic acid, tyramine, g-amino butyric acid, putrescine and cadaverine
+(Fig. 2f & Supplementary Fig. 1). The cheeses produced without S. thermophilus formed a separate cluster relative to the rest (Fig. 2e). This pattern can be ascribed to compounds with significantly different concentrations across strain removal conditions, especially when S. thermophilus was absent. The majority of these compounds are peptides that were either not accumulated, accumulated in lower amounts or accumulated in higher amounts without the addition of S. thermophilus. Lactose, galactose and lactic acid concentrations were also found to be significantly different when S. thermophilus was absent (Fig. 2g, ANOVA \( F(3,36) = 208 \pm 334, p < 0.001, \eta^2 = 0.86 \pm 0.08 \) & Supplementary Fig. 2). This difference was already notable in the samples from the second week of cheese ripening (Fig. 2h & Supplementary Fig. 3a,c&e). These results points towards the significant influence of S. thermophilus growth during milk fermentation and its non-growing yet active cells during the long term cheese ripening.
+
+The presence of S. thermophilus was found to benefit both the growth of the Lactococcus community and the final metabolic profile of the cheddar cheese. Notably, the growth benefit is not clearly visible before two weeks to three months, a time-frame that may be considered prohibitively long for experiments in the laboratory. Instead, these effects become more evident as the experimental time-frame expands to one year. The cheese metabolic profiles indicated differences in peptide composition between the conditions, a result likely stemming from the non-growing metabolic activity of S. thermophilus. In accordance with the literature, the presence of S. thermophilus reveals a few key metabolic changes from the milk fermentation; the complete consumption of lactose and the production of galactose [13, 14]. Together, the compositional and metabolic profiles led us to hypothesize that both early and late S. thermophilus activity is critical for the long term effect on Lactococcus population.
+
+2 S. thermophilus has strain-specific interactions with the members of Lactococcus lactis community
+
+To assess the role of S. thermophilus, we performed controlled experiments wherein additional strains were removed. Samples were taken when milk fermentation reached the transition from exponential to stationary phase (at pH 5) and gene expression was monitored using metatranscriptomics. To further investigate the microbial interactions within the SLAB culture, we also analyzed the genomes of the individual strains
+in the culture and their phylogenetic relationship (Fig. 3a). The phylogenomic analysis separated the species L. lactis and L. cremoris into two distinct groups. Two of the main Lactococcus strains, LC and LLm2, have no close relatives within the culture community. For the third main Lactococcus strain, LLm1, we identified two close relatives (LL-LB01 & LL-LB02), belonging to the Lactococcus blend (Fig. 3 & Supplementary Table 1). The rest of the strains in the Lactococcus blend formed distinct sub-clusters within the cremoris clade (Fig. 3). By employing pan-genome analysis on Lactococcus strains, we identified singletons (i.e. a gene found only in one genome) for each member of the culture. The highest number of singletons was harbored by the main LLm2 followed by LC, and LLm1 where its number of singletons was in the same range as the Lactococcus blend strains, with numbers 288, 134, 79 and average of 53.6 ± 18.6, respectively. In parallel, we evaluated both the presence and activity level (using metatranscriptomics) of the Lactococcus strains by calculating percentile of the transcribed singletons per genome. We found a high number of transcribed singletons on all the main Lactococcus strains, while this number varied considerably among the strains of the Lactococcus blend, raising questions regarding whether or not these strains were active (Fig. 3a & Supplementary Table 4). We hypothesized that the temperature dynamics during the experiment could influence the strain activity. To test this, by carrying out milk fermentation using each of the individual Lactococcus strains at different temperatures, we calculated a proxy for their temperature stress tolerance (Pearson correlation between curves of 30 against 37, 40 and 43 degrees). The results showed that the L. lactis strains have higher temperature stress tolerance in the tested ranges, including Lactococcus blend strains LL-LB01 and LL-LB02 (PCC 0.96 ± 0.05). The L. cremoris strains exhibited low temperature stress tolerance (PCC 0.7 ± 0.24), with the main LC being one of the most stress tolerant strains within this group (PCC range 0.84-0.99) (Supplementary Fig. 5a&b, Fig. 3a).
+
+We next explored the extent to which we could discriminate between the different strains. The number of unique k-mers in each genome ranged from 368 to 622455 and they were found to be dependent on the presence of close phylogenetic relative strains (Fig. 3a). The k-mer transcription abundance indicated the feasibility of discriminating between the three main Lactococcus strains (LC, LLm1 and LLm2) and the one S. thermophilus strain, but not between the strains in the Lactococcus blend (Supplementary Fig. 4). Using differential expression analysis, we investigated the interaction effect on S. thermophilus
+transcriptome by leaving out each *Lactococcus* strain/blend from the culture. We found the strongest response when LLm1 was left out followed by *Lactococcus* blend, LC and LLm2 with 291, 182, 21 and 1 gene(s) significantly differential expressed, respectively (Fig. 3b-e, Supplementary Table 6 & sections below). Although both LLm1 and LLm2 belong to *L. lactis* clade, the latter strain stands out by having the highest number of singletons as well as having the second highest number of unique k-mers (Fig. 3). By removing strains with a different degree of relatedness we identified that different *Lactococcus* strains have distinct effects on gene expression of *S. thermophilus*. This finding can be attributed to the different genetic background and the diverse phylogenetic relationships within the *Lactococcus* community, highlighting the diversity of potential microbial interactions between *S. thermophilus* and the *Lactococcus* community.
+
+**3 S. thermophilus proteolytic activity may benefit *Lactococcus* community by providing nitrogen sources**
+
+To investigate mechanisms of inter-species interactions, we generated genome-scale metabolic models for the individual strains in the culture and performed flux balance analysis -based simulations. To explore the effect of changes in growth medium composition, we generated models gap-filled on relevant variations of milk-mimicking media. Supplementary Fig. 6a shows overall model statistics for the different species, while Supplementary Fig. 6b shows the specific reactions added through gap-filling on different media across the models (see Supplementary Section 0.1). Flux balance simulations with individual models were carried out using the different model sets in aerobic and anaerobic variants of fermented and non-fermented milk (Fig. 3a,b, Supplementary Fig. 6c,d). All species consistently consumed lactose in simulations where growth was feasible. Several differences in amino acid uptake/secretion were observed among the strains. The branched chain amino acid valine was exported by *S. thermophilus* in all but one simulation condition, whereas LC, LLm1, and LLm2 consumed this amino acid in four, six, and three simulations conditions, respectively. Additional amino acid uptake/secretion predictions are reported at Supplementary Section 0.1
+
+To assess potential cross-feeding, we performed community simulations using the SMETANA framework (see Supplementary Section 0.1). We used the calculated SMETANA scores, ranging from 0 to
+1, as a measure of predicted interaction confidence (0 being lowest confidence and 1 being the highest) [15]. These simulation indicate that S. thermophilus provides valine to Lactococcus strains in all three community simulation conditions (gap-filled on rich aerobic milk & simulated on minimal aerobic milk, gap-filled on rich anaerobic milk & simulated on minimal aerobic milk, and gap-filled on rich anaerobic milk & simulated on minimal anaerobic milk) where exchanges were predicted between community members (Figure 3c). Not only was this interaction consistent, it also occurred with a SMETANA score of 1 in every case, suggesting that this is a key ecological interaction for maintaining the composition and function of the community. Serine exchanges from S. thermophilus to LLm1 were predicted in 2 out of 3 community simulation conditions, with an average SMETANA score of 0.38±0.01. Exchanges from S. thermophilus to LC involving glycine and ammonium were predicted in 1 out of the 3 simulation conditions, with SMETANA scores of 0.3 and 0.29, respectively (Figure 3c). Finally, an exchange from S. thermophilus to LLm2 involving alanine was predicted under only 1 simulation condition with a SMETANA score of 0.42. Overall, the metabolic simulations strongly indicate cross-feeding between the community members, with S. thermophilus standing out as a key donor (Figure 3c).
+
+To complement the genomic and metabolic modeling analysis, we investigated the changes in S. thermophilus and Lactococcus community transcriptome. The principal component analysis showed pattern consistency with the differential transcriptome analysis (Fig. 3c-f, Supplementary Section 0.2). Regarding changes in S. thermophilus’s transcriptome, we found up-regulated genes annotated as oligopeptide-binding protein (AmiA) and oligopeptide transport system permease protein (OppB, OppC, OppD & OppF) (see Supplementary Section 0.2). Seven down-regulated genes are annotated as transporters, such as glutamine ABC transporter permease protein (GlnP), cadmium, cobalt and zinc/H(+)-K(+) antiporter (Supplementary Section 0.2). Further, transcriptional repressors such as the catabolite control protein A (CcpA) and nitrogen-metabolism-regulating-proteins such as GTP-sensing transcriptional pleiotropic repressor (CodY) were found to be transcriptionally active. In line with the putative interactions predicted using metabolic modelling, we found a gene annotated as branched-chain amino acids (BCAA) transaminase expressed in S. thermophilus, suggesting that valine biosynthesis is active. Moreover, we found a gene annotated as BCAA transport system 2 carrier protein (BrnQ) as well expressed, suggesting that S. thermophilus is secreting valine. Complementary differential pathway enrichment analysis supported
+further that valine biosynthesis was enriched in S. thermophilus metatranscriptome (Fig. 4d-e). Overall, the metabolic modeling and metatranscriptome analysis together suggests that S. thermophilus can act as a branched chain amino acid donor to Lactococcus. It is well known that Lactococcus harbor a number of amino acid auxotrophies including valine [16]. In addition, the SMETANA simulations provide insights into the metabolic strategies that may be employed by community members under challenging media conditions.
+
+We next investigated transcriptional change in the Lactococcus community in the presence and absence of S. thermophilus. As it is not possible to distinguish between the transcriptome profiles of all the different Lactococcus strains, we merged these into one group for pan-genome analysis and pan-metabolic modelling. Strain-to-strain transcriptional changes were investigated when possible (see Supplementary Section 0.2). The most up-regulated OGs in Lactococcus when S. thermophilus is left out are involved in nitrogen assimilation, namely nitrogen regulatory protein P-II, related transcriptional regulator (GlnR), ammonium transporter (amtB), glutamine synthetase (glnA) and glutamine transport ATP-binding protein (glnQ). Additionally, we confirmed that GlnR in Lactococcus regulates the three operons, namely amtB, glnA, and glnQ (Fig. 4f-i, Supplementary Table 2 & Supplementary Section 0.2). The purine and pyrimidine biosynthesis pathways are deferentially expressed in a coordinated manner. Several genes in the pathways were up-regulated, in particular the genes associated with the reactions leading to uridine and guanine synthesis (Fig. 4j & supplementary). Both pathways are connected via L-glutamine synthetase. Expression patterns in amino acid metabolism show a mixed pattern of both up- and down- regulation (see Supplementary Section 0.2). Overall, the metatranscriptomics analysis further supports that S. thermophilus cross-feed Lactococcus community. One hypothesis that could explain this is that S. thermophilus has a higher proteolytic activity, thereby providing more peptides and amino acids (available nitrogen source) in the culture which Lactococcus takes advantage of. When S. thermophilus is absent, there is a lower degree of nitrogen availability causing upregulation of the nitrogen assimilation and nucleotide syntheses pathway in Lactococcus.
+4 Cheese flavor compounds are strongly influenced by the interactions within *Lactococcus* community
+
+To elucidate the role of interactions between the SLAB strains in the development of cheese flavor, we performed a targeted metabolomics analysis. Leaving out *L. cremoris* (LC) led to the strongest metabolic response followed by a weaker and different response when leaving out LLm1, *S. thermophilus* and the *Lactococcus* blend (see Supplementary Section 0.3). Surprisingly, leaving out LLm2 did not lead to a marked change in the metabolic profile of the acidified milk (Fig. 5a). Six flavor compounds, namely heptanal, hexanal, 2-ethyl-furan, 2,3-pentanedione, diacetyl and acetoin, were either detected at significantly higher concentrations or only produced when this main *L. cremoris* strain was removed (Fig. 5b-c & Supplementary Fig. 10). A graphical summary of the metabolome differences between the whole culture and the culture lacking LC is shown in Fig. 5d. Many of the flavor compounds found were significantly altered by the strain removal strategy, with the majority corresponding to byproducts of secondary metabolism with unknown microbial biosynthesis pathways or gene clusters. Yet, investigation of their patterns highlights the numerous ways these compounds may cross-feed within the SLAB culture. For example, 2,3-Pentanedione could be donated by *S. thermophilus* to LC while ethyl hexanoate, ethyl acetate and 2-Methyl-3-thiolanone likely produced by LC.
+
+C4 aroma compounds such as diacetyl and acetoin are known derivatives of citrate metabolism with characteristic contribution to buttery-like aroma. Although desirable in small quantities, e.g less than 0.05 rag. per 100 g of diacetyl content, higher amount could lead to off-flavors [17, 18]. Therefore, we investigated whether the increase in diacetyl and acetoin concentrations triggered by the removal of LC could be explained by the metatranscriptome profiles. Indeed, we found differential expression patterns in genes related to diacetyl and acetoin metabolism. Therefore, we examined the gene expression levels in the metabolic route from citrate towards diacetyl and acetoin through pyruvate as well as an alternative route towards \( \alpha \)-ketoglutarate. The citrate-sodium symporter that is present only on *Lactococcus* strains was highly expressed implying active uptake of the citrate available in milk by *L. lactis*. From citrate, *Lactococcus* strains showed expression of the genes associated with the pyruvate
+carboxylase and acetolactate synthase reactions, whereas only LC and LLm1 had the genes related to acetolactate decarboxylase expressed at high levels. On the other route towards \( \alpha \)-ketoglutarate and further, genes associated with aconitate hydratase and isocitrate dehydrogenase are only expressed in LC and *S. thermophilus* (Fig. 5e & Supplementary Fig. 11). Metabolic modeling simulations across the different medium conditions are largely corroborated by the metatranscriptomics results with the exception of three reactions involving acetoin. These are acetolactate decarboxylase (ACLDC), which does not carry a flux in the LC model, and 2,3-butanediol dehydrogenase (BTDD_RR) and diacetyl reductase (ACTD), which carry fluxes in LLm1 model but not in LC model (Fig. 5e & f). Taken together, the presence of the main *L. cremoris* strain in the cultures results in the prevention of an undesirable increase in the levels of some key flavor compounds, including diacetyl and acetoin. This can be attributed to the degradation of diacetyl, e.g., through the action of some alcohol dehydrogenases from LC. Alternatively, LC might be competing with the other strains in the culture on citrate and subsequently converting it to other products than diacetyl and acetoin, such as ethyl-hexanoate, ethyl-acetate and 2-methyl-3-thiolanone, which normally happens in mixed culture fermentation following the rapid drop of redox potential in the beginning of fermentation. This would result in the reduction of the overall amount of citrate available for diacetyl and acetoin formation.
+
+Discussion
+
+Interactions in microbial communities are reported in a wide range of ecosystems [19, 20, 21, 22, 23]. While such interactions are widely reported also in food microbial communities [24, 3, 25, 26, 27, 28], very few studies have provided insights into the molecular agents that mediate the interactions. Uncovering interactions *in situ* is particularly challenging due to complexity of the communities involved as well as that of the medium such as milk. Here, we combined genomics, metatranscriptomics, metabolomics, and metabolic modelling to uncover key microbial interactions in Cheddar cheese-making. Notably, we used industrial strains and a full cycle of year-long cheese ripening. We report competitive and cross-feeding metabolic interactions between the lactic acid bacteria used in Cheddar cheese production. Firstly, *L. cremoris* strain competes with *L. lactis* for the milk’s available citrate, resulting in accumulation of key
+metabolites such as diacetyl and acetoin in the final product. Such interactions take place within the first 5 hours of milk fermentation and strongly influences the final cheese flavor. Secondly, S. thermophilus provides the necessary nitrogen source to the Lactococcus community, explaining the significant one-year long growth benefit of the Lactococcus strain population and the different final cheese metabolome profile. In addition, Lactococcus strains affect the activity of S. thermophilus differently. While the presence of one L. lactis strain hardly influences the activity of S. thermophilus and the development of the final cheese flavor profile, the presence of a different L. lactis strain influences both significantly. Often in literature the attempts to study microbial interactions disregard the strain diversity. Yet, recent studies highlight the importance of strain interactions, claiming the major role of those in predicting eco-evolutionary dynamics [29]. Our results show how strain-specific metabolic interactions between microbes shape the biochemical profile of cheese, and provide targets towards the rational design and assembly of microbial communities with the aim of fine-tuning cheese flavor. More broadly, the study provides a blue-print to uncovering in situ interactions in complex food microbial ecosystems.
+
+Methods
+
+Starter lactic acid bacteria culture
+
+A cheese making experiment was designed to investigate the effect of removing selected strains from a defined-strain SLAB culture. Wild-type strains were obtained from the Chr. Hansen Culture Collection and were originally isolated from dairy products or cultures. By applying this strain removal strategy, it becomes possible to gain insights into the individual role of the strain but also its interactions within the microbial community. The original defined-strain SLAB culture consisted of 5 different bulks: 1 bulk with single strain S. thermophils, 1 bulk with a multi strain mixture of 21 Lactococcus strains belonging to both species of L. lactic and L. cremoris [30], which were inoculated as a single component (hereafter, Lactococcus blend), 2 bulks with single strain L. lactis, and 1 bulk with single strain L. cremoris. The original defined-strain SLAB culture was composed from of 25 strains and all bacterial and all strains applied are listed in Supplementary Table 1.
+
+To apply the strain removal strategy 4 defined-strain SLAB cultures were designed. Condition ALL,
+corresponded to the original defined-strain SLAB composed from of 25 strains packed under industrial conditions. Condition -HP, corresponded to the original defined-strain SLAB composed from of 25 strains packed under non-industrial condition by hand. Condition -LB, corresponded to the removal of the Lactococcus blend. Condition -ST, corresponded to the removal of the S. thermophils strain. The amount added varied between the conditions to target the same acidification profile in the cheese vat while the cheeses making process was kept constant. The target was to obtain a pH of 5.35 at milling and a final moisture in non-fat solids (MNFS) of the cheeses at 54% ± 0.5% measured after 2 weeks of ripening. The designed defined-strain SLAB for conditions -ALL, -HP, and -LB were all dosed using 8.1g of culture/100L, whereas the designed defined-strain SLAB for condition -ST was dosed at a concentration of 17.3 g/100L to compensated for missing contribution to acidification from S. thermophils.
+
+Cheese making
+
+Cheeses were produced at the Application and Technology Center of Chr. Hansen A/S (Hørsholm, Denmark). The composition of the pasteurized cheese milk (organic milk, Naturmælk, Denmark) was measured using a MilkoScan™ (FOSS, Hillerød, Denmark), and the fat level was adjusted with pasteurized cream (organic cream 38% fat, Naturmælk, Denmark) to obtain a protein to fat ratio of 0.90. The cheese milk was heated to 32°C and each cheese vat was filled with 150 kg. The milk was ripened for 40 min with the starter culture before the coagulant (CHY-MAX® Plus, Chr. Hansen A/S) was added. After 30 min of coagulation, the gel was cut into cubes 10x10 mm. The whey and the cheese grains were stirred for 10 min before heating to 38°C over 40 min and the final scalding and stirring period was 45 min before whey drainage at pH 6.4-6.5. After 20 min, the cheese curd was cut into 12 blocks and rearranged in blocks 3 times during the next 75 min until pH 5.35 was obtained. The blocks where then milled, and the chips were dry salted (1.7-2.0% salt in dry matter) by manually adding and mixing the salt 2 times during 15 min. The salted chips were mixed every 5 min over a period of 30 min. Following molding, the chips were pressed at 2 bars for 15 min followed by 5 bars for 17 hours. The cheeses were vacuum-packed and stored at 9°C until sampling after 0.5, 3, 6, 9 and 12 months. The cheese gross composition (moisture, fat, protein and total solids) was estimated after 2 weeks of ripening by FoodScan™ (FOSS, Hillerød, Denmark). The NaCl content was estimated by analyzing the chloride concentration by automated potentiometric
+endpoint titration (DL50, Mettler-Toledo A/S, Glostrup, Denmark). pH was measured potentiometrically (PHC2002-8, Radiometer Analytical SAS, Lyon, France) in a paste prepared by mixing 10 g of grated cheese with 10 ml of deionized water with a wooden spatula. All analysis were carried out in duplicate.
+
+In summary, for the one year-long ripening cheese making experiment we inoculated three variations of the SLAB cultures into four milk tanks. Two milk tanks were inculcated with the complete set of strains, one used as the control and the other to test a different approach of culture inoculations (All and All_HP, respectively). The two other variations of the culture were prepared using the ’strain removal strategy’ with the aim of testing the effect of L. lactis blend (All-LB) and of the single S. thermophilus (All-ST) during cheese-making. Following the acidification phase, we sampled the resulting cheddar cheeses in 5 time points over the one year period. We followed two batch experiments started on two consecutive days.
+
+Microbial community dynamics
+
+Grated cheese (5.0 g) was mixed with 45.0 g of autoclaved 2% sodium citrate buffer, pH 7.5, 46°C. The mixture was homogenized in a Stomacher® blender for 4 minutes at medium setting to dissolve the cheese and suspend the bacteria present. Sequential of 10-fold dilutions were prepared as required in 0.1% peptone/0.15M NaCl in water, pH 7.0. The total population of Lactococcus spp. was determined on pour plated M17 Agar (Difco, USA) after aerobic incubation for 5 days at 30°C. The total population of thermophilic cocci (ST) was determined on pour plated M17 Agar after aerobic incubation for 3 days at 37°C. The total population of non-starter lactic acid bacteria (NS-LAB) was determined on overlaid Rogosa Agar (Sigma Aldrich, USA) after incubation for 7 days at 37°C. The microbial analysis was carried out in duplicate at each sampling point.
+
+Carbohydrates and organic acids in cheese
+
+The content of lactose, glucose, galactose, lactate, citrate and acetate were quantified using a Dionex ICS-3000 RFIC-EG™ dual system equipped with an amperometric detector (Dionex, Sunnyvale, CA, USA). The separation was performed using an anion-exchange column (CarboPac® PA20, 3x150 mm, 6.5 µm) and an ion-exclusion column (IonPac® ICE-AS6, 9x250 mm, 8 µm). Grated cheese (3.0 g) was mixed with 15 ml 83 mM PCA containing 2 mM Na-EDTA, 33mM arabinose and 48mM 2-hydroxyisobutyric acid
+(internal standards) and rotated for 30 min at room temperature. This extract was subsequent centrifuged (5000xg, 30 min, 4°C) and the supernatant was filtered (0.45 µm). The supernatant was diluted (600-fold) before analysis. For analysis of carbohydrates 25 µL was injected and the separation was performed at 26°C with a flow rate of 1.3 mL/min increasing the KOH concentration as follow: 1 mM KOH for 5 min, from 1 mM to 20 mM KOH over 0.5 min, then kept at 20 mM KOH for 6.6 min min before re-equilibration to 1 mM KOH over 5 min. For analysis of organic acids 50 µL was injected and the separation was performed at 30°C with a flow rate of 1 mL/min using 0.4 mM heptfluorobutyric acid in 5% acetonitrile as eluent. The analysis was performed in duplicate at each sampling point.
+
+Proteolysis in cheese
+
+Grated cheese (2.0 g) was mixed with 18 ml distilled water and blended for 2 min at 25,000 rpm (Ultra-Turrax model T25). This suspension was used for determination of total nitrogen (TN), Non-Casein nitrogen (NCN), and non-protein nitrogen (NPN) using the fractionation procedure described in standard NF ISO 27871 (ISO, 2011). The nitrogen content for each fraction was determined using the Kjeldahl method as described in standard NF EN ISO 8968-1 (ISO, 2014). The level of primary proteolysis (NCN) and secondary proteolysis (NPN) were expressed as percentage of TN found in the cheese. The analysis was performed in duplicate at each sampling point. Finally, the metabolic profile of the cheeses was quantified by measuring 30, 3, 28 and 248 features belonging to the classes of acids, sugars, flavor-related organic compounds and peptides, respectively.
+
+Genome sequencing and assembly
+
+Genomic DNA for de novo short read whole-genome shotgun sequencing (WGS) was extracted from 1 mL of overnight culture (M17 Broth (Difco) supplemented with glucose) of each of the strains (at OD600 = 1) with DNeasy Blood and Tissue kit on QiaCube system (Qiagen, Germany) following the manufacturer’s protocol. Prior to extraction, cell pellets were washed twice in TES buffer (50 mM TRIS pH 8.0, 1 mM EDTA pH 8.5, 20% sucrose) and then resuspended in 180 uL of pre-lysis TET buffer (20 mM TRIS-Cl pH 8.0, 2 mM EDTA pH 8.5, 1,2% Triton X-100, 20 mg/mL lysozyme, 2 µL of 25U/µL mutanolysin, 4 µl of 100 mg/mL RNase A). Genomic libraries were prepared using KAPA LTP Library
+Preparation Kit KK8230 (Roche, Switzerland) on a Biomek 4000 Liquid Handler (Beckman Coulter, USA). A portion of 500 ng of genomic DNA diluted in EB buffer (Tris-Cl, pH 8.0) was mechanically fragmented on Bioruptor® Standard (Diagenode, USA) with 12 sonication cycles (30 sec ON/OFF) to obtain an average fragment size of 300 bp (CV%: 1.5). Fragmented DNA was processed following the KK8230 kit manufacture’s protocol. Following the adapter ligation step, adapter-modified DNA fragments were enriched by 8-cycle PCR. AMPure XP (Beckman Coulter) paramagnetic beads were used for clean-ups to purify fragments at average size between 450 to 550 bp. Concentration of gDNA and double stranded DNA libraries were measured by Qubit® Fluorimeter using Qubit dsDNA Broad range and Qubit 1x dsDNA HS assays (Thermo Fisher Scientific, USA), respectively. Average dsDNA library size distribution was determined using an Agilent HS NGS Fragment (1-6000 bp) kit on an Agilent Fragment Analyzer (Agilent Technologies, USA). Libraries were normalized and pooled in NPB solution (10 mM Tris-Cl, pH 8.0, 0.05% Tween 20) to the final concentration of 10 nM. Following denaturation in 0.2 N NaOH, 10 pM of pooled libraries in 600 μL ice-cold HT1 buffer were loaded onto the flow cell provided in the MiSeq Reagent kit v3 (600 cycles) and sequenced on a MiSeq platform (Illumina Inc., San Diego, USA) with a paired-end protocol and read lengths of 301 nucleotides.
+
+All processing of the short reads was done in either CLC Genomics Workbench versions 9.5.3, 9.5.4 or 10.1.1. The short reads were mapped with default parameters to the reference sequence of the phage Phi X 174 using the tool “Map reads to reference”. Unmapped reads from the mapping were trimmed for quality using the PHRED score 23 as the threshold and with the non-default parameter of discarding reads that were less than 50 base pairs long using the tool “Trim Sequences”. The trimmed reads were de novo assembled with default parameters except for the minimum contig length was set to 600 base pairs using the tool “De Novo Assembly”. Afterwards, a decontamination step was performed where contigs with low depth of coverage were removed using a custom plugin written by Qiagen. The decontamination step first removes all contigs where the depth of coverage is below 15X and afterwards removes all contigs where the depth of coverage is below 25% of the median depth of coverage for the entire genome assembly. Gene calling of the filtered contigs was done with Prodigal version 2.6.2 using the default parameters. Finally, the genome assemblies with annotated genes were functionally annotated with BLAST against a local annotation database using a custom plugin written by Qiagen.
+Comparative genomics and Phylogenomics
+
+A L. lactis and L. cremoris pangenome was created without strain LC-LB16 which found to have 6 times more singletons than all the other strains. The number of core, accessory, and unique (singletons) genes in the pangenome is 1247, 3308, and 2323, respectively. The OGs from this pangenome were used in the comparative genomics and differential metatranscriptome analyses as well as in the creation of a pan-metabolic network of L. lactis and L. cremoris for microbial community modeling. We identified all k-mers of length up to 31 bps (k=31) in the genomes of all the strains and unique k-mers in each strain were filtered. For the set of unique k-mers we mapped the metatranscriptomic reads to get their abundance and kept only those above 7 counts. Normalization of the number of counts per strain performed based on the number of unique k-mers identified in each of the strains. From the total pool of 54667 genes 15.8 % was removed as the cluster above 96 cd-hit identity was mixed among the three taxa. still 84.2 % consider significant number to look on L. cremoris and L. lactis differences. For the phylogenetic analysis 22 genomes from the SLAB culture were used alongside with 107 complete genomes of L. lactis and L. cremoris, which were retrieved from NCBI results in total to 129 genomes. CD-HIT on 80% amino acid identity thresholds along with a custom-made R script were used to identify a set of 464 monocore marker gene sets [31, 32]. The trees were constructed using PhyloPhlAn v3.0 [33] using the PhyloPhlAn parameters "-accurate" and "-diversity low", which translate to the usage of a pfasum60 substitution matrix. In addition, multiple sequencing alignments (msa) were performed with muscle and trimming was performed by removing columns with at least one nucleotide appearing above a threshold of 0.99. The final maximum likelihood tree was constructed on the concatenated DNA msa using raxmlHPC, 100 bootstrap and GTRGAMMAI model [34, 35].
+
+Generation and simulation of genome-scale metabolic models
+
+Genome-scale metabolic models were generated for S. thermophilus, Lactococcus LLm1, LLm2, and LC using CarveMe v1.4.1 [36] based on the protein sequences of assembled genomes. First, Prodigal v2.6.3 [37] was used to generate open-reading-frame (ORF) annotated protein sequence files from the corresponding DNA fasta files. Additionally, milk media from a recent kefir publication was used for
+gap-filling during model generation [28]. More specifically, we formulated four biologically relevant variations of the milk composition, including aerobic rich milk, anaerobic rich milk, aerobic depleted milk, and anaerobic depleted milk. The former two media represent the initial composition of milk, while the latter two media represent milk after it has been depleted by fermentation. A version of each of the four species was generated by gap-filling on each of the four milk media variations, as well as without gap-filling, resulting in a total of 20 models. All models were generated using the default CarveMe universal bacterial model template. Note that although gram positive bacteria are generally not modeled with a periplasmic compartment due to its smaller size relative to gram positive bacteria, all models generated with the automated CarveMe tool contain such a compartment. Individual model simulations were carried out for all models on each of the four media variations using the reframed v1.2.1 and cobrapy v.0.20.0 metabolic modeling packages [38]. Community simulation of genome-scale metabolic models was also carried out in the different media variations, using SMETANA v1.2.0 [15]. All associated data, including code used for generating and plotting results, is available on GitHub.
+
+Inoculations in milk
+
+All milk fermentations were carried out in organic low fat (1.5%) milk from the organic dairy NATUR-MÆLK, Tinglev, Denmark. which consists of skim milk powder at a level of dry matter of 9.5% (w/v) reconstituted in distilled water and pasteurized at 99°C for 30 min, followed by cooling to 30°C (Sørensen et al., 2016). Starting material for all inoculations of were concentrated F-DVS® (Direct Vat Set) strains and cultures for direct inoculation into. All cultures were inoculated 0,02% with F-DVS material by firstly transfer 2.0 g F-DVS material to 200 ml cold milk. After mixing, 4 ml were transferred to 200 ml pre-warmed milk (32°C) for the final fermentation. The addition percentages of the different strains and cultures for the 6 different culture blends is given in Supplementary Table 15.
+
+Milk acidifications and sampling
+
+After inoculation into bottles with milk placed at 32°C in a waterbath, pH measurement was started and acidification was followed using a CINAC system (Corrieu et al., 1988). After 59 min, the temperature was raised to 38°C and at 183 min, the temperature was reduced to the final temperature, 35°C, for 20
+hours. The 6 different culture blends were each run in triplicate. After around 5 hours of acidification during the exponential growth phase when reaching pH 5, 1 g fermented milk from each fermentation and 1 g unfermented milk were sampled and added 200 \( \mu l \) 4 N H2SO4 to prepare for analysis for organic acids, Carbohydrates and Volatiles. In addition 1 g was samples for mRNA extraction.
+
+Carbohydrates, acids and volatiles in milk
+
+For the preparation of fermented milk samples used for carbohydrate and small organic acid analysis, 1 g or 1 mL of sample was quenched with 200 \( \mu L \) 4N sulfuric acid (H2SO4) in a 7 mL glass tube, mixed and stored at -20 \( ^\circ \)C until analysis. The analytes for carbohydrate analysis were extracted from the sample and proteins get deproteinated and precipitated by treatment with an aqueous perchloric acid (PCA) solution. The samples get further diluted to fit into the dynamic range of the quantification. Arabinose is added as an internal standard. The diluted samples are analyzed on a Dionex ICS-3000 system (Thermo Fischer Scientific, Waltham (MA), USA) using an analytical anion-exchange column and a pulsed amperometric detector (PAD). For quantification a one-point calibration curve is used. Concentrations are calculated based on the chromatographic peak heights after normalizing to the internal standard (arabinose). The analytes for analysis of small organic acids were extracted from the sample and proteins get deproteinated and precipitated adding an aqueous PCA solution. The samples get further diluted to fit into the dynamic range of the quantification. Adipic acid is added as an internal standard. The diluted samples are analyzed on a Dionex ICS-3000 or ICS-5000 system (Thermo Fischer Scientific, Waltham, MA, USA) using an analytical ion exclusion column and a suppressed conductivity detector (SCD). For quantification, an 8-point calibration curve is used. Concentrations are calculated based on the chromatographic peak heights after normalizing to the internal standard (adipic acid).
+
+For the preparation of the volatile organic compounds (VOC’s) samples, 1 g or 1 mL of each sample was transferred to a headspace vial (20 mL) with 200 \( \mu L \) of 4N H2SO4 and sealed with teflon-lined aluminium caps. The samples were then identified using a static head space sampler connected to a Gas Chromatograph with Flame Ionization Detector (GC-FID) (Perkin Elmer, MA, USA) and equipped with a HP-FFAP column. The identification of VOC´s was based on retention time in comparison with that of standards. The injector and detector were maintained at 180 \( ^\circ \)C and 220 \( ^\circ \)C, respectively. The oven
+was initially heated to 60 °C and held for 2 min, then increased to 230 °C and held for 0.5 min. The calculation of the concentration of each compound in each sample was based on the peak height divided by the response factor. The response factor is established suing standard solutions by the quotient of the peak height divided by the known sample concentration. All of the chemicals and analytes used for the standard solutions were provided by Sigma–Aldrich, Munich, Germany. Overall with these approaches, we measured a total of 43 metabolites comprising five carbohydrates, eight organic acids and 30 important flavor compounds.
+
+RNA extraction, sequencing and analysis
+
+Total RNA was extracted using the RNeasy Mini kit (Qiagen 74104) and rRNA was depleted with the NEBNext rRNA Depletion Kit (Bacteria). The sequencing library was then prepared using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina and the libraries were sequenced by EMBL Genomic Core Facility using the Illumina NextSeq 500 system, read length 150bp. The RNA-reads are preprocessed and filtered based on minimum quality and length using the NGLess pipeline and substrim. Given a read it finds the longest substring, such that all bases are of at least the given quality. The parameters minimum quality and minimum length were set to 25 and 100, respectively. The filtered reads are mapped to the genomes of all strains present in the Culture using bbmap – if one read has multiple hits random distribution is used (selecting one top-scoring site randomly). The total number of reads ranges from 4,855,868 to 18,812,096 in samples 4B and 1B, respectively, and the percentage of mapped reads is in the range of 93-95% (Supplementary Table 5). htseq-count was used to count the number of reads per gene. The resulting counts grouped per orthology group (custom python script) based on a previously created pangenome created for the pan-metabolic network. Normalization was performed and differential gene expression with DESeq2 and SARTools in R [39, 40]. Further differentially expressed genes, Lactococcus pan-metabolic network information and KEGG pathways were based on custom analysis [41, 21]. The data was normalized using DESeq2 to correct systematic technical biases and make it possible to compare read counts across samples. The median scaling factor for each sample was used. This is done for the data containing all genes. RegPrecise database was use for validation of key functional regulation [42].
+Acknowledgments
+
+The authors wish to thank Jannik Vindeloev and Ana Rute Neves for helpful discussions. Karsten Hellmuth, Eric Johansen and Mads Bennedsen for their help with project management and funding acquisition. Also, Kosai Al-Nakeeb, Anna Koza, Jacbo Bælum, and Martin Abel-Kistrup for there significant contribution at genome sequencing and assembly. Lisandra Zepeda and Ida Bærholm Schnell for their help during cheese sampling. Finally, Gunnar Øregaard for providing acidification data for individual strains. The research of CM was supported by a Grand Solution grant from Innovation Fund Denmark (grant no. 6150-00033B), The FoodTranscriptomics project and by the Dutch Research Council, as part of the MiCRop Consortium (NWO/OCW grant no. 024.004.014)
+
+Data Availability Statement
+
+Processed data, including metabolomics, metatranscriptomics count tables as well as the code to reproduce the results are available in https://github.com/Chrats-Melkonian/mi_cheese. Genomes and raw metatranscriptomics data will become available before publication.
+
+Author contributions statement
+
+CM, organizing ideas, writing, comparative genomics, phylogenomics, transcriptomics and bioinformatics analyses. FZ, genome-scale metabolic modeling, community simulations and writing. IK, initial transcriptomics, bioinformatics analyses and genome-scale metabolic modeling. MJ and KSr, carried out the experiments and drafted the manuscript part relating to sections “Inoculations in milk” and “Milk acidifications and sampling”. SB, mRNA extraction of the milk fermentation samples for mRNA sequencing. DM, analysed initial metatranscriptomics data, performed model reconstruction and community simulations. LtA, participated in conceptualization of the study, design and supervise the year-long cheddar-making experiment. KrP, participated in conceptualization of the study, funding acquisition, supervision of modeling and critically revising the manuscript. AZ, participated in conceptualization of the study, funding acquisition, supervision of experimental design, data analysis and bioinformatics, and
+Declaration of competing interest
+
+CM, IK, MJ, KSr, LtA and AAZ are present or previous employees at Chr. Hansen A/S, a global supplier microbial cultures for food fermentation. The authors’ views presented in this study, however, are solely based on scientific grounds and do not reflect the commercial interest of their employer.
+
+References
+
+1. Smith J, Yeluripati J, Smith P, Nayak DR. Potential yield challenges to scale-up of zero budget natural farming. Nature Sustainability. 2020 Mar;3(3):247–252. Available from: https://doi.org/10.1038/s41893-019-0469-x.
+
+2. Wastyk HC, Fragiadakis GK, Perelman D, Dahan D, Merrill BD, Yu FB, et al. Gut-microbiota-targeted diets modulate human immune status. Cell. 2021;184(16):4137–4153. 4137. Available from: https://doi.org/10.1016/j.cell.2021.06.019.
+
+3. Blaya J, Barzideh Z, LaPointe G. Symposium review: Interaction of starter cultures and nonstarter lactic acid bacteria in the cheese environment. Journal of Dairy Science. 2018 Apr;101(4):3611–3629. Available from: https://doi.org/10.3168/jds.2017-13345.
+
+4. Song AA, In LLA, Lim SHE, Rahim RA. Erratum to: A review on Lactococcus lactis: from food to factory. Microbial Cell Factories. 2017 Aug;16(1). Available from: https://doi.org/10.1186/s12934-017-0754-1.
+
+5. Markakiou S, Gaspar P, Johansen E, Zeidan AA, Neves AR. Harnessing the metabolic potential of Streptococcus thermophilus for new biotechnological applications. Current Opinion in Biotechnology. 2020 Feb;61:142–152. Available from: https://doi.org/10.1016/j.copbio.2019.12.019.
+
+6. Rau MH, Gaspar P, Jensen ML, Geppel A, Neves AR, Zeidan AA. Genome-Scale Metabolic Modeling Combined with Transcriptome Profiling Provides Mechanistic Understanding of Streptococcus ther-
+mophilus CH8 Metabolism. Applied and Environmental Microbiology. 2022 Aug;88(16). Available from: https://doi.org/10.1128/aem.00780-22.
+
+7. Pastink MI, Teusink B, Hols P, Visser S, de Vos WM, Hugenholtz J. Genome-Scale Model of Streptococcus thermophilus LMG18311 for Metabolic Comparison of Lactic Acid Bacteria. Applied and Environmental Microbiology. 2009 Jun;75(11):3627–3633. Available from: https://doi.org/10.1128/aem.00138-09.
+
+8. Oliveira AP, Nielsen J, Förster J. Modeling Lactococcus lactis using a genome-scale flux model. BMC Microbiology. 2005 Jun;5(1). Available from: https://doi.org/10.1186/1471-2180-5-39.
+
+9. Champagne CP, Gagnon D, St-Gelais D, Vuillemard JC. Interactions between Lactococcus lactis and Streptococcus thermophilus strains in Cheddar cheese processing conditions. International Dairy Journal. 2009 Nov;19(11):669–674. Available from: https://doi.org/10.1016/j.idairyj.2009.06.002.
+
+10. Duru IC, Laine P, Andreevskaya M, Paulin L, Kananen S, Tynkkynen S, et al. Metagenomic and metatranscriptomic analysis of the microbial community in Swiss-type Maasdam cheese during ripening. International Journal of Food Microbiology. 2018 Sep;281:10–22. Available from: https://doi.org/10.1016/j.ijfoodmicro.2018.05.017.
+
+11. Salazar JK, Carstens CK, Ramachandran P, Shazer AG, Narula SS, Reed E, et al. Metagenomics of pasteurized and unpasteurized gouda cheese using targeted 16S rDNA sequencing. BMC Microbiology. 2018 Nov;18(1). Available from: https://doi.org/10.1186/s12866-018-1323-4.
+
+12. Walsh AM, Macori G, Kilcawley KN, Cotter PD. Meta-analysis of cheese microbiomes highlights contributions to multiple aspects of quality. Nature Food. 2020 Aug;1(8):500–510. Available from: https://doi.org/10.1038/s43016-020-0129-3.
+
+13. Hutkins R, Halambeck SM, Morris HA. Use of Galactose-Fermenting Streptococcus thermophilus in the Manufacture of Swiss, Mozzarella, and Short-Method Cheddar Cheese. Journal of Dairy Science. 1986 Jan;69(1):1–8. Available from: https://doi.org/10.3168/jds.s0022-0302(86)80361-7.
+14. Proust L, Haudebourg E, Sourabié A, Pedersen M, Besançon I, Monnet V, et al. Multi-omics Approach Reveals How Yeast Extract Peptides Shape Streptococcus thermophilus Metabolism. Applied and Environmental Microbiology. 2020 oct;86(22). Available from: https://doi.org/10.1128%2Faem.01446–20.
+
+15. Zelezniak A, Andrejev S, Ponomarova O, Mende DR, Bork P, Patil KR. Metabolic dependencies drive species co-occurrence in diverse microbial communities. Proceedings of the National Academy of Sciences. 2015 May;112(20):6449–6454. Available from: https://doi.org/10.1073/pnas.1421834112.
+
+16. Nugroho ADW, Kleerebezem M, Bachmann H. Growth, dormancy and lysis: the complex relation of starter culture physiology and cheese flavour formation. Current Opinion in Food Science. 2021;39:22–30.
+
+17. Smid EJ, Kleerebezem M. Production of Aroma Compounds in Lactic Fermentations. Annual Review of Food Science and Technology. 2014 feb;5(1):313–326. Available from: https://doi.org/10.1146%2Fannurev-food-030713-092339.
+
+18. Calbert HE, Price WV. A Study of the Diacetyl in Cheese. I. Diacetyl Content and Flavor of Cheddar Cheese. Journal of Dairy Science. 1949 jun;32(6):515–520. Available from: https://doi.org/10.3168%2Fjds.s0022-0302%2849%2992073-1.
+
+19. Bäumler AJ, Sperandio V. Interactions between the microbiota and pathogenic bacteria in the gut. Nature. 2016 Jul;535(7610):85–93. Available from: https://doi.org/10.1038/nature18849.
+
+20. Lara EG, van der Windt I, Molenaar D, de Vos MGJ, Melkonian C. Using Functional Annotations to Study Pairwise Interactions in Urinary Tract Infection Communities. Genes. 2021 aug;12(8):1221. Available from: https://doi.org/10.3390%2Fgenes12081221.
+
+21. Melkonian C, Fillinger L, Atashgahi S, da Rocha UN, Kuiper E, Olivier B, et al. High biodiversity in a benzene-degrading nitrate-reducing culture is sustained by a few primary consumers. Communications Biology. 2021 may;4(1). Available from: https://doi.org/10.1038%2Fs42003-021-01948-y.
+22. Melkonian C, Seidl MF, van der Hooft JJJ, de Vos MGJ. Metabolic interactions shape a community's phenotype. Trends in Microbiology. 2022 jul;30(7):609–611. Available from: https://doi.org/10.1016%2Fj.tim.2022.05.001.
+
+23. Machado D, Maistrenko OM, Andrejev S, Kim Y, Bork P, Patil KR, et al. Polarization of microbial communities between competitive and cooperative metabolism. Nature Ecology & Evolution. 2021 Jan;5(2):195–203. Available from: https://doi.org/10.1038/s41559-020-01353-4.
+
+24. Irlinger F, Mounier J. Microbial interactions in cheese: implications for cheese quality and safety. Current Opinion in Biotechnology. 2009 Apr;20(2):142–148. Available from: https://doi.org/10.1016/j.copbio.2009.02.016.
+
+25. Jiang N, Wu R, Wu C, Wang R, Wu J, Shi H. Multi-omics approaches to elucidate the role of interactions between microbial communities in cheese flavor and quality. Food Reviews International. 2022 Apr;1–13. Available from: https://doi.org/10.1080/87559129.2022.2070199.
+
+26. Parente E, Zotta T, Ricciardi A. A review of methods for the inference and experimental confirmation of microbial association networks in cheese. International Journal of Food Microbiology. 2022 May;368:109618. Available from: https://doi.org/10.1016/j.ijfoodmicro.2022.109618.
+
+27. Ponomarova O, Gabrielli N, Sévin DC, Mülleder M, Zirngibl K, Bulyha K, et al. Yeast Creates a Niche for Symbiotic Lactic Acid Bacteria through Nitrogen Overflow. Cell Systems. 2017 Oct;5(4):345–357.e6. Available from: https://doi.org/10.1016/j.cels.2017.09.002.
+
+28. Blasche S, Kim Y, Mars RA, Machado D, Maansson M, Kafkia E, et al. Metabolic cooperation and spatiotemporal niche partitioning in a kefir microbial community. Nature microbiology. 2021;6(2):196–208.
+
+29. Goyal A, Bittleston LS, Leventhal GE, Lu L, Cordero OX. Interactions between strains govern the eco-evolutionary dynamics of microbial communities. eLife. 2022 feb;11. Available from: https://doi.org/10.7554%2Felife.74987.
+30. Oren A, Garrity GM. Notification that new names of prokaryotes, new combinations and new taxonomic opinions have appeared in volume 71, part 3 of the IJSEM. International Journal of Systematic and Evolutionary Microbiology. 2021 Jun;71(6). Available from: https://doi.org/10.1099/ijsem.0.004812.
+
+31. Fu L, Niu B, Zhu Z, Wu S, Li W. CD-HIT: accelerated for clustering the next-generation sequencing data. Bioinformatics. 2012 oct;28(23):3150–3152. Available from: https://doi.org/10.1093%2Fbioinformatics%2Fbts5565.
+
+32. R Core Team. R: A language and environment for statistical computing. Vienna, Austria; 2020. Available from: https://www.R-project.org/.
+
+33. Asnicar F, Thomas AM, Beghini F, Mengoni C, Manara S, Manghi P, et al. Precise phylogenetic analysis of microbial isolates and genomes from metagenomes using PhyloPhlAn 3.0. Nature Communications. 2020 may;11(1). Available from: https://doi.org/10.1038%2Fs41467-020-16366-7.
+
+34. Edgar RC. MUSCLE: multiple sequence alignment with high accuracy and high throughput. Nucleic Acids Research. 2004 mar;32(5):1792–1797. Available from: https://doi.org/10.1093%2Fnar%2Fgkh340.
+
+35. Stamatakis A. RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylogenies. Bioinformatics. 2014 jan;30(9):1312–1313. Available from: https://doi.org/10.1093%2Fbioinformatics%2Fbttu033.
+
+36. Machado D, Andrejev S, Tramontano M, Patil KR. Fast automated reconstruction of genome-scale metabolic models for microbial species and communities. Nucleic Acids Research. 2018 06;46(15):7542–7553. Available from: https://doi.org/10.1093/nar/gky537.
+
+37. Hyatt D, Chen GL, LoCascio PF, Land ML, Larimer FW, Hauser LJ. Prodigal: prokaryotic gene recognition and translation initiation site identification. BMC bioinformatics. 2010;11(1):1–11.
+
+38. Ebrahim A, Lerman JA, Palsson BO, Hyduke DR. COBRAPy: constraints-based reconstruction and analysis for python. BMC systems biology. 2013;7(1):1–6.
+39. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology. 2014 dec;15(12). Available from: https://doi.org/10.1186%2Fs13059-014-0550-8.
+
+40. Varet H, Brillet-Guéguen L, Coppée JY, Dillies MA. SARTools: A DESeq2- and EdgeR-Based R Pipeline for Comprehensive Differential Analysis of RNA-Seq Data. PLOS ONE. 2016 jun;11(6):e0157022. Available from: https://doi.org/10.1371%2Fjournal.pone.0157022.
+
+41. Melkonian C, Gottstein W, Blasche S, Kim Y, Abel-Kistrup M, Swiegers H, et al. Finding Functional Differences Between Species in a Microbial Community: Case Studies in Wine Fermentation and Kefir Culture. Frontiers in Microbiology. 2019 jun;10. Available from: https://doi.org/10.3389%2Ffmicb.2019.01347.
+
+42. Novichkov PS, Kazakov AE, Ravcheev DA, Leyn SA, Kovaleva GY, Sutormin RA, et al. Reg-Precise 3.0 – A resource for genome-scale exploration of transcriptional regulation in bacteria. BMC Genomics. 2013;14(1):745. Available from: https://doi.org/10.1186/1471-2164-14-745.
+
+figures and tables
+Figure 1. Schematic representation of the experimental design and methods (a) The 1 year long cheddar-making experiment. Microbial population dynamics were quantified during cheese ripening using a selective method of viable cell counting, which discriminates between thermophilic cocci, mesophilic cocci and non-starter lactic acid bacteria (NSLAB). Additionally, metabolic changes in the cheeses were measured using several targeted analytical chemistry approaches. (b) Schematic representation of the controlled milk experiment in the laboratory as well as of the methods (1-5) used in the controlled milk experiment. This second experiment involves the removal of additional strains, namely the individual exclusion of three major *Lactococcus* strains. The numbers indicate the different type of data and analysis as follows: 1. metatranscriptomics, 2. metabolomics, 3. genomics, 4. phylogenomics and 5. genomes-scale metabolic models (GEMs) and community simulations. Our integrative systems biology approach combined: i) the analysis of the SLAB community’s genomes, ii) the generation and simulation of their respective GEMs, iii) the analysis of the metatranscriptomes across the different strain removal conditions and iv) the quantification of key metabolites.
+Figure 2. S. thermophilus benefit the growth of lactococci community and influence the final flavor of cheese. (a-d) Microbial population dynamics during cheese ripening. Relationship between CFUs and time (months) for (a) Mesophilic, (c) Thermophilic cocci, and (d) non-starter lactic acid bacteria (NSLAB). (b) Presents the boxplot comparison of the different condition on Mesophilic cocci at 12 months. Note, significant reduction only on Mesophilic cocci is observed when S. thermophilus is absent. Condition All corresponds to whole culture which is composed from 24 number of strains. HP corresponds to an alternative method (hand packed) of the whole culture inoculation. -LB corresponds to the removal of a L. lactis blend population. -ST corresponds to the removal of the S. thermophilus strain. (e-h) Metabolome dynamics during cheese ripening. (e) 2-dimensional representation with the usage of UMAP of all samples based on the metabolomics measurements (Acids, Carbohydrates and Peptides). The colour indicates the time when the sample was taken and the shape indicates the four different conditions. Note, the stronger change is observed between 2 week and 3 months while the removal of ST (cross symbol) has a strong effect on the later time of cheese ripening. (f) Relative change of the metabolites on different time intervals including all the conditions, both colour and shape indicates the class of the metabolite. Note, the higher relative change of peptides followed by acids at the interval between 2 week and 3 months, later acids exhibit higher changes. The three measured sugars (Glucose, Lactose and Galactose) are excluded from this panel. (g) Selection of the six most discriminative metabolites out of 50 for the four conditions, colour indicate the condition while the right side shape the class of the metabolite. Note, the majority of peptides concentration are significant different when S. thermophilus is not present as well as with galactose, lactose and lactic acid. (h) Galactose concentration over time highlights the absence of galactose when S. thermophilus is not present. The galactose concentration remains steady from the start till the 9 month of cheese ripening and then shows signs of decline.
+Figure 3. The removal of different Lactococcus strains has a distinct effect on gene expression in S. thermophilus. (a-e) relationship of the Lactococcus phylogeny with the S. thermophilus transcription change on different dropout conditions. (a) Lactococcus phylogenetic tree separates the L. cremoris and L. lactis strains. The two clades are colored with transparent yellow and grey, respectively. The colors and shape of the tree tips indicate the strains presents in the SLAB culture. Tree empty tips correspond to the obtained complete NCBI genomes of Lactococcus. Additional information is presented in the outer layers of the tree. The first layer presents the unique k-mer content based on the SLAB culture’s genomes (purple-scale color) and indicates higher number for the strains who have no close relative within the culture. The second layer presents the proxy of temperature stress tolerance based on Pearson Correlation Coefficient (PCC) between individual acidification curve of 30 and 40 degree °C (red-scale color). High values correspond to high temperature tolerance. The clade L. cremoris shows lower temperature tolerance, with few exceptions including the main LC strain (more elaborate data are presented in Supplementary Fig. 5). The third layer presents the percentile of transcribed sigletons (green-scale color) as well as the total number of the strains sigletons (outer bars). (b-e) Volcano plots presents the S. thermophilus transcription between the whole community (All) and the different Lactococcus dropout conditions, respectively. The colors corresponds to the 4 Lactococcus components. The S. thermophilus transcription observed to change more when LLm1 was left-out followed by LB, with 291 and 182 number of total significant up/down regulated genes, respectively. A smaller effect of 21 genes was observed when LC was left-out and only 1 gene when LLm2 was left out.
+Figure 4. S. thermophilus provides nitrogen in form of amino acids to L. lactis community, which is necessary for de novo nucleotide biosynthesis. (a) Summary of metabolic modeling analysis including individual model and community-based simulations. (b) Selected exchange fluxes predicted across individual flux balance analysis simulations carried in milk media variations for each set of metabolic models, in particular fermentation products, amino acids, and carbon sources highlighting differences in metabolic strategies across species. (c) Alluvial diagram showing predicted metabolic exchanges from community simulations in milk media variants, highlighting the fact the amino acid valine is strongly predicted across all simulation conditions where metabolic cross-talk is expected. Refer to Supplementary Figure 6 for more details on the metabolic models and simulation. (d-i) Transcriptomics profiles between All and -ST (S. thermophilus is left out) conditions of key functions for all the strains in SLAB culture. (d-f) Transcriptomics boxplots of branched chain amino acids (BCAA) aminotransferase, BCAA transport system 2 carrier protein and ammonium transporter (T) supports the predictions of metabolic exchanges. Note, (e) the higher transcription profile of S. thermophilus on BCAA trasport system as well as the high transcription profile of S. thermophilus and the increase of LC transcription profile when S. thermophilus was left out. (g-j) Up-regulation of Lactococcus glutamine and nucleotide metabolism when S. thermophilus is left out. (g-i) Boxplots present the activities of nitrogen regulatory protein P-II, glutamine synthetase (glnA) and related transcriptional regulator (TR) (GlnR) for all community members. Note, the up-regulation of all the enzymes derived from Lactococcus strains when S. thermophilus is left out. (j) Metatranscriptomics represents the increased activity of Lactococcus reactions, which are incorporated into Escher maps. The pattern shows a coordinate up-regulation (with green) of the enzymes towards guanine biosynthesis (and uridine). (k) Schematic compilation of the compounds S. thermophilus may provide to Lactococcus community as well as selected transcriptional changes of the Lactococcus community when S. thermophilus was left out. The color and direction of the arrows represent the up- and down- regulation with green/up and red/down, respectively.
+Figure 5. Cheese flavor compounds are strongly influenced by the interactions within Lactococcus community The measured metabolome from the controlled milk experiment in response to stain removal conditions (a-d). (a) The PCA of key carbohydrates, acids and cheese flavor compounds. The strongest response observed from the removal of L. cremoris followed by coordinated response of LLm1, LB and S. thermophilus. Removal of LLm2 has no effect. The arrows represents the top 5 compounds (2,3-Pentanedione, Hexanal, Diacetyl, Acetoin and Ethyl acetate) responsible for the observed changes. Panels (b-c) represents the signal-to-noise ratio (S/N) of diacetyl and acetoin, respectively. Regarding those two compounds, absence of S. thermophilus from the community led to significant lower accumulation, while absence of L. cremoris leads to significant higher accumulation. The black horizontal line indicate the average value of the two compounds at milk prior to acidification. Paired t-test for each condition against the whole community (All) were performed with the symbols indicating the statistical significance results: ns: p > 0.05, *: p <= 0.05 and **: p <= 0.01. (d) Schematic compilation of the compounds that changed between the whole SLAB culture and when L. cremoris was left out (for detailed boxplots see supplementary). The color represent the removal conditions; yellow indicates the removal of L. cremoris while red represents the whole SLAB culture. The upwards arrow indicates the increase of the compounds relative to the compounds concentration in milk. The X symbol indicates the lack of production of the compounds in the respective condition. (e) Simplified metabolic escher maps from citrate to α-ketoglutarate as well as to diacetyl through pyruvate and (S)-2-Acetolactate. Metabolites and reactions are indicated with black and grey color in the graph, respectively. Full colored circles and open colored circles indicate the high and low transcription per strain, respectively. Lack of circle visualization indicates the lack of the corresponding orthologous group genes. Note, the transcriptional fluxes from citrate-sodium symporter towards diacetyl, which are present to all members in Lactococcus community. In addition, LC and LLm1 manifest higher transcription among the Lactococcus community on the genes related to acetolactate decarboxylase. Only LC strain manifests transcriptional flux through diacetyl reductase and butanediol dehydrogenase as well as in reactions towards α-ketoglutarate. (f) Predicted fluxes based on metabolic modeling across the different simulated media.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• MICheeseSUPP.pdf

1003c99ade06e9dbb20a5837a88e4801c0fa9dce9184fba16a6a52a8b20ccb57/preprint/preprint.md

@@ -0,0 +1,375 @@
+Maternal N-acetyltransferase 10 (NAT10) orchestrates oocyte meiotic cell-cycle progression and maturation in mice
+
+Jianqiang Bao (jqbao@ustc.edu.cn)
+The First Affiliated Hospital of University of Science and Technology of China https://orcid.org/0000-0003-1248-2687
+
+Xue Jiang
+The First Affiliated Hospital of University of Science and Technology of China
+
+Yu Cheng
+University of Science and Technology of China
+
+Yuzhang Zhu
+University of Science and Technology of China
+
+Caoling Xu
+The First Affiliated Hospital of University of Science and Technology of China
+
+Qiaodan Li
+University of Science and Technology of China
+
+Xuemei Xing
+The First Affiliated Hospital of University of Science and Technology of China
+
+Wenqing Li
+The First Affiliated Hospital of University of Science and Technology of China
+
+Jiaqi Zou
+The First Affiliated Hospital of University of Science and Technology of China
+
+Lan Meng
+The First Affiliated Hospital of University of Science and Technology of China
+
+Muhammad Azhar
+The First Affiliated Hospital of University of Science and Technology of China
+
+Yuzhu Cao
+The First Affiliated Hospital of University of Science and Technology of China
+
+Xianhong Tong
+The First Affiliated Hospital of University of Science and Technology of China
+
+Weibing Qin
+Xiaoli Zhu
+The First Affiliated Hospital of University of Science and Technology of China
+Article
+
+Keywords: Nat10, meiosis, oocyte growth, oocyte maturation, PolyA Tail length assay (PAT)
+
+Posted Date: September 15th, 2022
+
+DOI: https://doi.org/10.21203/rs.3.rs-2033653/v1
+
+License: This work is licensed under a Creative Commons Attribution 4.0 International License.
+Read Full License
+Maternal N-acetyltransferase 10 (NAT10) orchestrates oocyte meiotic cell-cycle progression and maturation in mice
+
+Xue Jiang1,†, Yu Cheng2,†, Yuzhang Zhu3,†, Caoling Xu1,†, Qiaodan Li4,†, Xuemei Xing5, Wenqing Li1, Jiaqi Zou1, Lan Meng1, Muhammad Azhar1, Yuzhu Cao5, Xianhong Tong5, Weibing Qin6*, Xiaoli Zhu5*, Jianqiang Bao1,7*
+
+1The First Affiliated Hospital of USTC, School of Basic Medical Sciences, Division of Life Sciences and Medicine, University of Science and Technology of China (USTC), Hefei, Anhui 230001, China; 2School of Information Science and Technology ,University of Science and Technology of China (USTC), Hefei, Anhui 230001, China; 3Division of Life Sciences and Medicine, University of Science and Technology of China (USTC), Hefei, Anhui 230001, China, 4Laboratory animal centre, University of Science and Technology of China (USTC), Hefei, Anhui 230001, China; 5 Reproductive and Genetic Hospital, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China (USTC), Hefei, Anhui 230001, China; 6NHC Key Laboratory of Male Reproduction and Genetics, Guangdong Provincial Reproductive Science Institute (Guangdong Provincial Fertility Hospital), Guangzhou 510600, China; 7Reproductive and Genetic Hospital, The First Affiliated Hospital of USTC, Hefei National Research Center for Physical Sciences at the Microscale, Biomedical Sciences and Health Laboratory of Anhui Province, School of Basic Medical Sciences, Division of Life Sciences and Medicine, University of Science and Technology of China (USTC), Hefei, Anhui 230001, China.
+
+*To whom correspondence should be addressed. Tel: +86 0551 63606389; Email: jqbao@ustc.edu.cn
+Correspondence may also be addressed to Xiaoli Zhu. Tel: +86 0551 63607625; Email: xiaolizh@ustc.edu.cn
+Correspondence may also be addressed to Weibing Qin. Tel: +86 020 87692825; Email: guardqin@163.com
+†The authors wish it to be known that, in their opinion, the first five authors should be regarded as Joint First Authors.
+
+Running Title: Nat10 is essential for mouse oocyte development
+
+Keywords: Nat10; meiosis; oocyte growth; oocyte maturation; PolyA Tail length assay (PAT)
+Abstract
+In mammals, the production of mature oocytes necessitates rigorous regulation of the discontinuous meiotic cell-cycle progression at both the transcriptional and post-transcriptional levels; however, the factors underlying this sophisticated but explicit process during oocyte development remain largely unclear. Here we characterized the function of N-acetyltransferase 10 (Nat10), which was previously recognized as a “writer” for N4-acetylcytidine (ac4C) deposited on RNA molecules. We generated two germline-specific Nat10 knockout mouse models in the embryonic gonad and in postnatal growing oocytes, and another two tamoxifen-inducible Nat10 deletion models. We provided genetic evidence showing that Nat10 is essential for oocyte meiotic prophase I progression, oocyte growth and maturation in mice. Intriguingly, we discovered that Nat10 is required for sculpting the maternal transcriptome through timely degradation of polyA tail mRNAs during the Maternal-to-Zygotic transition (MZT). Importantly, we developed a novel method that outperformed the conventional methods for examining the polyA tail length (PAT), termed Hairpin Adaptor-polyA tail length (HA-PAT), in terms of the cost, sensitivity and efficiency. In summary, these findings altogether provide solid genetic evidence that unveils the indispensable role of maternal Nat10 in oocyte development, and lay a solid foundation for future mechanistic studies of varied domains of NAT10.
+Introduction
+
+In mammals, the germ cells are distinct from somatic cells in that they entail two successive cell divisions following one round of DNA replication, producing haploid gametes with parentally exchanged, but equal amounts of genetic DNA transmitted to the offspring \( ^1 \). This sexually dimorphic event is achieved through a highly conserved and tightly controlled process, namely, meiosis. In mice, the primordial germ cells (PGCs) migrate and colonize the genital ridge prior to embryonic day 10.5 (E10.5). Subsequently, stimulated by the microenvironment signaling emanating from the surrounding soma, the PGCs are sexually determined to commit to either male (XY) or female (XX) germ cells on E11.5. In the female, the ovary is morphologically distinguishable from the male testis by the lack of cord-like structure on E12.5 \( ^2 \). Unlike the prolonged mitotic arrest of male germ cells in embryonic testis, female germ cells (herein referred to as oogonia) initiate meiosis quickly starting from E13.5, and sequentially undergo stages of meiotic prophase I (leptotene, zygotene and pachytene), but are fully arrested at diplotene stage until pubertal LH signaling \( ^3 \). It is well-known that the topoisomerase SPO11-mediated double-strand breaks (DSBs) are a prerequisite for the timely pairing, synapsis, and recombination between homologous chromosomes. Nonetheless, little is known about how genome-wide DSBs are specifically generated and repaired and which factors are involved in this extended stage of meiotic prophase I \( ^4 \).
+
+After birth, individual oocytes arrested at the diplotene of prophase I are encircled by a flattened layer of granulosa cells, which together are called primordial follicles. A clutch of primordial follicles is periodically recruited to the growth phase, which develops sequentially through stages of primary, secondary, early antral, antral, and large antral follicles, ensued by the ovulation upon LH surge \( ^2 \). This whole process is exquisitely coordinated, as evidenced by massive gene transcription in conjugation with selective degradation of specific transcripts, which together build up the maternal transcriptome. The growing oocytes acquire the meiotic competence, *i.e.*, the capability of resuming meiosis to MII stage, during the transition from the secondary to early antral follicles, while the acquisition of developmental competence occurs in the late stage of antral follicles that gives rise to the matured MII oocytes with the ability to be fertilized and develop to term (referred to as oocyte maturation) \( ^2, 5, 6, 7 \). Since the zygote is transcriptionally silent prior to zygotic gene activation, all the initial developmental events during early embryonic reprogramming are dependent on the maternal RNA transcriptome inherited from the oocyte. Remarkably, the maternal transcriptome is particularly rich in mRNA, which is stably stored and occupies up to 20% of the total RNA in a fully grown GV oocyte, in contrast to the average ~2% mRNA in somatic cells \( ^8, 9 \). The transcriptional activity peaks in the early growing oocytes, but gradually decreases and is considered silent in the fully grown oocyte. Interestingly, this process is accompanied by DNA configuration transition from the less condensed, non-surrounded nucleolus (NSN) to the surrounded nucleolus (SN) state, wherein fully condensed chromatin DNA encompasses the nucleolus \( ^{10} \). Of note, compared with SN oocytes, NSN oocytes exhibit higher activity of transcription and lower developmental competence, with most zygotes arresting at the 2-cell stage. Moreover, genome-wide RNA transcriptome analyses revealed that SN oocytes display different gene expression profiles and metabolic pathway enrichment compared with NSN oocytes \( ^{10, 11} \). These studies revealed the significance of maternal transcriptome integrity in coordinating the timely oocyte growth and
+maturation, and implicated its pathogenic roles, when disrupted, underling female infertility. Nevertheless, which factors define the maternal transcriptome and how they coordinate with each other through continuous stages of folliculogenesis remain poorly understood.
+
+Our interest in exploring the role of Nat10 in oocyte development was initially piqued for two reasons. First, during oocyte maturation, there is a profound polyA-shortening mechanism-mediated mRNA decay that dramatically reshapes the maternal transcriptome. How particular mRNAs are selected for destruction is not well-understood \(^{6, 12}\). Nonetheless, this event must occur at the post-transcriptional level since the DNA transcription is progressively shut down concurrent with the oocyte growth. Previous studies have shown that multiple RNA modifications, such as m6A, m5C and m1A, are involved in post-transcriptional RNA metabolism, including alternative splicing, mRNA decay, and mRNA translation \(^{13, 14}\). Nat10 is highly expressed in the mouse ovary and is thus far the only known “writer” for epi-transcriptomic modification- N4-acetylcytidine (ac4C). In HeLa cells, it has been shown to enhance the stability and protein translation for mRNAs \(^{15}\). Second, as described above, oocyte development is under tight, spatiotemporally specific regulation through discontinuous meiotic cell-cycle progression – rapid progression at early meiotic prophase I in the embryonic gonad, lengthened late prophase I arrest at diplotene during prepubertal development, and quick cytoplasmic and nuclear maturation during the GV-MII transition. Compelling studies have shown evidence related to Nat10's function in cell-cycle control in somatic cells \(^{16, 17, 18}\). Interestingly, specific deletion of *Nat10* in the male germline elicited severe defects resulting in male infertility owing to corrupted meiotic cell-cycle progression in mice \(^{19}\). This evidence altogether is reminiscent of the critical role of Nat10 in female oocyte development.
+
+In this study, we generated two germline-specific *Nat10* knockout (KO) mouse models by *Stra8*- and *Zp3*-driven Cre expression and two additional tamoxifen (TMX)-inducible *Nat10* KO mouse models using Ddx4-CreERT2 and Ubc-CreERT2. We revealed that Nat10 plays a profound role and is indispensable for oocyte meiotic cell-cycle progression in embryonic gonads and in postnatal oocyte growth and maturation. Importantly, we designed and optimized a novel method, termed HA-PAT, that outperformed previous conventional methods adopted for polyA tail length examination in single oocytes. Taking advantage of this approach, we uncovered that Nat10-mediated polyA tail shortening is a critical mechanism that defines the oocyte maternal transcriptome. Together, we provided compelling genetic evidence that validated the essential roles of Nat10 in mouse oocyte development.
+Results
+
+NAT10 is highly expressed and localized to the nucleolus in mouse oocytes
+
+To explore the potential function of Nat10 during oogenesis, we first evaluated the multiple-tissue expression pattern of Nat10 in an array of mouse organs. NAT10 protein was abundantly expressed in the ovary, with the highest levels detected in the thymus, spleen, and testis, which is consistent with a recent study (Fig. 1A) \(^{19}\). We further re-analyzed the published bulk RNA-seq datasets in the growing oocytes and pre-implantation embryos (GSE71434) \(^{22}\). It showed that Nat10 mRNA was dynamically regulated in the postnatal growing follicles and preimplantation embryos, with the highest mRNA levels detected in GV oocytes and the lowest levels in 2-cell embryos (Fig. 1B). In agreement with this finding, quantitative real-time PCR (qPCR) validated the similar expression trend of Nat10 mRNA levels in different stages of oocytes and embryos (Fig. 1C). Next, we performed the fluorescent immunostaining (IF) with a NAT10 antibody in isolated oocytes at various stages as well as in ovary cryosections. As shown in Fig. 1D, NAT10 protein is abundantly present in the nucleus of the GV oocytes. The intensity of NAT10 signal is reduced and dispersed in the nucleus of MI and MII oocytes owing to the breakdown of the nuclear membrane (Fig. 1D). On the basis of the morphology and the number of surrounding granulosa cells, the follicles can be categorized into various stages during postnatal folliculogenesis in mice, including primordial follicle (PrF), primary follicle (PF), secondary follicle (SF), early antral follicle (EAF), and antral follicle (AF) (Fig. 1E) \(^{23}\). Consistent with its mRNA expression trend, IF revealed abundant NAT10 protein being detected in the central nucleus of the growing oocytes at various stages, with the highest intensity in the nucleus center encircled by a layer of highly condensed chromatin (Fig. 1E). This distinctive expression pattern raised the possibility that it might be localized in the nucleolus, as reported by a few previous studies \(^{18, 24, 25}\). To verify this possibility, we co-immunostained NAT10 and a nucleolus-specific marker, nucleophosmin (NPM). As shown in Fig. 1F, NAT10 is well co-localized to the nucleolus with NPM in the oocytes through various stages of folliculogenesis. It is worth noting that NAT10 is also highly expressed in the nucleolus of granulosa cells, particularly in antral follicles, as revealed by its perfect co-localization with NPM (Fig. 1F). This evidence together implies that NAT10 might have important physiological roles in mouse oocyte development *in vivo*.
+
+Pre-meiotic deletion of *Nat10* caused follicular developmental arrest at primary follicles and premature ovarian failure (POF)
+
+Next, to decipher the physiological function of Nat10 *in vivo*, we generated pre-meiotic stage-specific *Nat10* knockout (KO) mice by crossing floxed Nat10 (*Nat10*lox/lox) alleles with Stra8-GFPCre knockin alleles to obtain oocyte conditional Nat10 KO females (*Nat10*lox/-; Stra8-GFPCre, hereafter called Nat10-ScKO) (Fig. S1A-C). The floxed Exons #4 and #5 (E4/5) reside in the DUF1726 domain of NAT10 protein (Fig. S1A). Stra8-GFPCre is specifically activated around embryonic day 12 (E12) in the primordial germ cells (PGCs) prior to meiosis in the female embryonic gonad (Fig. 2A). When crossing these deleter mice with Nat10*lox/lox* mice, E4/5 of the *Nat10* gene were removed, resulting in the frame-shift translation and presumably nonsense-mediated mRNA decay (NMD) (Fig. 2A-D, Fig. S1B and C). Using isolated GV oocytes, both western blotting and qPCR showed that the protein and mRNA levels
+of Nat10 were markedly reduced in Nat10-ScKo oocytes compared with WT oocytes (Fig. 2C and D). Fertility testing showed that female Nat10-ScKo mice were completely sterile when crossed with WT males during 6 months of breeding (Fig. 2E). At 1 month, the size of Nat10-ScKo ovaries was reduced by eightfold compared with that in WT ovaries (Fig. 2F). To trace at which stage the Nat10 abrogation impacted follicle development, we next carried out the histological examination of Hematoxylin&Eosin (HE)-stained sections of paraffin-embedded ovaries during the first wave of postnatal follicle development. This unveiled that the follicles managed to proceed but arrested at the stage of primary follicles before P21 in the Nat10-ScKo females, unlike the WT ovaries where antral follicles were frequently observed (Fig. 2G). After P21, the Nat10-ScKo ovaries progressively lost all the characteristic follicular structure but harbored tubule-like structures filled with homogeneous, immature granulosa cells, which resembled the seminiferous tubules in the male testis (Fig. 2G). All these tubules were devoid of oocytes in the center, presumably owing to the quick degeneration of the Nat10-null oocytes, reminiscent of premature ovarian failure (POF) (Fig. 2G). Indeed, this phenotype of transdifferentiation of ovarian cells to Sertoli-like cells has been previously reported, wherein deletion of Mtor in the primordial oocytes induced the conversion of granulosa cells to Sertoli-like cells, displaying a seminiferous tubule-like testicular structure present in the oocyte-specific Mtor KO ovary 26. To test this possibility, we next designed a panel of granulosa cell- and Sertoli/Leydig cell-specific primers. As shown in Fig. 2H, while the expression levels of granulosa cell-specific markers (Amh, Cyp19a1 and Esr2) were significantly reduced in Nat10-ScKo ovaries compared with WT ovaries, we only observed the mRNA expression levels of two markers, Sox9 in Sertoli cells and Cldn11 in Leydig cells, among a selection of testis-enriched markers (Cyp11b1, Hsd3b6, Gata1, data not shown) were markedly elevated. This evidence implicated the partial conversion of granulosa cells to Sertoli cells upon Nat10 KO. In summary, pre-meiotic ablation of Nat10 in the female gonad led to female infertility due to defective follicular developmental arrest at the primary follicle stage and premature ovarian failure.
+
+Premiotic loss of Nat10 led to oocyte meiotic arrest at pachytene stage owing to disturbed DSB repair
+
+In contrast to the male germline, the female germ cells initiate meiotic prophase I division early following sex determination in the embryonic gonad, and sequentially undergo leptotene, zygotene, and pachytene, but are finally arrested at the diplotene stage perinatally until further hormonal stimulation (Fig. 2A). Nat10 has recently been shown to be essential for meiotic divisions during spermatogenesis in the testis 19, thus we next investigated whether Nat10 is required for female meiotic prophase I progression in vivo. Co-staining of the nuclear chromosome spreads by SYCP1 and SYCP3 markers showed the accumulation of aberrant pachytene-like cells with partially synapsed homologous chromosomes in the perinatal Nat10-ScKo ovaries (Fig. 3A). Statistical comparison validated the elevated percentage of pachytene/pachytene-like cells and, as a result, the decreased proportion of oocytes at the diplotene stage in Nat10-ScKo ovaries (Fig. 3A and B), suggesting the meiotic arrest of Nat10-ScKo oocytes at the pachytene stage.
+Premature oocyte death during the first wave of folliculogenesis, as described above (Fig. 2G), is most often a consequence of a self-surveillance mechanism for the host to safeguard genome integrity against unsynapsed chromosomes or excessive double-strand DNA breaks (DSBs) \(^{27, 28, 29}\). There was a much higher occurrence of aberrant chromosomes that were not fully synapsed, as described above, in the perinatal Nat10-ScKO ovaries (Fig. 3A and B), we thus next assessed the causative factors that account for the premature oocyte loss upon *Nat10* KO. IF staining by \( \gamma \)H2AX revealed the elevated DSB signals in oocytes at the pachytene stage, but not at the diplotene stage, in embryonic Nat10-ScKO ovaries (Fig. 3C and D), suggesting defective DSB repair in Nat10-null pachytene oocytes \(^{30}\). Further examination by staining with RPA2, a marker that exclusively labels unrepaired DSBs, unveiled that more RPA2 foci were present in the Nat10-ScKO oocytes at pachytene stage, rather than at diplotene stage (Fig. 3E and F), which presumably correspond to oocytes with elevated \( \gamma \)H2AX staining resulting from DSB repair deficiency in Nat10-ScKO ovaries \(^{30}\). Together, this evidence suggests that Nat10 is essential for meiotic prophase I progression in the female embryonic gonad.
+
+**Nat10 is essential for the chromatin configuration NSN-SN transition in growing oocytes during postnatal folliculogenesis**
+
+At birth, the oocytes arrest at the diplotene stage in meiotic prophase I, and each is surrounded by a single layer of flattened granulosa cells, which together constitute the primordial follicle pool (Fig. 2A). Upon pubertal stimulation by FSH and LH, the follicles sequentially enter the growth and maturation stages (Fig. 2A). We thus next evaluated whether Nat10 is required for oocyte growth through generation of a *Nat10*-specific deletion mouse model in growing oocytes by crossing the Nat10lox/lox alleles with female Zp3-Cre (henceforth termed Nat10-ZcKO), which is specifically activated in the oocytes of primary follicles (Fig. 2A, Fig. 4A)\(^{31}\). IF staining using P21 ovaries and isolated GV oocytes showed that Nat10 protein is specifically eliminated from the oocytes, but not in the granulosa cells (Fig. 4B and C, Fig. S2A and B). During a half-year fertility testing, Nat10-ZcKO females were completely sterile, suggesting Nat10 is indispensable for oocyte growth during postnatal ovarian development (Fig. 4D-E). We next performed H&E staining and counted the average numbers of follicles at various stages in the postnatal ovary sections. At 1 month, there was a slight reduction in the size of the Nat10-ZcKO ovary (Fig. 4E), but the morphological features and the proportion of follicles at various stages were indistinguishable between the Nat10-ZcKO and WT ovaries (Fig. 4E-G). Nevertheless, after the first wave of folliculogenesis, Nat10-ZcKO oocytes appeared to quickly degenerate, resulting in developmental arrest at secondary follicles in Nat10-ZcKO ovaries, as compared with WT ovaries (Fig. 4E-G, Fig. S2C and D).
+
+On the other hand, the permissive sterile phenotype in all the Nat10-ZcKO females implies that there exits functional deficiency in Nat10-ZcKO ovaries despite their morphological similarity to WT ovaries at the age of 1 month. Therefore, we next collected and stained oocytes from the PMSG-primed females at P21. The average numbers of GV oocytes retrieved were comparable between Nat10-ZcKO and WT ovaries (Fig. 5A and B). Notably, more bulged granules appeared to be observed in the cytoplasm of Nat10-ZcKO oocytes (Fig. 5A). In growing oocytes, cytoplasmic maturation is accompanied by the Non-Surrounded Nucleolus (NSN)-to-Surrounded Nucleolus (SN) DNA configuration transition \(^{10}\). SN
+ooocytes mostly dominate the GV stage in late antral follicles and are considered meiotically competent. Compelling studies have previously shown that DNA is transcriptionally inert in SN oocytes, characterized by the enhanced modifications of H3K4me3 and H3K9me3, while NSN oocytes exhibit active DNA transcription with decreased H3K4me3 and H3K9me3 modifications \(^{11}\). Indeed, further counting of NSN versus SN oocytes uncovered that the ratio of NSN to SN was distorted between Nat10-ZcKO and WT ovaries (Fig. 5C and D). IF staining by H3K4me3 showed markedly decreased H3K4me3 intensity in both NSN and SN oocytes in Nat10-ZcKO oocytes compared with WT oocytes (Fig. 5E and F). In contrast, H3K9me3 staining revealed dispersed and elevated chromatin signals in both NSN and SN oocytes in Nat10-ZcKO oocytes (Fig. 5G and H). This evidence suggests that the transcriptional machinery might be disrupted resulting in the compromised meiotic competence in Nat10-ZcKO oocytes. Together, these studies demonstrate that Nat10 is required for oocyte growth in developing follicles and hence for female fertility.
+
+Nat10 ablation impaired meiotic GV-MII progression owing to transcriptome disruption
+
+The sterility and the premature oocyte death in the Nat10-ZcKO females suggested that Nat10 is pivotal for meiotic maturation (GV-MII progression). To test this possibility, we next sought to isolate and culture GV oocytes *in vitro* and determine whether Nat10-ZcKO oocytes are capable of resuming meiosis. At 3 h after IBMX release in M16 medium, most WT GV oocytes (~90%) resumed meiotic division and entered the pro-metaphase I, as evidenced by the nuclear membrane breakdown of the germinal vesicles (GVBD), while a lower percentage of Nat10-ZcKO GV oocytes (~58%) managed to complete GVBD (Fig. 6A and B, Fig. S4A-D). Consistently, the number of Nat10-ZcKO oocytes that progressed to MII stage significantly declined compared to that of WT oocytes (23±2.5 vs 65±1.51) (Fig. 6C). To determine the precise arrested stage and how meiotic divisions were impacted in Nat10-ZcKO oocytes, we collected the superovulated oocytes *in vivo* after PMSG/hCG injection and performed the co-staining with DAPI and a cytoskeleton marker, tubulin, in the formaldehyde-fixed oocytes. The average number of collected oocytes significantly declined in Nat10-ZcKO ovaries compared with WT ovaries (3.8±1.15 vs 30.57±0.92) (Fig. 6D). In agreement with previous findings, most Nat10-ZcKO oocytes were arrested at the MI stage, with a small fraction of oocytes exhibiting anaphase-to-telophase arrest in Prophase I (AI-TI) in Nat10-ZcKO ovaries compared to WT oocytes (Fig. 6E and F). Furthermore, we carried out an *In vitro* fertilization (*IVF*) assay using superovulated MII oocytes. Nat10-ZcKO oocytes appeared to be fertilized and developed to the 2-cell stage, similar to the WT oocytes (Fig. 6G and H). However, the proportion of 4-cell stage embryos markedly declined in the maternal *Nat10* KO group. Together, these studies suggest that Nat10 is indispensable for oocyte meiotic maturation (Fig. 6).
+
+Oocyte growth is under stringent transcriptional regulation, with peak transcription occurring in growing oocytes at the early stage of antral follicles \(^{6, 11, 32}\). To address the factors underlying defective GV oocytes in Nat10-ZcKO ovaries, we carried out RNA-seq analyses with GV oocytes retrieved from WT and Nat10-ZcKO ovaries. In agreement with the sterile severity, a total of 1615 differentially expressed genes (DEGs) were identified (Cutoff: Fold change (FC)≥2, p<0.05), with similar numbers of genes up-regulated (839) and down-regulated (776) in the Nat10-ZcKO GV oocytes relative to the WT
+oocytes (Fig. S3A and C, Supplementary Table S4). Interestingly, Gene ontology (GO) analyses revealed that most down-regulated genes were enriched in transcription-related pathways, whereas the up-regulated genes were related to tRNA processing and meiotic cell cycles (Fig. S3B and D). Moreover, there were a total of 583 genes related to cell-cycle progression showing an alternative splicing pattern in Nat10-ZcKO oocytes (Fig. S3E-G). This evidence altogether demonstrated that the maternal transcriptome was disrupted in Nat10-ZcKO oocytes at the GV stage resulting in the impaired oocyte maturation.
+
+Mini-bulk SMART-seq2 identified defective maternal mRNA decay in Nat10-null MII oocytes
+
+Progression through oocyte development is accompanied by tightly regulated RNA transcription and degradation \(^{6, 33, 34, 35, 36}\). Transcription is active in growing oocytes, but is shut down when oocytes progress to the SN-type GV stage. Once meiosis resumes, the GV maternal transcriptome undergoes a global but selective degradation of ~20% polyA mRNAs, culminating in a characteristic maternal transcriptome in MII oocytes that differs from that in GV oocytes \(^{7, 12, 37, 38}\). To interrogate the molecular mechanism underlying the disrupted oocyte maturation in Nat10-ZcKO oocytes, we performed RNA-seq analyses. Given that one Nat10-ZcKO female only superovulated ~4 MII oocytes on average, we further optimized an *in-house* mini-bulk SMART-seq2 protocol that utilized 3–5 oocytes for each biological replicate for RNA-seq (Fig. 7A and B). We first verified the validity of our mini-bulk SMART-seq2 method by comparing our data with published bulk RNA-seq result in WT oocytes. On average, our method detected ~13337 genes in GV and ~12071 genes in MII, which are comparable to the ~13629 genes and ~12045 genes detected in GV and MII (Cutoff: TPM≥1), respectively, in the bulk oocyte RNA-seq datasets (Fig. 7B, Supplementary Table S4 and 5) \(^{22}\). This result confirmed the validity and sensitivity of our mini-bulk SMART-seq2 protocol.
+
+Next, we conducted the mini-bulk SMART-seq2 using 5 oocytes at MII stage from WT and Nat10-ZcKO females. This revealed a higher number of genes (1196) to be up-regulated than down-regulated (555) in Nat10-ZcKO MII oocytes (Fig. 7C, Fig. S4E and F, Supplementary Table S5). GO analyses showed that the up-regulated genes were mostly enriched in translation- and mRNA processing-related biological processes (Fig. 7D-F). By comparison, the down-regulated genes were enriched in transcription-related GO terms (Fig. S4G). A comparison of the relative expression levels of the transcripts showed that a higher number of transcripts was present in the Nat10-ZcKO MII oocytes than in the WT (TPM≥1) (Fig. 7G and H), suggesting an aberrant accumulation of maternal transcripts. To distinguish what specific type of transcripts was affected, the expressed transcripts were divided into five bins according to their expression levels in the WT MII oocytes. This revealed that *Nat10* KO caused global transcript up-regulation regardless of their expression abundance (Fig. 7I). Since there is global polyA mRNA degradation during GV-MII transition in WT oocytes, the expressed transcripts in WT MII oocytes were allocated into three types: Up- [FC(MII/GV) ≥2, p<0.05], Down- [FC(MII/GV) ≤2, p<0.05], and Stable-type (remaining transcripts). A Sankey plot showed that the majority of up-regulated genes (965/1196) in the Nat10-ZcKO MII oocytes overlapped with the Down-type genes in the WT MII oocytes, suggesting that the 965 transcripts destined for decay during meiotic maturation failed to be eliminated in the Nat10-ZcKO MII oocytes (Fig. 7J). To determine which transcripts were susceptible to
+degradation in the presence of Nat10, we defined a total of 2011 transcripts down-regulated in WT oocytes during the GV-MII transition and 1206 transcripts down-regulated in Nat10-ZcKO oocytes through GV-MII transition. Interestingly, a large fraction of 1416 transcripts (among 2011 transcripts in total) did not overlap with down-regulated transcripts in Nat10-ZcKO oocytes, suggesting that they were not timely degraded but aberrantly accumulated in Nat10-ZcKO oocytes (Fig. 7K). The degradation trend profiling verified that the 1416 transcripts indeed displayed a decreased degradation speed in the absence of Nat10 (Fig. 7L). Of note, a total of 442 genes displayed an aberrantly alternative splicing pattern (Fig. S4H and I). Taken together, these studies suggest that Nat10 is essential for the maintenance of normal maternal transcriptome by the timely degradation of selective maternal mRNAs.
+
+Hairpin Adaptor-PolyA Tail length (HA-PAT) assay, a simple, low-cost and sensitive method for validation of polyA mRNA degradation in single oocytes
+
+CCR4-NOT deadenylase complex mediated polyA mRNA degradation has been well-documented to be responsible for maternal mRNA degradation during the MZT \(^{33, 39, 40}\). Indeed, we discovered that ~65% (782/1195) of transcripts up-regulated in Nat10-ZcKO MII oocytes were also accumulated in the Cnot6l (CCR4-NOT subunit) KO MII oocytes (Fig. 8A). Intriguingly, these up-regulated genes were especially enriched in ribosomal subunit components that were normally degraded through polyA tail-shortening mediated mRNA decay via the CCR4-NOT complex, a phenotype that was also observed in CCR4-NOT-deficient oocytes (Fig. 8B) \(^{33, 41}\). Further examination by qPCR analysis revealed that the mRNA levels of three members (Cnot6l, Cnot7 and Btg4) of CCR4-NOT pathway were significantly down-regulated in the Nat10-ZcKO MII oocytes compared with WT oocytes (Fig. 8C). This evidence led us to hypothesize that Nat10-driven maintenance of CCR4-NOT deadenylation activity is required for the polyA mRNA elimination and consequently the establishment of the maternal transcriptome.
+
+We thus next tested whether the accumulated transcripts harbor long polyA tails in Nat10-ZcKO MII oocytes. Currently, two methods are commonly adopted for PolyA tail length (PAT) examination, including Ligase-mediated PAT (LM-PAT) and extension PAT (ePAT) \(^{42, 43, 44}\). The original LM-PAT method relies on oligo(dT)\(_{12-18}\) hybridization and T4 ligation, followed by oligo-(dT) anchor primer-mediated PCR amplification \(^{43, 44}\); ePAT exploits a hybridized oligo(dT) anchor primer as a DNA template for Klenow enzyme-mediated 3' extension ensued by PCR amplification \(^{42}\). Given the limited availability of Nat10-null MII oocytes, we initially tried both methods using only 5–10 oocytes as input materials; however, the PCR amplification either failed or produced very weak, unsatisfactory results, presumably owing to the low sensitivity of both methods when dealing with low-input samples (data not shown). Therefore, we next sought to develop a new PAT assay that can circumvent the drawbacks of previous approaches (Fig. 8D). After several rounds of optimization, we termed this novel method the Hairpin Adaptor-PolyA Tail length (HA-PAT), which is more sensitive, low-cost and time-saving than previous methods \(^{42, 44}\). HA-PAT utilizes an exquisitely designed hairpin adaptor that can self-hybridize plus an extended oligo (dT)\(_8\) with two additional degenerate W nucleotides at the 3' end (Fig. 8D). The loop sequence is connected through a C3 spacer, which provides a sufficiently flexible linker to enhance the self-hybridization of stem sequences. In practice, this strategy was optimized to facilitate specific anchoring at the 3' end of polyA tail and to reduce the random dT hybridization background. Moreover,
+given that the 3' ends of polyA mRNA tails dominate with two U nucleotides in the oocytes 7, the two degenerate "WW" nucleotides will further capture and stabilize the binding between the hairpin adaptor and the "U"-terminated polyA tails. Therefore, this single hairpin adaptor not only provides an anchor primer for reverse transcription, but also harbors the full reverse primer sequence for subsequent cDNA PCR amplification. Thanks to its integral design strategy, the total reagent cost and hands-on operations were significantly reduced compared with the other two common methods (Fig. S5A). To test its efficiency, we selected two known genes (Rpl35a and Chchd2) that accumulated in MII Nat10-null oocytes and the house-keeping Gapdh gene as a control. As shown in Fig. S5B, using 5 oocytes as input material without PCR preamplification, neither LM-PAT nor ePAT could detect Rpl35a and Chchd2 genes. In contrast, our HA-PAT assay yielded clear, differential smear bands between WT and Nat10-null oocytes, suggesting that the HA-PAT approach is more sensitive in detecting low-input mRNA samples (Fig. S5B). Under PCR preamplification for 16 cycles using five oocytes, LM-PAT showed strong signals for the smear polyA tails, whereas ePAT failed to produce satisfactory results (Fig. S8D). Nonetheless, our HA-PAT method outperformed the other two methods since it gave rise to stronger and clearer bands (Fig. S5B). This evidence validated the sensitivity and efficiency of the LM-PAT assay in detecting polyA tail length using a minute amount of input RNA material from single oocytes.
+
+Taking advantage of the HA-PAT approach, we next selected a panel of representative genes showing up-regulation in each GO cluster identified by mini-bulk SMART-Seq2 for validation. We collected five hormone-primed oocytes at GV and MII stages for each sample and executed 16 cycles of PCR amplification. As shown in Fig. 8E-F, all the polyA tails of the seven genes were lengthened in MII Nat10-null oocytes, as judged by the elevated, smear PCR bands. Additionally, as a cross-validation, we further performed LM-PAT, which corroborated the similar findings (Fig. S6). Taken together, these results indicate that Nat10-mediated polyA-tail shortening and the resulting mRNA degradation sculpted the transcriptome of MII oocytes.
+
+Evidence that maternal Nat10 is translationally required for pre-implantation embryo development
+
+As an RNA acetyltransferase, Nat10 harbors a C-terminal tRNA-binding domain and was originally identified and experimentally validated to modify two well-known eukaryotic RNA substrates: tRNAsSer/Leu and 18S rRNA 24, 45. Both types of RNA are canonical noncoding RNAs that are involved in mRNA translation to protein products. We next asked whether maternal Nat10 is physiologically important for protein translation. Oocytes are unique in their properties, as they amass abundant mRNAs, of which most are stored for translation during later maternal-zygotic transition (MZT), owing to the uncoupling of transcription and translation in maturing oocytes 46. As such, oocytes were often broadly exploited as a model to study cis-element mediated mRNA translation. To this end, we generated an in-house knockin, tamoxifen (TMX)-inducible Ddx4-cre mouse model (Ddx4-CreERT2), in which the Cre activity is specifically turned on in Ddx4-expressing germ cells upon TMX injection (data not shown, manuscript in preparation). Ddx4-CreERT2 females were crossed with Nat10lox/lox males to attain Nat10lox/lox; Ddx4-CreERT2 offspring (henceforth termed Nat10-DcKO following TMX injection) (Fig. S7A). We performed the TMX injection for three consecutive days starting from P17,
+when the oocytes had already proceeded to the GV stage in the antral follicles during the first wave of postnatal folliculogenesis. At P21, we injected PMSG and retrieved the oocytes 48 h later for culture *in vitro*. As shown in Fig. S7B, the mRNA levels of Nat10 were significantly reduced in Nat10-DcKO oocytes compared with WT oocytes, indicative of the successive deletion of *Nat10* in oocytes. However, we did not observe any distinguishable difference in terms of the gross morphology, the average numbers, or the percentage of PMSG-primed oocytes at GV stage, and at MI or MII stages when cultured in vitro (Fig. S7C-I). The morphology or average numbers of superovulated MII oocytes following PMSG/hCG priming were also indistinguishable between WT and Nat10-DcKO females (Fig. S7J). Interestingly, we performed IVF using superovulated WT and Nat10-DcKO MII oocytes with WT sperm, which unveiled that the pre-implantation embryos were arrested at the 2-cell stage in Nat10-DcKO zygotes, suggesting that maternal Nat10 is essential for pre-implantation embryo development (Fig. S7K and L). Considering the high prevalence of Nat10 mRNA in oocytes at MI, MII and at 2-cell embryos (Fig. 1C), the absence of phenotypic outcome in Nat10-DcKO oocytes presumably hinted that the residual Nat10 mRNA or protein was still functional after TMX-induced Cre activation.
+
+To further corroborate this finding, we utilized another well-established global Cre-inducible mouse model, namely Ubc-CreERT2, to perform similar crossings to obtain Nat10*lox/lox*, Ubc-CreERT2 (termed Nat10-UcKO) mice after TMX injection (Fig. S8A). Following TMX injection, Ubc-Cre was activated in all cell types, including somatic granulosa cells and oocytes in the ovary. Next, we carried out all subsequent experiments similar to those for Ddx4-CreERT2 as described above. In line with previous studies, all results were identical to the conclusions as drawn in Nat10-DcKO mice with only one exception – the preimplantation embryos were arrested at 1-cell zygote stage (Fig. S8). The discrepancy of earlier zygotic arrest between Nat10-DcKO- and Nat10-UcKO-derived embryos was likely due to the defective function of surrounding granulosa cells since Nat10 protein is also highly abundant in the nucleolus of the supporting granulosa cells (Fig. 1F). Given that no active transcription but exclusive translation occurs during late oocyte-to-zygote transition, these studies were reminiscent of an integral physiological role of Nat10 in coordination of mRNA translation during MZT.
+
+Next, we sought out to examine how Nat10 impacted mRNA translation. Given the limited availability of oocytes, we decided to generate stable, Cre-inducible cell lines by exclusively utilizing our established Nat10*lox/lox*, Ubc-CreERT2 mice. We followed a “3T3” protocol and successfully generated two stable, 4-hydroxytamoxifen (4’-OHT)-inducible, mouse embryonic fibroblast (MEF) cell lines from E13 embryos (Nat10*lox/lox*, Ubc-CreERT2) after 2-month in vitro culture (Fig. 9A) 20. Immunoblotting and qPCR analyses demonstrated that Nat10 protein and mRNAs were vastly declined upon OHT induction, respectively (Fig. 9B and C), indicative of the successful establishment of two stable cell lines. Apparently, Nat10 depletion significantly reduced the cell proliferation without any effect on cell apoptosis, as evidenced by the Ki67 staining and the CCK8 assays (Fig. 9D-H). Reciprocally, rescue by Nat10 overexpression in Nat10-inducible KO MEF cells recovered and enhanced the cell division (Fig. 9F-H). Noteworthily, we consistently observed that longer induction of Nat10-UcKO MEF cells by 4’-OHT treatment led to complete cell cycle arrest and resulting cell death, suggesting that Nat10 is essential for cell survival. The permissive cell death in Nat10-UcKO MEF cells contrasted with the
+viability observed in Nat10-null HeLa cells \(^{15}\). Next, we pursued to explore how translation was impacted in Nat10 KO MEF cells. We collected similar numbers of MEF cells without and with 4'-OHT treatment for three days for polysome profiling. Consistently, we found that Nat10 KO cells displayed lower levels of mRNA-bound polysomes than non-treated MEF cells, indicating the translational efficiency was repressed in the Nat10-null MEFs (Fig. 9I). Intriguingly, both the levels of the 40S subunit and 80S ribosome were concurrently reduced in KO cells compared with WT cells (Fig. 9I and J). We reasoned that this presumably resulted from the assembly defect of the 40S ribosomal subunit protein binding to 18S rRNA, whose ac4C modification is essential for ribosome assembly as previously reported \(^{46}\). In summary, these studies provided evidence that Nat10 is required for mouse pre-implantation embryo development, at least in part, through modulating mRNA translation.
+
+Discussion
+Oocyte development, RNA modification, and ac4C modification
+
+Dysregulated expression of Nat10 has been recently linked to numerous diseases, such as Hutchinson-Gilford progeria syndrome (HGPS) \(^{47}\), epithelial ovarian cancer \(^{48}\), breast cancer \(^{16}\), and spermatogenesis \(^{19}\), and is most recently implicated in oocyte *in vitro* maturation (IVM) \(^{49}\). Mature oocytes are specialized, transcriptionally inert cells that almost exclusively contribute to the cytoplasm of zygotes when fertilized by sperm. Therefore, all the early developmental events of preimplantation embryos occur at the post-transcriptional level and are dependent on the stored RNA content derived from oocytes, called the maternal transcriptome \(^{6, 32, 35, 36}\). In mammals, a hallmark of gene expression for the maternal transcriptome is characterized by the uncoupling of transcription and translation – some accumulated mRNA species are immediately translated to support oocyte growth, while a large stock of other species is stabilized as ribonucleoprotein particles (RNPs) in a translationally inactive state, in growing follicles \(^{6, 38, 39}\). When fully-grown GV oocytes resume meiosis, known as a process called meiotic maturation, however, the situation substantially changes - many previously active mRNAs become translationally silent, whereas some previously “dormant” mRNA species are translationally reactivated. A massive wave of RNA elimination is considered as a hallmark and a major driving force underlying oocyte-to-embryo transition (OET). It has been estimated that up to ~40% polyA RNA is degraded during the GV-to-MII oocyte transition; however, the molecular mechanisms underlying this RNA remodeling event are not well understood \(^{6, 38, 39}\).
+
+Post-transcriptional RNA modifications constitute an exciting layer of so-called “epi-transcriptome” that modulates gene expression without altering RNA sequences. Recent evidence showing N6-methyladenosine (m6A) modification abundantly present in mRNA species is an attractive model that selectively marks any mRNA for degradation at the post-transcriptional level. For instance, m6A is specifically deposited by a large, heterogeneous multiprotein complex, known as the m6A “writer”, which comprises a core catalytic member of METTL3, along with METTL14, WTAP and KIAA1429 (VIRMA) cofactors \(^{50}\). Genetic evidence in conjunction with m6A antibody-mediated RNA-immunoprecipitation sequencing (m6A-seq, or MeRIP-seq) showed that the m6A mark specifically decorates the 3'UTR close to the STOP codon region, and defective MZT transition is tightly associated with reduced m6A levels in m6A-deficient oocytes, suggestive of an essential role of m6A in active RNA
+degradation through the MZT \(^{5, 14, 34, 51}\). On the other hand, to be functional, the m6A mark on the substrate mRNA must first be recognized by the “reader” proteins, e.g., YTH domain-containing protein family (YTHDF1-3, YTHDC1-2), which interpret and relay substrates to the downstream signaling. However, unexpectedly, more recent compelling evidence validated that, either the “writer” METTL3/METTL14 or YTHDC2, exerts their critical biological functions independent of the m6A modification in somatic cells and the germline, respectively \(^{52, 53, 54}\). For example, METTL3/METTL14 drives local chromatin remodeling through binding to target genomic loci, whereas YTHDC2 binds a class of mRNAs containing U-rich motif in their 3'UTRs and coding sequences. This evidence indicates that we might overinterpret the role of m6A in mediating its diverse biological functions observed in either the “writer” or the “reader” knockout mouse models. Indeed, this mechanistic discrepancy could be explained by methodological limitations - most previous studies that used high-throughput sequencing for mRNA m6A identification were highly dependent on the anti-m6A antibody, which is known to suffer from notable limitations, such as non-specificity, low resolution, and high input RNA materials\(^{55}\). In addition, those conclusions were most often indirectly drawn based on the correlation between the phenotypic outcome and the declined overall levels of m6A, but lacked the direct evidence gauging the m6A stoichiometry to its functional readout.
+
+In contrast to m6A modification, which is mostly prevalent (~0.5%) in mRNA, the abundance of ac4C modification in mammalian mRNAs is currently controversial \(^{56}\). An initial study identified that ac4C tends to enrich in the 5'UTR region in thousands of coding mRNAs, on the basis of ac4C antibody-mediated immunoprecipitation and sequencing (acRIP-seq), similar to the m6A-seq strategy, in HeLa cells \(^{15, 57}\). However, surprisingly, a later study unveiled no even a single ac4C site in mammalian mRNAs through a convincing chemistry-catalyzed strategy with quantitative and nucleotide resolution \(^{58, 59}\). Importantly, as a positive control, they successfully identified the previously known ac4C sites in both rRNA and tRNAs, as well as low levels of a few hundred of ac4C acetylation sites in Nat10-overexpressing human cells, implying that their adopted method is sensitive and feasible for detecting ac4C modification \(^{58, 59}\). Hence, the ac4C modification levels are, if present, extremely low in human mRNAs. The discrepancy between the two studies likely arises from the promiscuous binding of the ac4C antibody, as described above, which suffers from poor characterization and intrinsic cross-reactivity. We have found that, consistent with a recent study \(^{19}\), the ac4C/C ratio approximates ~0.04% in mRNAs from mouse testicular cells, but is roughly 1/10 of the m6A abundance in mouse testis (data not shown), based on mass-spectrometry (MS) measurement \(^{58, 59}\). This prompted extremely low levels, or negligible levels of ac4C modification in mammalian mRNAs, and hence, we reasoned that Nat10 exerts its critical functions independent of ac4C acetylation. Nonetheless, owing to the large mRNA requirement, we were not able to perform the whole-transcriptome acRIP-seq to assess the effect of ac4C modification using isolated RNA from Nat10-null oocytes.
+
+The pleiotropic roles of Nat10 through oocyte meiotic progression and maturation
+
+The NAT10 is a highly conserved protein from E. coli and yeast to mammalian species, and is the solely known enzyme responsible for ac4C modification on RNA substrates. It is a relatively large protein that comprises four distinct domains, including DUF1726, Helicase, GNAT and tRNA binding
+domain. In this study, we employed four Cre deleter mouse models to cross with floxed Nat10 alleles to attain germline-specific or inducible Nat10 KO mice, yielding full Nat10-null mouse offspring without any of the four domains owing to premature frame-shift reading. Consistent with the high expression levels of Nat10 in the female germline, pre-meiotic ablation of Nat10 caused apparent meiotic arrest at the pachytene stage, resulting in premature ovarian failure and female infertility in adults (Fig. 2). We showed that this defect was likely caused by the deficient DSB repair as judged by the high persistent γH2AX intensity and enhanced RPA2 remnant (Fig. 3). This conclusion agrees with a previous study showing that DNA-damaging agents induced Nat10 expression in a dosage- and time-dependent fashion \(^{60}\). Nat10 depletion in growing oocytes of primary follicles disrupted the NSN-SN transition of GV oocytes and damaged meiotic maturation, as evidenced by the aberrant GV to MII transition \(^{10}\). By optimized mini-bulk SMART-seq2 analyses, we discovered that a large number of genes enriched in cell cycles and DNA transcription were dysregulated in Nat10-null GV oocytes (Fig. S3). In particular, we revealed that many genes destined for degradation in the MII oocytes were aberrantly accumulated in Nat10-null MII oocytes, which was corroborated by a novel, *in-house* developed HA-PAT approach. Further evidence showed that Nat10 deletion caused the down-regulated expression of important members of the CCR4-NOT complex, at least in part, at the transcriptional level \(^{33,39}\). These studies demonstrate that NAT10 is transcriptionally essential to maintain transcriptomic homeostasis to facilitate oocyte growth and maturation. Finally, we provided both genetic and *in vitro* evidences showing that NAT10 is translationally required for oocyte developmental competence, since Nat10 depletion caused a remarkable reduction in translational efficiency, as evidenced by the polysome profiling. This most likely resulted from the assembly defects of the ribosomes, a conclusion that is consistent with a previous study wherein down-regulation of Nat10 disrupted the biogenesis of 40S ribosomal subunit due to abolished ac4C modification on 18S rRNA \(^{24,45}\).
+
+In conclusion, we provide genetic and *in vitro* studies showing that Nat10 is transcriptionally required for the prophase I progression and meiotic resumption during oocyte growth and meiotic maturation. Nat10 is also translationally indispensable for the developmental competence of the oocytes. However, further experiments are urgently needed to elucidate the pathogenic roles of the distinct domains of NAT10 and to decipher whether there is a causative relationship, if present, between the ac4C RNA modification and the phenotypic outcome.
+
+Methods
+Mice
+
+The floxed Nat10 (Nat10\(^{lox/lox}\)) alleles were generated by GemPharmatech Co., Ltd. Conditional Nat10 knockout mice were achieved by crossing Nat10\(^{lox/lox}\) mice with Stra8-GFPCre and with Zp3-Cre mice to attain the Nat10-ScKO and Nat10-ZcKO offspring, respectively. Induced Nat10 KO mice were generated by tamoxifen injection for three consecutive days in Ubc-CreERT2 or Ddx4-CreERT2 mice. The Stra8-GFPCre knock-in (KI) mouse line was generated in Ming-Han Tong’s Lab at the Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences. Zp3-Cre and Ubc-CreERT2 KI mice were obtained from Jackson Laboratory. Ddx4-CreERT2 KI mouse line was generated *in-house*.
+All mice were from the C57BL/6J background, and were bred in a specific pathogen-free (SPF) facility with a 12h light/dark cycle and with free access to food and water. All animal experiments were approved by the Animal Care and Research Committee of the University of Science and Technology of China (USTC).
+
+Inducible Nat10 KO cell lines and culture
+
+We generated two tamoxifen-inducible, stable cell lines following a standard 3T3 protocol \(^{20}\). In brief, the pregnant female mice were euthanized, and the embryos at embryonic day 13.5 (E13.5) were carefully dissected. After removing somatic organs, the embryonic body was chopped into small pieces and digested with DMEM medium containing 0.25% trypsin-EDTA in a 37 °C water bath for 30 min. After filtering through a 70 μm cell strainer, the single cells were cultured in MEF medium (DMEM medium supplemented with 10% FBS (VivaCell, C04001) and 1% Penicillin-Streptomycin (Biosharp, BL505A)). The cells were subcultured and the medium was replenished every three days, with the passage number recorded. Cells were grown in a 5% CO\(_2\) cell culture incubator (Heal Force) at 37°C. Two stable MEF lines were achieved after passage 23 following recovery of the MEF cells from the crisis around passage ~10-14 (total time is ~2 months).
+
+Lentiviral transduction
+
+A mouse Nat10 cDNA plasmid clone (EX-Mm12636-M45) was purchased from GeneCopoeia, Inc. and cloned into pCDH-CMV vectors in-frame with a FLAG tag. The lentiviral vector was co-transduced into 293T cells alongside a packaging vector psPAX2 and a helper vector pCMV-VSVG using LIP2000 transfection reagent (BioSharp, BL623B) for lentivirus production. Viral particles were collected to infect MEF cells with 8 mg/ml Polybrene (Solarbio, H8761). The infected cells were positively selected with puromycin (2.5 mg/ml) (Solarbio, P8230) for 48 hours.
+
+Oocyte collection and in vitro culture
+
+The 21-day-old female (P21) mice were injected with 5 IU of pregnant mare’s serum gonadotropin (PMSG) (Ningbo Sansheng Pharmaceutical). After 48 h, the mice were euthanized and the oocytes at the GV stage were harvested in M2 medium (Nanjing Aibei Biotechnology Co., Ltd, M1250) and further cultured in M16 medium (Sigma, M7292) covered with mineral oil (Sigma, M5310) at 37 °C in 5% CO\(_2\). Samples were imaged with a microscope (SZX7, Olympus).
+
+Superovulation and in vitro fertilization
+
+For superovulation, 21-day-old female mice were injected with 5 IU of PMSG. After 46 h, the mice were injected with 5 IU of human chorionic gonadotropin (hCG) (Ningbo Sansheng Pharmaceutical). At post-hCG 16 h, the cumulus-oocyte complex (COC) was retrieved from the oviducts, and the oocytes were counted after digestion with 0.3% hyaluronidase (Sigma, H4272). For *in vitro* *fertilization* (IVF), superovulated female mice were euthanized 15 h after hCG injection, and the ampulla parts of the oviducts were collected in HTF medium covered by mineral oil. COCs were released and fertilized with WT sperm in HTF for 30 min at 37°C in a 5% CO\(_2\) incubator, and further cultured in KSOM medium in vitro (Sigma, MR-101).
+
+RNA isolation and real-time RT-PCR
+Total RNA was isolated from mouse tissues with TRIzol Reagent following the manufacturer’s instruction as described previously 21. In brief, freshly collected or frozen tissues were homogenized in TRIzol reagent. For MEF cells, total RNA was isolated with a SPARKeasy Cell RNA kit (Shandong Sparkjade Biotechnology Co., Ltd., AC0205). The quantity and quality of RNA samples were determined by measurement using a NanoPhotometer N50 (Implen, Germany). The RNA samples with OD values of 260/280 ≥ 1.9 were selected for downstream analyses. Equal amounts of total RNA were loaded to synthesize cDNAs using the Hiscript III Reverse Transcriptase (Vazyme, R302-1). Quantitative PCR (qPCR) was performed using HiEff® qPCR SYBR Green Master Mix (Yeasen, 11201ES03) on a Q2000B Real-Time PCR machine (LongGene). For oocytes, 5~10 oocytes were lysed in 2 μl lysis buffer (0.2% Triton X-100 and 2 IU/μl RNase inhibitor) followed by reverse transcription and PCR pre-amplification for 8~16 cycles. The PCR products were diluted and used for qPCR. The primers are listed in Supplementary Table S1.
+
+Histological analysis
+
+Hematoxylin & Eosin (HE) staining was performed following the standard procedure as described previously 21. In brief, ovary samples were freshly collected, and fixed in Bouin’s solution at room temperature overnight. Paraffin-embedded samples were cut into slides with 5 μm thickness. Slides were de-paraffinized with xylene and re-hydrated, followed by staining with HE. The slides were then dehydrated and mounted with neutral resins. Images were taken on a light microscopy (MShot) with a MSX2 camera.
+
+Western blot analysis
+
+Samples were freshly collected and lysed in RIPA solution [100mM Tris-HCl (PH7.4), 1% Triton X-100, 1% Sodium deoxycholate, 0.1% SDS, 0.15M NaCl, supplemented with Protease inhibitor cocktail]. Protein concentrations were determined using a BCA protein assay kit. All protein samples were run in 8% of denatured sodium dodecyl sulfate polyacrylamide (SDS-PAGE) gel with Trelief® Prestained Protein Ladder (TSINGKE, TSP021), followed by wet-transfer to PVDF membranes. Subsequently, the membranes were blocked in 1XPBS with 5% non-fat milk and incubated with primary antibody followed by secondary antibody. Signals were visualized using an imaging system (SHST, Hangzhou, China). The following antibodies were used: rabbit anti-NAT10 (ZENBIO, 389412, 1:1000), mouse anti-PCNA (Proteintech, 60097-1-Ig, 1:2000), mouse anti-GAPDH (Proteintech, 60004-1-Ig, 1:10000), rabbit anti-Tubulin (Proteintech, 11224-1-AP, 1:5000).
+
+Chromosome spreads analysis and immunofluorescent staining
+
+For oocyte chromosome spreading analyses, newborn pups were sacrificed, and ovaries were dissected in hypotonic buffer [30mM Tris-HCL (pH=8.2), 50mM sucrose, 17mM trisodium citrate dihydrate, 5mM EDTA (pH=8.0), 1mM dithiothreitol (DTT), and 1mM phenylmethylsulfonyl fluoride (PMSF)] for 25 min. Next, ovaries were transferred to 100 mM sucrose, and single cells were released into sucrose solution using syringe needles. Single cells were spread and fixed in 1% PFA solution containing 0.15% Triton X-100 on slides, followed by washing with 0.4% Photo-Flo. For immunofluorescent staining, cells were permeabilized with 0.3% Triton X-100, blocked with 5% normal goat serum (Solarbio, SL038) in PBST, and incubated with primary antibodies diluted in blocking
+solution at 4°C overnight. Antibodies used were as follows: mouse anti-SYCP3 (Abcam, ab97672, 1:1000), rabbit anti-SYCP1 (Abcam, ab15090, 1:1000), rabbit anti-SYCP3 (Proteintech, 23024-1-AP, 1:200), mouse anti-yH2AX (Millipore, 05-636, 1:1000), rabbit anti-RPA2 (Proteintech, 10412-1-AP, 1:400). After washing in 1XPBS (0.1% Tween) for three times, the samples were incubated with TRITC-conjugated Goat Anti-Rabbit IgG (Proteintech, gb2AF488) or 488-conjugated secondary antibodies (Proteintech, gb2AF488). For ovarian immunofluorescent staining, ovaries were fixed in 4% paraformaldehyde (PFA) overnight at 4°C on a rocker. Slides were dehydrated with 10% and 20% sucrose for 2 h each. Ovary samples were cut into 8 μm slides. For immunofluorescence staining of MEF cells and oocytes, they were fixed in 4% PFA for 30min. The primary antibodies used were as follows: rabbit anti-NAT10 (Proteintech, 13365-1-AP, 1:400), rabbit anti-NAT10 (ZENBIO, 389412, 1:500), rabbit anti-Nucleophosmin (Abcam, ab10530, 1:1000), rabbit anti-H3K4me3 (Abclonal, A2357, 1:200), rabbit anti-α-Tubulin (Proteintech, 11224-1-AP, 1:200), rabbit anti-KI67 (Servicebio, GB111141, 1:400), rabbit anti-FLAG (Proteintech, 80010-1-RR, 1:400). Slides were imaged by a Leica THUNDER Imager Live Cell with a K5 camera driven by the Leica Application Suite Software. Image processing was performed by ImageJ software.
+
+PolyA-tail (PAT) length assay
+
+For the Hairpin-Adaptor PAT (HA-PAT) assay, ten denuded oocytes were freshly lysed in 2 μl lysis buffer (0.2% Triton X-100 and 2 IU/μl RNase inhibitor). The hairpin adaptor was pre-annealed in 1Xoligo annealing buffer (50mM Tris, pH8.0, 50mM NaCl, 1mM EDTA). PolyA-tail mRNAs in the oocyte lysates were denatured at 72 °C for 3 min and then hybridized with the annealed hairpin adaptor at 25 °C for 10min. The reverse transcription mix contained 100 U of SuperScript IV, 10 U of RNase Inhibitor, 5 mM of DTT, 1 M of Betaine, 6 mM of MgCl2, 1 mM of dCTP (for template-switching) and 1.5 M of P5TSO. The 1st strand cDNA was synthesized by reverse transcription at 42 °C for 90 min. The full-length cDNAs were pre-amplified through semi-suppressive PCR for 8 or 16 cycles. The pre-amplified cDNA products were diluted and used for gene-specific PCR reactions using gene-specific primers (GSP) and polyA reverse primer 2. For the Ligation-Mediated PolyA Test (LM-PAT), oocyte samples were hybridized with oligo(dT)20 at 42 °C for 30min, and then ligated with dT anchor primer by T4 DNA ligase (Sangon, B600511) at 12 °C for 2h. Reverse transcription was performed with P5TSO and 1mM dCTP for 1h, ensued by PCR pre-amplification for 8 or 16 cycles using dT anchor and P5TSO. The diluted PCR products were used for gene-specific amplification. For extension PolyA Test (ePAT), oocyte polyA-tail RNAs were 3'-prime extended using ePAT anchor primer as a template with Klenow enzyme (New England Biolabs, K0210) at 25 °C for 1h and 80 °C for 10min. Reverse transcription was performed with P5TSO and 1mM dCTP for 1h, followed by pre-amplification for 8 or 16 cycles using ePAT anchor and P5TSO primers. PCR products were analyzed on a 2% agarose gel. All PCR primers are listed in Supplementary Table S2.
+
+Cell proliferation and apoptosis assays
+
+For the cell proliferation assay, 2000 MEF and OHT-treated MEF cells were plated in 96-well plates with 200 μL of fresh complete DMEM medium supplemented with 10% FBS. After incubation for 1, 3, 5 and 7 days, cell viabilities were measured using Cell Counting Kit-8 (MedChemExpress, HY-K0301)
+following the manufacturer’s instructions. For apoptosis assays, an Annexin V-FITC/PI Apoptosis Detection Kit (YEASEN, 40302ES50) was used following the manufacturer’s protocol. Samples were detected by the BD Accuri C6 flow cytometry, and the results were analyzed with FlowJo V10 software.
+
+Polysome profiling
+
+For polysome profiling, cells were cultured in 10-cm dishes and treated with OHT for 3 days. Cells were washed with cold PBS supplemented with 100 µg/ml cycloheximide and collected by centrifugation. Cell pellets were lysed in lysis buffer [50 mM HEPES, 2 mM MgCl₂, 100 mM KCl, 100 µg/ml cycloheximide, 1 mM DTT, 0.5% Triton X-100, 10% glycerol, and 20 U/ml EDTA-free protease inhibitor cocktail (Sigma, 11836170001)]. The lysate was cleared by centrifugation at 12,000g for 10 min at 4 °C, and the supernatant was loaded onto a 20–50% density gradient of sucrose cushion [30 mmol/l Tris–HCl (pH 7.5), 100 mmol/l NaCl, 10 mmol/l MgCl₂, protease inhibitor cocktail (EDTA-free), and 100 units/ml RNase inhibitor (APExBIO, K1046)], ensued by ultracentrifugation in a rotor at 38,000 rpm for 3 h at 4 °C. After centrifugation, the gradient was fractionated and the absorbance at 254 nm was continuously recorded using an ISCO fractionator (Teledyne ISCO).
+
+Mini-bulk SMART-seq2 for RNA-seq library preparation
+
+The oocyte mini-bulk SMART-seq2 protocol was based on the well-established SMART-seq2 protocol with in-house optimized modifications as indicated below. In brief, after hormone challenge, five oocytes retrieved from each animal were washed at least five times in 1XPBS containing 0.5% BSA/PVP, and directly lysed in Lysis Buffer MasterMix (0.3% Triton, 40U/µL RNase inhibitor, 2.5µM oligodT₃₀VN, and 2.5µM dNTP mix). The oocyte Lysis mixture was allowed to undergo at least one freeze-thaw cycle at -80°C to facilitate complete cytosolic lysis. The 1st strand cDNA synthesis was performed at 42°C for 90 min, followed by 14 cycles of PCR preamplification to attain the full-length cDNA products through ISPCR primer-mediated semi-suppressive PCR. 1 ng of size-selected full-length cDNAs was used for Tn5-guided library preparation using HiEff NGS Fast Tagment DNA Library Prep Kit for Illumina (Yeasen, 12206ES96). Final dual-barcoded libraries were achieved through PCR amplification for 8 cycles with both index i5 and i7 primers prior to pooled library sequencing on the NovaSeq 6000 platform with PE150 mode (Novagene).
+
+RNA-Seq data analysis
+
+Raw reads were processed to remove adaptor contaminants and low-quality bases. The clean reads were aligned to the mouse genome (mm10) using STAR, and uniquely mapped reads were counted with RSEM by default parameters (Supplementary Table S3). We quantified gene expression levels with TPM. For each sample, the expressed genes were defined with cutoff: TPM≥1. Differentially expressed genes (DEGs) were assessed with the DESeq2 package with a cutoff: Padj <0.05 and fold change (FC) ≥2. Gene Ontology (GO) enrichment was performed using DAVID (https://david.ncifcrf.gov/). rMATS was used to analyze the alternative splicing events. Statistical analyses were performed using R software (http://www.rproject.org). The RNA-seq data in this work have been submitted in the NCBI Gene Expression Omnibus (GEO) under accession number SRP392832.
+Statistical analysis
+
+All experiments were performed at least in biological triplicates unless otherwise indicated. Statistical analysis was performed using Student’s t-test unless otherwise stated. Values of \( p < 0.05 \) were deemed statistically significant. Statistically significant values of \( p < 0.05 \), \( p < 0.01 \), \( p < 0.001 \) and \( p < 0.0001 \) are indicated by one, two, three and four asterisks, respectively. Statistical data were calculated by R or GraphPad Prism 6.
+
+Data availability
+
+All the raw data and processed files have been deposited in the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) with the accession number: SRP392832
+References
+
+1. Xu C, Cao Y, Bao J. Building RNA-protein germ granules: insights from the multifaceted functions of DEAD-box helicase Vasa/Ddx4 in germline development. Cell Mol Life Sci **79**, 4 (2021).
+
+2. Jagarlamudi K, Rajkovic A. Oogenesis: transcriptional regulators and mouse models. *Mol Cell Endocrinol* **356**, 31-39 (2012).
+
+3. Koubova J, Menke DB, Zhou Q, Capel B, Griswold MD, Page DC. Retinoic acid regulates sex-specific timing of meiotic initiation in mice. *Proc Natl Acad Sci U S A* **103**, 2474-2479 (2006).
+
+4. Guan Y, *et al.* SCF ubiquitin E3 ligase regulates DNA double-strand breaks in early meiotic recombination. *Nucleic Acids Res* **50**, 5129-5144 (2022).
+
+5. Hu Y, *et al.* Oocyte competence is maintained by m(6)A methyltransferase KIAA1429-mediated RNA metabolism during mouse follicular development. *Cell Death Differ* **27**, 2468-2483 (2020).
+
+6. Svoboda P, Franke V, Schultz RM. Sculpting the Transcriptome During the Oocyte-to-Embryo Transition in Mouse. *Curr Top Dev Biol* **113**, 305-349 (2015).
+
+7. Morgan M, *et al.* mRNA 3' uridylation and poly(A) tail length sculpt the mammalian maternal transcriptome. *Nature* **548**, 347-351 (2017).
+
+8. Mole MA, Weberling A, Zernicka-Goetz M. Comparative analysis of human and mouse development: From zygote to pre-gastrulation. *Curr Top Dev Biol* **136**, 113-138 (2020).
+
+9. Timmers HTM, Tora L. Transcript Buffering: A Balancing Act between mRNA Synthesis and mRNA Degradation. *Mol Cell* **72**, 10-17 (2018).
+
+10. Ma JY, *et al.* Maternal factors required for oocyte developmental competence in mice: transcriptome analysis of non-surrounded nucleolus (NSN) and surrounded nucleolus (SN) oocytes. *Cell Cycle* **12**, 1928-1938 (2013).
+
+11. Dumdie JN, *et al.* Chromatin Modification and Global Transcriptional Silencing in the Oocyte Mediated by the mRNA Decay Activator ZFP36L2. *Dev Cell* **44**, 392-402 e397 (2018).
+
+12. Su YQ, *et al.* Selective degradation of transcripts during meiotic maturation of mouse oocytes. *Dev Biol* **302**, 104-117 (2007).
+
+13. Hu L, *et al.* m(6)A RNA modifications are measured at single-base resolution across the mammalian transcriptome. *Nat Biotechnol* **40**, 1210-1219 (2022).
+
+14. Zhao BS, *et al.* m(6)A-dependent maternal mRNA clearance facilitates zebrafish maternal-to-zygotic transition. *Nature* **542**, 475-478 (2017).
+
+15. Arango D, *et al.* Acetylation of Cytidine in mRNA Promotes Translation Efficiency. *Cell* **175**, 1872-1886 e1824 (2018).
+
+16. Liu HY, *et al.* Acetylation of MORC2 by NAT10 regulates cell-cycle checkpoint control and resistance to DNA-damaging chemotherapy and radiotherapy in breast cancer. *Nucleic Acids Res* **48**, 3638-3656 (2020).
+
+17. Zheng J, *et al.* NAT10 regulates mitotic cell fate by acetylating Eg5 to control bipolar spindle assembly and chromosome segregation. *Cell Death Differ* **29**, 846-860 (2022).
+
+18. Shen Q, *et al.* NAT10, a nucleolar protein, localizes to the midbody and regulates cytokinesis and acetylation of microtubules. *Exp Cell Res* **315**, 1653-1667 (2009).
+19. Chen L, et al. NAT10-mediated N4-acetylcytidine modification is required for meiosis entry and progression in male germ cells. Nucleic Acids Res, (2022).
+
+20. Xu J. Preparation, culture, and immortalization of mouse embryonic fibroblasts. Curr Protoc Mol Biol Chapter 28, Unit 28 21 (2005).
+
+21. Jiang X, et al. The Spin1 interactor, Spindoc, is dispensable for meiotic division, but essential for haploid spermatid development in mice. Reprod Biol Endocrinol 19, 144 (2021).
+
+22. Zhang B, et al. Allelic reprogramming of the histone modification H3K4me3 in early mammalian development. Nature 537, 553-557 (2016).
+
+23. Ivanova I, et al. The RNA m(6)A Reader YTHDF2 Is Essential for the Post-transcriptional Regulation of the Maternal Transcriptome and Oocyte Competence. Mol Cell 67, 1059-1067 e1054 (2017).
+
+24. Sharma S, Langhendries JL, Watzinger P, Kotter P, Entian KD, Lafontaine DL. Yeast Kre33 and human NAT10 are conserved 18S rRNA cytosine acetyltransferases that modify tRNAs assisted by the adaptor Tan1/THUMPD1. Nucleic Acids Res 43, 2242-2258 (2015).
+
+25. Liu X, et al. Deacetylation of NAT10 by Sirt1 promotes the transition from rRNA biogenesis to autophagy upon energy stress. Nucleic Acids Res 46, 9601-9616 (2018).
+
+26. Guo J, et al. Oocyte stage-specific effects of MTOR determine granulosa cell fate and oocyte quality in mice. Proc Natl Acad Sci U S A 115, E5326-E5333 (2018).
+
+27. Bolcun-Filas E, Rinaldi VD, White ME, Schimenti JC. Reversal of female infertility by Chk2 ablation reveals the oocyte DNA damage checkpoint pathway. Science 343, 533-536 (2014).
+
+28. Rinaldi VD, Bolcun-Filas E, Kogo H, Kurahashi H, Schimenti JC. The DNA Damage Checkpoint Eliminates Mouse Oocytes with Chromosome Synapsis Failure. Mol Cell 67, 1026-1036 e1022 (2017).
+
+29. MacQueen AJ, Hochwagen A. Checkpoint mechanisms: the puppet masters of meiotic prophase. Trends Cell Biol 21, 393-400 (2011).
+
+30. Cloutier JM, Mahadevaiah SK, Ellnati E, Nussenzweig A, Toth A, Turner JM. Histone H2AFX Links Meiotic Chromosome Asynapsis to Prophase I Oocyte Loss in Mammals. PLoS Genet 11, e1005462 (2015).
+
+31. Lan ZJ, Xu X, Cooney AJ. Differential oocyte-specific expression of Cre recombinase activity in GDF-9-iCre, Zp3cre, and Msx2Cre transgenic mice. Biol Reprod 71, 1469-1474 (2004).
+
+32. Pan H, O'Brien MJ, Wigglesworth K, Eppig JJ, Schultz RM. Transcript profiling during mouse oocyte development and the effect of gonadotropin priming and development in vitro. Dev Biol 286, 493-506 (2005).
+
+33. Sha QQ, et al. CNOT6L couples the selective degradation of maternal transcripts to meiotic cell cycle progression in mouse oocyte. EMBO J 37, (2018).
+
+34. Wu Y, et al. N(6)-methyladenosine regulates maternal RNA maintenance in oocytes and timely RNA decay during mouse maternal-to-zygotic transition. Nat Cell Biol 24, 917-927 (2022).
+
+35. Ma J, Flemr M, Strnad H, Svoboda P, Schultz RM. Maternally recruited DCP1A and DCP2 contribute to messenger RNA degradation during oocyte maturation and genome activation in mouse. Biol Reprod 88, 11 (2013).
+
+36. Li L, Lu X, Dean J. The maternal to zygotic transition in mammals. Mol Aspects Med 34, 919-938 (2013).
+37. Reyes JM, Ross PJ. Cytoplasmic polyadenylation in mammalian oocyte maturation. Wiley Interdiscip Rev RNA 7, 71-89 (2016).
+
+38. Tora L, Vincent SD. What defines the maternal transcriptome? Biochem Soc Trans 49, 2051-2062 (2021).
+
+39. Yu C, et al. BTG4 is a meiotic cell cycle-coupled maternal-zygotic-transition licensing factor in oocytes. Nat Struct Mol Biol 23, 387-394 (2016).
+
+40. Horvat F, et al. Role of Cnot6l in maternal mRNA turnover. Life Sci Alliance 1, e201800084 (2018).
+
+41. Ma J, Fukuda Y, Schultz RM. Mobilization of Dormant Cnot7 mRNA Promotes Deadenylation of Maternal Transcripts During Mouse Oocyte Maturation. Biol Reprod 93, 48 (2015).
+
+42. Janicke A, Vancuylenberg J, Boag PR, Traven A, Beilharz TH. ePAT: a simple method to tag adenylated RNA to measure poly(A)-tail length and other 3' RACE applications. RNA 18, 1289-1295 (2012).
+
+43. Salles FJ, Strickland S. Rapid and sensitive analysis of mRNA polyadenylation states by PCR. PCR Methods Appl 4, 317-321 (1995).
+
+44. Bilska A, Krawczyk PS, Dziembowski A, Mroczek S. Measuring the tail: Methods for poly(A) tail profiling. Wiley Interdiscip Rev RNA, e1737 (2022).
+
+45. Ito S, et al. Human NAT10 is an ATP-dependent RNA acetyltransferase responsible for N4-acetylcytidine formation in 18 S ribosomal RNA (rRNA). J Biol Chem 289, 35724-35730 (2014).
+
+46. Luong XG, Daldello EM, Rajkovic G, Yang CR, Conti M. Genome-wide analysis reveals a switch in the translational program upon oocyte meiotic resumption. Nucleic Acids Res 48, 3257-3276 (2020).
+
+47. Larrieu D, Britton S, Demir M, Rodriguez R, Jackson SP. Chemical inhibition of NAT10 corrects defects of laminopathic cells. Science 344, 527-532 (2014).
+
+48. Tan TZ, et al. Functional genomics identifies five distinct molecular subtypes with clinical relevance and pathways for growth control in epithelial ovarian cancer. EMBO Mol Med 5, 1051-1066 (2013).
+
+49. Xiang Y, et al. NAT10-Mediated N4-Acetylcytidine of RNA Contributes to Post-transcriptional Regulation of Mouse Oocyte Maturation in vitro. Front Cell Dev Biol 9, 704341 (2021).
+
+50. Poh HX, Mirza AH, Pickering BF, Jaffrey SR. Alternative splicing of METTL3 explains apparently METTL3-independent m6A modifications in mRNA. PLoS Biol 20, e3001683 (2022).
+
+51. Sui X, et al. METTL3-mediated m(6)A is required for murine oocyte maturation and maternal-to-zygotic transition. Cell Cycle 19, 391-404 (2020).
+
+52. Liu P, et al. m(6)A-independent genome-wide METTL3 and METTL14 redistribution drives the senescence-associated secretory phenotype. Nat Cell Biol 23, 355-365 (2021).
+
+53. Saito Y, et al. YTHDC2 control of gametogenesis requires helicase activity but not m(6)A binding. Genes Dev 36, 180-194 (2022).
+
+54. Su R, et al. METTL16 exerts an m(6)A-independent function to facilitate translation and tumorigenesis. Nat Cell Biol 24, 205-216 (2022).
+
+55. Helm M, Motorin Y. Detecting RNA modifications in the epitranscriptome: predict and validate. Nat Rev Genet 18, 275-291 (2017).
+56. Wiener D, Schwartz S. The epitranscriptome beyond m(6)A. Nat Rev Genet **22**, 119-131 (2021).
+
+57. Arango D, Sturgill D, Oberdoerffer S. Immunoprecipitation and Sequencing of Acetylated RNA. *Bio Protoc* **9**, e3278 (2019).
+
+58. Sas-Chen A, *et al.* Dynamic RNA acetylation revealed by quantitative cross-evolutionary mapping. *Nature* **583**, 638-643 (2020).
+
+59. Thomas JM, Bryson KM, Meier JL. Nucleotide resolution sequencing of N4-acetylcytidine in RNA. *Methods Enzymol* **621**, 31-51 (2019).
+
+60. Liu H, Ling Y, Gong Y, Sun Y, Hou L, Zhang B. DNA damage induces N-acetyltransferase NAT10 gene expression through transcriptional activation. *Mol Cell Biochem* **300**, 249-258 (2007).
+Acknowledgements
+We are grateful to Prof. Li He (School of Basic Medical Sciences, Division of Life Sciences and Medicine, University of Science and Technology of China (USTC), Hefei, Anhui) for generous assistance with oocyte imaging. We also thank all members of Bao laboratory for helpful discussion. This work was supported by grants from the Ministry of Science and Technology of China (2019YFA0802600); National Natural Science Foundation of China (31970793, 32170856); The open project of NHC Key Laboratory of Male Reproduction and Genetics (No. KF201901); the Fundamental Research Funds for the Central Universities" (WK9110000181, WK2070000156) and Startup funding (KY9100000001) from USTC.
+
+Author contributions
+J.Q.B., X.L.Z and W.B.Q. conceived, designed, and supervised the work. J.Q.B. wrote the manuscript. X.J. performed mouse crossing, chromosome spreads analysis and immunofluorescent staining. Y. C. collected equal amounts of mouse oocytes used in this study and western blot analysis. Y. Z. Z. did polyA-tail length assay. C.L.X analyzed the RNA-seq data and generated the figures. Q.D.L. collected mouse oocytes used for RNA-seq. X.M.X. did HE staining of mouse ovaries with the help of W.Q.L. J.Q.Z. and L.M. performed mouse genotyping. M.A. and Y.Z.C. helped with the immunofluorescence.
+
+Supplementary Data
+Supplementary Fig. S1~8; Supplementary Table S1~5
+
+Competing interests
+The authors declare no competing interests.
+
+Correspondence
+Correspondence and requests for materials should be addressed to Weibing Qin, Xiaoli Zhu or Jianqiang Bao.
+Fig.1. Expression and Localization of Nat10 enriched in the nucleolus of oocytes in mice. (A) Western blot showing the relative expression levels of NAT10 protein among multiple organs in adult WT mice. GAPDH served as a loading control. (B) Dynamic mRNA expression levels of Nat10 from RNA-seq analyses in oocytes and preimplantation embryos in mice (GSE71434). ICM, Inner cell mass. (C) Quantitative RT-PCR results showing the relative expression levels of mouse Nat10 mRNA in oocytes, and preimplantation embryos. Data were presented as mean± SEM, n=3. GO, growing oocytes collected from postnatal 14-day-old (P14) female mice. (D) Immunofluorescence (IF) staining of NAT10 in growing (GO), GV, MI, and MII oocytes as indicated. Dashed circle indicates cellular membrane of oocytes. DNA was counterstained with 4',6-diamidino-2-phenylindole (DAPI). Scale bar, 20 μm. (E) IF images of 21-day-old WT ovarian cryosections stained with anti-NAT10 antibody (Red) and DAPI (Blue) for follicles at various stages (primordial, primary, secondary, early antral, and antral stages) as
+indicated. Scale bar, 20 \( \mu \)m. Boxed insert area is a magnified view of the oocyte in the respective follicles. Arrows point to NAT10-positive nucleus of the oocyte in mouse developing follicles. (F) IF images of 21-day-old WT ovarian sections co-stained with NAT10 antibody (Red), Nucleophosmin (NPM, Green) and DAPI (Blue) in the follicles as indicated. Scale bar, 20 \( \mu \)m. Bottom panel is a magnified view of the oocyte in the respective follicles. Arrows point to the co-localization of NAT10 and NPM in the oocyte nucleolus at varied stages of follicles.
+Fig.2. Pre-meiotic deletion of Nat10 caused follicular developmental arrest and premature ovarian failure (POF). (A) Schematic diagram showing the landmark timeline of oocyte development from embryonic meiotic cell-cycle progression to postnatal oocyte growth and maturation. Stra8-GFPcre is activated prior to Embryonic day 13.5 (E13.5); Zp3-cre is active starting from P5 in the primary follicles; Both Ubc-CreERT2 and Ddx4-CreERT2 lines are Cre-inducible in all tissues and specifically in the germline, respectively, upon tamoxifen injection. (B) A breeding scheme by crossing Nat10lox/lox with
+Stra8-GFPCre to generate Nat10lox/-; Stra8-GFPCre (Nat10-ScKO) offspring. (C) Western blot analyses of the NAT10 protein levels in adult WT and Nat10-ScKO ovary. α-TUBULIN was used as a loading control. (D) Quantitative RT–PCR (qPCR) assay showing the relative expression levels of Nat10 mRNA in adult WT and Nat10-ScKO mouse ovary. Data are presented as mean± SEM, n=3. ****, p<0.0001 by two-tailed Student’s t-test. (E) Fertility test showing the cumulative average numbers of pups from breeding of WT and Nat10-ScKO females with WT males. Data are presented as the mean± SEM, n=5, ****, p<0.0001 by two-tailed Student’s t-test. (F) The gross morphology of ovaries derived from WT and Nat10-ScKO mice at 1M. Scale bar, 200 μm. (G) H&E staining of paraffin-embedded ovarian sections showing the histology of WT and Nat10-ScKO ovaries at postnatal days as indicated. Scale bar, 200 μm. High-resolution view of the boxed area is shown in parallel. Scale bar, 20 μm. Arrows point to follicles at stages as indicated. PrF, Primordial Follicle; PF, Primary Follicle; SF, Secondary Follicle; EAF, Early Antral Follicle; AF, Antral Follicle; LAF, Late Antral Follicle; (H) qPCR analyses of the relative expression levels for a cohort of genes showing specific or characteristic expression in ovarian granulosa cells (Left) or testicular Sertoli/Leydig cells (Right) in 1-month-old WT and Nat10-ScKO ovaries. Data are presented as the mean±SEM, n=3.
+Fig.3. Embryonic Nat10 loss caused oocyte meiotic prophase I arrest at pachytene stage owing to deficient DSB repair. (A) Immunofluorescence staining of oocyte nuclear chromosome spreads by SYCP3 and SYCP1 markers in WT and Nat10-ScKO mouse ovaries at birth. Scale bar, 10 μm. Arrows point to the asynapsed structure of the lateral and central axes. (B) The statistic counts showing the percentage of oocytes at various stages as indicated. Data are presented as the mean± SEM, n=3, *, p<0.05 by two-tailed Student’s t-test. (C) IF staining by SYCP3 and γH2AX on surface-spread oocytes at pachytene and diplotene stages from WT and Nat10-ScKO mouse oocytes at birth. Scale bar, 10
+μm. (D) The statistic counts showing the relative γH2AX signal intensity calculated by ImageJ in pachytene and diplotene oocytes. Data are presented as mean± SEM, n=3; ****, p<0.0001; n.s., not significant by two-tailed Student’s t-test. (E) IF staining on surface-spread oocytes by SYCP3 and RPA2 in WT and Nat10-ScKO mouse oocytes at birth. Scale bar, 10 μm. (F) Quantification of the numbers of RPA2 foci (representative of the unrepaired DSBs) in WT and Nat10-ScKO mouse oocytes at birth. Zyg, Zygote; Pac, Pachytene; Pac-like, Pachytene-like; Dip, Diplotene. Data are presented as the mean± SEM, n=3; *, p <0.05. n.s., not significant by two-tailed Student’s t-test.
+Fig.4. Postnatal Nat10 depletion caused ovarian developmental arrest at secondary follicles. (A) A breeding scheme for Nat10 KO in growing oocytes of primary follicles by crossing Nat10lox/lox with Zp3-Cre deleter to attain Nat10lox/-; Zp3-Cre (Nat10-ZcKO) female offspring. (B-C) Immunofluorescence staining by NAT10 (Red), NPM (Green) and Hoechst 33342(Blue) in the secondary follicles (B) and GV oocytes (C) from WT and Nat10-ZcKO ovaries. Scale bar, 20 μm. (D) Fertility test showing the cumulative numbers of pups from breedings of WT and Nat10-ZcKO females with WT males during a half-year caging. Data are presented as the mean± SEM, n=5; ****, p<0.0001 by two-tailed Student’s t-test. (E) The gross morphology of ovaries derived from WT and Nat10-ZcKO mice at age of 1 month (M) (left) and 2 months (right). Scale bar, 200 μm. (F) H&E staining showing ovarian histology from WT and Nat10-ZcKO mice at 1 M (Top) and 2 M (Bottom). Scale bar, 50 μm. Follicles are indicated by arrows. (G) Comparison of the average numbers of follicles at indicated stages in the ovaries of WT and Nat10-ZcKO mice at 1M (Top) and 2M (Bottom). Follicles were counted on serial ovarian sections
+after H&E staining. Data are presented as the mean± SEM, n=3; *, p<0.05; **, p<0.01; ***, p<0.001 by two-tailed Student’s t-test. PO, Preovulatory Follicle.
+Fig.5. Postnatal Nat10 deficiency impedes oocyte chromatin NSN–SN configuration transition.
+(A) The gross morphology of oocytes at Germinal vesicle (GV) stage collected from PMSG-primed WT and Nat10-ZcKO females at P21. Scale bar, 100 μm. (B) Quantification of collected average numbers of GV oocytes. Data are presented as the mean± SEM, n=4. n.s., non-significant by two-tailed Student’s t-test. (C) Hoechst 33342 (Blue) staining of the GV oocytes with non-surrounded nucleolus (NSN) and surrounded nucleolus (SN) chromatin configurations in WT and Nat10-ZcKO oocytes. Scale bar, 20 μm. (D) The percentage of NSN-type and SN-type oocytes isolated from WT and Nat10-ZcKO mice at P21. Data are presented as the mean± SEM, n=3. **, p<0.01 by two-tailed Student’s t-test. (E-F) Immunofluorescence staining by H3K4me3 in NSN-type (Left) and SN-type (Right) oocytes from PMSG-primed WT and Nat10-ZcKO mice (E), and quantification of H3K4me3 intensity (F). Scale bar, 10 μm. Data are presented as the mean± SEM, n=3; ***, p<0.001 by two-tailed Student’s t-test. (G-H) Immunofluorescence staining by H3K9me3 in NSN-type (Left) and SN-type (Right) oocytes from PMSG-primed WT and Nat10-ZcKO mice (G), and quantification of H3K9me3 intensity (H). Scale bar, 10 μm. Data are presented as the mean± SEM, n=3; *, p<0.05 by two-tailed Student’s t-test.
+Fig.6. Postnatal Nat10 ablation led to defective oocyte meiotic maturation. (A) The gross morphology of oocytes collected at the time points as indicated for GV (0h), and cultured in vitro for MI
+(6h) and MII (16h) from PMSG-primed WT and Nat10-ZcKO females. Scale bar, 100 μm. (B-C) Percentage of oocytes at GVBD (B) and MII (C) after release of GV oocytes cultured in IBMX-containing medium from PMSG-primed WT and Nat10-ZcKO females. Data are presented as mean± SEM, n=3. **, p<0.01, ***, p<0.001 by two-tailed Student’s t-test. (D) Average numbers of superovulated oocytes at MII from WT (30.57±0.92) and Nat10-ZcKO (3.8±1.15) mice following PMSG and hCG injection in vivo. Data are presented as the mean± SEM, n=5. ***, p<0.001 by two-tailed Student’s t-test. (E-F) Immunofluorescence staining images of superovulated oocytes collected at 16h after hCG injection by α-TUBULIN staining. Oocytes with MI arrest, Anaphase-to-telophase arrest in prophase I (AI-TI), and aberrant spindles were observed (E) and counted (F) in Nat10-ZcKO mice. Scale bar, 40μm. Data are presented as the mean± SEM, n=3. ***, p<0.001 by two-tailed Student’s t-test. (G-H) Representative gross morphology of preimplantation embryos at various stages as indicated derived from superovulated WT and Nat10-ZcKO oocytes (after hCG priming) fertilized with WT sperm (G). Arrows point to the blastocysts; Quantitative comparison of the average numbers of preimplantation embryos at varied stages was shown (H). Scale bar, 100 μm. Data are presented as the mean± SEM, n=5. ****, p<0.0001 by two-tailed Student’s t-test.
+Fig.7. Mini-bulk SMART-seq2 analyses identified the dysregulated maternal transcriptome in Nat10-ZcKO oocytes. (A) A diagram showing mouse oocyte samples collected for mini-bulk SMART-seq2 analyses. (B) Bar graph showing the numbers of transcripts detected in WT and Nat10-ZcKO
+oocytes at GV and MII stages (TPM \( \geq 1 \)). Data are presented as the mean± SEM, n=3. n.s., non-significant by two-tailed Student's t-test. (C) Scatter plot of mini-bulk SMART-seq2 data showing differentially expressed genes (DEGs) in Nat10-ZcKO MII oocytes. Red color: Up-regulated; Blue color: Down-regulated; Cutoff: fold change (FC) \( \geq 2 \), adjusted p<0.05. The TPMs of WT and Nat10-ZcKO MII oocytes are listed in Supplementary Table S5. (D) Gene Ontology (GO) enrichment analysis of up-regulated genes in Nat10-ZcKO MII oocytes (Cutoff: FC \( \geq 2 \), adjusted p<0.05). (E) Heatmap of representative genes from four major functional GO categories showing up-regulated expression in Nat10-ScKO MII oocytes. The color intensity gradient from red to blue indicates the relative gene expression levels from high to low. (F) Bar plots showing the qPCR analyses of relative mRNA expression levels for a panel of up-regulated genes identified by mini-bulk SMART-seq2 in WT and Nat10-ZcKO MII oocytes. (G-H) Box plots showing the relative expression levels of the transcripts in WT and Nat10-ZcKO oocytes at the GV and MII stages as indicated (G); box plot showing the relative fold changes in mRNA levels in MII versus GV oocytes from WT and Nat10-ZcKO, respectively (H). The box indicates the upper and lower quartiles; the thick line in the box indicates the median, n = 3. P-values by a two-tailed Student's t-test are indicated. (I) Box plot showing gene expression levels in WT and Nat10-ZcKO oocytes at the GV and MII stages. Genes were divided into 5 bins according to their relative expression abundance in the WT MII oocytes. The box indicates upper and lower quartiles, n=3. (J) Sankey diagram showing the overlapping of the DEGs (1196 up-regulated vs 555 down-regulated) with genes exhibiting up- [FC(MII/GV) \( \geq 2 \), p<0.05], down- [FC(MII/GV) \( \leq -2 \), p<0.05], or stable expression patterns in WT MII relative to GV stage oocytes (TPM \( \geq 1 \)). (K) Venn diagram showing the overlapping of down-regulated transcripts between WT MII oocytes relative to GV oocytes (2011, Cutoff: TPM \( \geq 1 \), FC[GV/MII] \( \geq 5 \)), and Nat10-ZcKO MII oocytes relative to GV oocytes (1206, Cutoff: TPM \( \geq 1 \), FC[GV/MII] \( \geq 5 \)). The total 595 overlapping transcripts represent those that were concurrently down-regulated in both WT MII and Nat10-ZcKO MII oocytes. In other words, they were degraded regardless of Nat10 presence (Nat10-unrelated). (L) Degradation trend patterns of mouse maternal transcripts during the GV-MII transition in WT and Nat10-ZcKO oocytes. Each light-yellow line represents the expression levels of one gene, and the middle blue and red lines represent the median expression levels in WT and Nat10-ZcKO, respectively. Transcripts with TPM \( \geq 1 \) at the GV stage were selected and analyzed.
+Fig.8. Hairpin Adaptor-PolyA Tail length (HA-PAT) assay validated the deficient maternal mRNA decay in Nat10-ZcKO MII oocytes. (A) Venn diagram showing the overlapping of transcripts that were stabilized during GV-to-MII transition in Cnot6l-/- and Nat10-ZcKO MII oocytes (FC=[WT MII/Nat10-ZcKO MII] ≥ 2, p<0.05). (B) Fold change of relative expression levels of transcripts encoding ribosomal protein subunits in Nat10-ZcKO relative to WT oocytes at MII stage. The values of log2(FC[Nat10-ZcKO/WT]) are listed on the right column. (C) qPCR results showing the relative levels of indicated transcripts (Cnot6l, Cnot7 and Btg4) in WT and Nat10-ZcKO oocytes at MII stage. Data are presented as the mean±SEM, n=3. ****, p<0.0001 by two-tailed Student’s t-test. (D) A schematic illustration depicting the design strategy and the key steps for Hairpin Adaptor-PolyA Tail length (HA-PAT) assay. The 1st strand of cDNA was synthesized with the hairpin adaptor (HA) primer in conjunction with a P5TSO primer containing three “G”, via a mechanism of “template-switching”. GSP, Gene-specific primer; A0, the PCR product resulting from the amplification with a gene-specific pair of GSPxF and
+GSPxR primers; GSPxR primer was designed against an mRNA's 3' terminal preceding the polyA sequence; polyA-containing PCR products were amplified with GSPxF and a fixed HAPrimerR primers. The full sequence for hairpin adaptor (HA) is listed at the bottom. W indicates degenerate nucleotides (A or T); * The asterisk indicates the phosphorothioate modification. (E-F) HA-PAT assay results showing changes in poly(A)-tail lengths of indicated transcripts in WT and Nat10-ZcKO oocytes at GV and MII stages. Experiments were performed in triplicates; a representative image is shown in the 2% agarose gel (E) and the length distribution shown in the densitometric curves (F).
+Fig.9. Nat10 is translationally required for cell cycle progression. (A) Schematic illustration showing the procedures for generation of two stable, TMX-inducible MEF cell lines (Nat10+/+Cre; lox/lox) from the Ubc-cre; Nat10lox/lox embryos following a 3T3 protocol (see Methods and Materials). Inducible Nat10 KO was achieved by 2 μM OHT treatment for three consecutive days. Rescue of Nat10 was attained by overexpression (OE) of WT Nat10 ORF plasmid. (B) Western blot displaying the NAT10 and PCNA protein levels in WT and OHT (Ubc-cre; Nat10lox/lox) MEF line. α-TUBULIN was used as a loading control. (C) Quantitative RT–PCR results showing the relative expression levels of Nat10 in WT and OHT MEF line. Data are presented as mean± SEM, n=3. ****, p<0.0001 by two-tailed Student’s t-test. (D) Cell cycle analysis by flow cytometry between WT and OHT MEF cells. G1, p=0.0083; S, p=0.0017; G2/M, p<0.0001; (E) Annexin V apoptosis detection of WT and OHT MEF cells. Q4, Viable cells, p= 0.2983; Q3, Early apoptosis, p= 0.2147; Q2, Late apoptosis, p= 0.3122; Q1, Necrotic cells, p=
+0.7863. (F) CCK8 assay showing the cell proliferation rates among WT, OHT treatment and OHT plus Nat10 overexpression (OE) groups. (G-H) Comparison of cell proliferation as visualized by Ki67 labelling (G) and quantification (H) among WT (empty vector), OHT treatment and Nat10 overexpression groups. Scale bar, 50μm. Data are presented as mean±SEM, n=3. *, p<0.05, **, p<0.01 by two-tailed Student’s t-test. (I-J) The polysome profiling displaying the translational efficiency and ribosome assembly in MEF cells analyzed by sucrose density gradient centrifugation. The graphic curves showed the polysome profiles of MEF cells treated by mock (Blue) or OHT (Red) (I). Comparison of the ratios of 60S to 40S (Left) and of 80S to 40S (Right) in MEF cells with mock treatment (Blue) or OHT (Red) (J). Data are presented as mean±SEM, n=3, ***, p<0.001 by two-tailed Student’s t-test.
+Supplementary Files
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+• SupplementaryTableS4.DEGsbetweenWTandNat10ZcKOGVoocytes.csv
+• SupplementaryTableS5.DEGsbetweenWTandNat10ZcKOMIloocytes.xlsx
+• Supplementarymaterials.SupplementaryFigureS18andSupplementaryTablesS13.pdf

1014f2957f8a5a80f5a91706c1899e6dcb17ad8f638e3ae989055850203d6b05/peer_review/peer_review.md

@@ -0,0 +1,457 @@
+Peer Review File
+Ultrastiff metamaterials generated through a multilayer strategy and topology optimization
+
+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/.
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+This manuscript provides a timely account of a significant breakthrough in architected metamaterials, unveiling novel mechanical properties characterized by remarkable stiffness and energy absorption capabilities. The robust numerical computations establish a firm groundwork for the proposed concept, while the integration of 3D printing technology further reinforces these findings. The meticulously designed methodology, coupled with the emergence of novel phenomena, underpinned by a compelling argument, has unequivocally convinced me of the urgency of publishing this manuscript in a rapid manner, thereby offering valuable insights to the broader research community. Nonetheless, before final acceptance, I kindly request the authors to address the following raised queries:
+
+(1) In Figure 3, a noticeable disparity between simulations and experimental results is evident. The authors should provide a detailed explanation for this variance, shedding light on the underlying reasons.
+
+(2) The authors have delved into the topic of energy absorption in the proposed design. However, it has been observed that the structures were compressed by less than 10%, which might be insufficient for a comprehensive evaluation of energy absorption. An elaboration on this limitation is necessary.
+
+(3) Figure 5 presents data on energy absorption in the proposed designs. To enhance the comprehensibility of the findings, the authors are encouraged to include a discussion on the mechanisms responsible for energy dissipation, as well as the evolution of deformation patterns. This additional insight will provide a more thorough understanding of the results.
+
+Reviewer #2 (Remarks to the Author):
+
+The authors propose a novel topology optimization method for designing multilayered mechanical structures. Some experimental results suggest that the proposed multilayered designs are superior to non-optimized structures in terms of static and dynamic mechanical properties. I find the manuscript to be well-written and interesting to a wide audience. However, the authors should address the following comments before I recommend the manuscript for publication in Nature Communications:
+
+1. Although I understand that the manuscript focuses on cellular lattice structures, I think that the authors should explain the advantages of the proposed multilayer strategy compared to standard optimized designs. Specifically, why do the authors need the multilayer strategy to increase the design
+flexibility? For example, to the best of my knowledge, the standard and conventional SIMP method provides maximum design flexibility.
+
+2. In Figure 2a, the left-most Opt-IWP performs worse than the non-optimized IWP. Please add a discussion on this anomaly.
+
+3. I do not comprehend how to interpret Figure 3b and Figure 4c, 4d. The authors present the experimental results for each model (P set and Opt-P set) with four different printing materials. However, the numerical results are presented for only one case. In this numerical simulation, which printing material is assumed? From these results, I do not observe any consistency between the experimental and numerical results, despite the authors claiming that “… shows much consistency with corresponding numerical results, yet …”.
+
+4. In Supplementary information, please add some references or more detailed descriptions on the material characterization. For instance, how the effective elastic tensor, Zener's ratio, etc. are derived or determined.
+
+5. There are typos in the section title and Eq.(8) in Supplementary information.
+Response to the comments of the reviewers
+
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+This manuscript provides a timely account of a significant breakthrough in architected metamaterials, unveiling novel mechanical properties characterized by remarkable stiffness and energy absorption capabilities. The robust numerical computations establish a firm groundwork for the proposed concept, while the integration of 3D printing technology further reinforces these findings. The meticulously designed methodology, coupled with the emergence of novel phenomena, underpinned by a compelling argument, has unequivocally convinced me of the urgency of publishing this manuscript in a rapid manner, thereby offering valuable insights to the broader research community. Nonetheless, before final acceptance, I kindly request the authors to address the following raised queries:
+
+(1) In Figure 3, a noticeable disparity between simulations and experimental results is evident. The authors should provide a detailed explanation for this variance, shedding light on the underlying reasons.
+
+(2) The authors have delved into the topic of energy absorption in the proposed design. However, it has been observed that the structures were compressed by less than 10%, which might be insufficient for a comprehensive evaluation of energy absorption. An elaboration on this limitation is necessary.
+
+(3) Figure 5 presents data on energy absorption in the proposed designs. To enhance the comprehensibility of the findings, the authors are encouraged to include a discussion on the mechanisms responsible for energy dissipation, as well as the evolution of deformation patterns. This additional insight will provide a more thorough understanding of the results.
+
+Reviewer #2 (Remarks to the Author):
+
+The authors propose a novel topology optimization method for designing multilayered mechanical structures. Some experimental results suggest that the proposed multilayered designs are superior to non-optimized structures in terms of static and dynamic mechanical properties. I find the manuscript to be well-written and interesting to a wide audience. However, the authors should address the following comments before I recommend the manuscript for publication in Nature Communications:
+
+(1) Although I understand that the manuscript focuses on cellular lattice structures, I think that the authors should explain the advantages of the proposed multilayer strategy compared to standard optimized designs. Specifically, why do the authors need the multilayer strategy to increase the design flexibility? For example, to the best of my knowledge, the standard and conventional SIMP method provides maximum design flexibility.
+
+(2) In Figure 2a, the left-most Opt-IWP performs worse than the non-optimized IWP. Please add a discussion on this anomaly.
+
+(3) I do not comprehend how to interpret Figure 3b and Figure 4c, 4d. The authors present the experimental results for each model (P set and Opt-P set) with four different printing materials. However, the numerical results are presented for only one case. In this numerical simulation, which printing material is assumed? From these results, I do not observe any consistency between the experimental and numerical results, despite the authors claiming that "... shows much consistency with corresponding numerical results, yet ..." .
+(4) In Supplementary information, please add some references or more detailed descriptions on the material characterization. For instance, how the effective elastic tensor, Zener's ratio, etc. are derived or determined.
+
+(5) There are typos in the section title and Eq.(8) in Supplementary information.
+Detailed response to the Reviewers
+
+The authors are grateful to the editors and reviewers for your time and effort spent on reviewing our manuscript. We would like to thank you very much for your recognition to the manuscript, and your constructive and insightful comments. The comments are very helpful for improving the quality of the manuscript. All comments and questions have been carefully taken into consideration, and we have revised our manuscript accordingly. The amendments on the paper are indicated in red. The specific responses to each point raised by the reviewers are itemized below.
+
+Reviewer #1 (Remarks to the Author):
+
+This manuscript provides a timely account of a significant breakthrough in architected metamaterials, unveiling novel mechanical properties characterized by remarkable stiffness and energy absorption capabilities. The robust numerical computations establish a firm groundwork for the proposed concept, while the integration of 3D printing technology further reinforces these findings. The meticulously designed methodology, coupled with the emergence of novel phenomena, underpinned by a compelling argument, has unequivocally convinced me of the urgency of publishing this manuscript in a rapid manner, thereby offering valuable insights to the broader research community. Nonetheless, before final acceptance, I kindly request the authors to address the following raised queries:
+
+Response: The authors would like to thank the reviewer for the comprehensive and supportive comments which give us great encouragement. We have tried our best to address these issues one by one as follows.
+
+Comment (1): In Figure 3, a noticeable disparity between simulations and experimental results is evident. The authors should provide a detailed explanation for this variance, shedding light on the underlying reasons.
+
+Response: Thanks for the suggestion. In the physical experimental part, we aimed to test the mechanical properties of the printed models. We only care about the relative proportional ratios of Young's moduli of the P set and the Opt-P set, which means that the experimental results should be material-independent. Thus, we chose four different printing materials to verify the simulation results. In our numerical simulation, the material iron is used (Young's modulus: 210GPa, yield strength: 400MPa), and the material properties are verified using both static (Fig. R1a) and dynamic methods (Fig. R1b). Here, different printing materials have different mechanical properties, plastic, brittle, or quasi-brittle. The yield strength can vary for different materials. Therefore, we only consider the effective Young's moduli of the printed models with the different materials, i.e., the stiffness, which can be obtained by calculating the slope of the linear part of the constitutional engineering stress-strain curve (Fig. R2). Table R1 summarizes the effective Young's moduli of the printed models. Table R2 presents the percentages of effective Young's moduli of the printed models
+
+![Material property verification. a, static method. b, dynamic method.](page_186_1047_1077_355.png)
+
+Fig. R1 | Material property verification. a, static method. b, dynamic method.
+Fig. R2 | Linear fit for effective Young's modulus with different materials. a, TPU. b, PA12. c, SS316. d, AlSi10Mg.
+
+normalized by the result of P-1. Different printing materials may result in different normalized values, but the tendency is consistent with the simulation results. As can be seen in Table R2, the proportional ratios among the P set and Opt-P set for the non-metal model (TPU and Nylon (PA12)) show larger values as a whole, but the variation tendency compared with the normalized P-1 is consistent. It seems that the variance between the experiment and the simulation is a proportional factor. For example, if the effective Young's moduli of the TPU result (except P-1) multiply a proportional factor of 75.3%, we obtain results very close to the numerical simulation results (Table R3). If the effective Young's moduli of the PA12 result (except P-1) multiply a proportional factor of 52.3%, we obtain results very close to the numerical simulation results (Table R3). For the metal result (SS316, AlSi10Mg), the proportional ratios show smaller values for P-4, -5 and Opt-P-4, -5. The variances might be caused mainly by printing technics and prototyping quality [1]. For the metal printing technology that requires high temperatures, the residue stress from cooling down from high temperatures may lead to many small defects and cracks, as a result, weakening the mechanical properties of the printed model [1]. For cell models, the local printing quality may affect the experimental results to some extent. For the scale of the 4X4X4 case, the local defects can be alleviated significantly (Fig. R3). For the PA12 model, the printed scale is 120mmX120mmX120mm. Such a length scale can ensure better printing quality (Fig. R4a). As a result, the experimental results are consistent with corresponding simulation results (Table R4). The FBC-Opt-P-2 model is an exception because the mass loss (around 27.4g (17.8%) compared with the P-1) is significant after removing the support
+Fig. R3 | Linear fit for effective Young's modulus with different materials. a, PA12 (FBC). b, PA12 (PBC). c, AlSi10Mg (FBC). d, AlSi10Mg (PBC).
+
+structures and powder remainder. For the AlSi10Mg model, since large-size printing for metal models can be very expensive, we printed the metal models with the scale of 60mmX60mmX60mm (Fig. R4b), and the printing quality is relatively lower (large-size printing for metal models can be very expensive). Again, the metal results (AlSi10Mg) of Opt-P-4 and -5 cannot reach the simulation level due to the reason of printing technics itself and prototyping quality (Table R4).
+
+We have revised the ‘Results’ part in the main text, and the ‘Supplementary Note 6’ in SI as follows,
+• ‘… uniaxial stiffness in linear elasticity …’
+• ‘… Thus, the relative proportional ratios of Young’s moduli of the P set and Opt-P set should be material-independent. In comparison with the numerical simulations (Figure 3b), the basic trend of the physical experimental results shows consistency with corresponding numerical results, yet still some minor differences remain for different kinds of printing material. The discrepancies are mainly caused by printing technics and prototyping quality (detailed discussion in Supplementary Note 6 in SI) …’
+• ‘… Supplementary Note 6: Disparity between experiment and simulation …’
+Table R1 Numerical and experimental Young’s moduli comparison for fabricated cell models (MPa)
+
+<table>
+  <tr>
+    <th>Model</th>
+    <th>Numerical simulation</th>
+    <th>TPU</th>
+    <th>PA12</th>
+    <th>SS316</th>
+    <th>AISi10Mg</th>
+  </tr>
+  <tr>
+    <td>P-1</td>
+    <td>813.24</td>
+    <td>0.064</td>
+    <td>1.83</td>
+    <td>320.40</td>
+    <td>176.00</td>
+  </tr>
+  <tr>
+    <td>P-2</td>
+    <td>465.46</td>
+    <td>0.039</td>
+    <td>1.79</td>
+    <td>258.05</td>
+    <td>103.83</td>
+  </tr>
+  <tr>
+    <td>P-4</td>
+    <td>8087.7</td>
+    <td>0.98</td>
+    <td>34.8</td>
+    <td>1403.60</td>
+    <td>993.67</td>
+  </tr>
+  <tr>
+    <td>P-5</td>
+    <td>6842.8</td>
+    <td>0.69</td>
+    <td>31.5</td>
+    <td>1574.31</td>
+    <td>781.48</td>
+  </tr>
+  <tr>
+    <td>Opt-P-1</td>
+    <td>1880.3</td>
+    <td>0.21</td>
+    <td>8.85</td>
+    <td>634.24</td>
+    <td>404.19</td>
+  </tr>
+  <tr>
+    <td>Opt-P-2</td>
+    <td>917.09</td>
+    <td>0.085</td>
+    <td>3.65</td>
+    <td>431.42</td>
+    <td>212.25</td>
+  </tr>
+  <tr>
+    <td>Opt-P-4</td>
+    <td>12607</td>
+    <td>1.65</td>
+    <td>59.35</td>
+    <td>1938.88</td>
+    <td>1472.81</td>
+  </tr>
+  <tr>
+    <td>Opt-P-5</td>
+    <td>10274</td>
+    <td>1.09</td>
+    <td>45.8</td>
+    <td>2001.96</td>
+    <td>1350.25</td>
+  </tr>
+</table>
+
+Table R2 Normalized Young’s moduli comparison for fabricated cell models
+
+<table>
+  <tr>
+    <th>Model</th>
+    <th>Numerical simulation</th>
+    <th>TPU</th>
+    <th>PA12</th>
+    <th>SS316</th>
+    <th>AISi10Mg</th>
+  </tr>
+  <tr>
+    <td>P-1</td>
+    <td>100.0%</td>
+    <td>100.0%</td>
+    <td>100.0%</td>
+    <td>100.0%</td>
+    <td>100.0%</td>
+  </tr>
+  <tr>
+    <td>P-2</td>
+    <td>57.2%</td>
+    <td>60.9%</td>
+    <td>97.8%</td>
+    <td>80.5%</td>
+    <td>59.0%</td>
+  </tr>
+  <tr>
+    <td>P-4</td>
+    <td>994.5%</td>
+    <td>1531.3%</td>
+    <td>1901.6%</td>
+    <td>438.1%</td>
+    <td>564.6%</td>
+  </tr>
+  <tr>
+    <td>P-5</td>
+    <td>841.4%</td>
+    <td>1078.1%</td>
+    <td>1721.3%</td>
+    <td>491.4%</td>
+    <td>444.0%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-1</td>
+    <td>231.2%</td>
+    <td>328.1%</td>
+    <td>483.6%</td>
+    <td>198.0%</td>
+    <td>229.7%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-2</td>
+    <td>112.8%</td>
+    <td>132.8%</td>
+    <td>199.5%</td>
+    <td>134.7%</td>
+    <td>120.6%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-4</td>
+    <td>1550.2%</td>
+    <td>2578.1%</td>
+    <td>3243.2%</td>
+    <td>605.1%</td>
+    <td>836.8%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-5</td>
+    <td>1263.3%</td>
+    <td>1703.1%</td>
+    <td>2502.7%</td>
+    <td>624.8%</td>
+    <td>767.2%</td>
+  </tr>
+</table>
+
+Table R3 Normalized Young’s moduli comparison for fabricated cell models
+
+<table>
+  <tr>
+    <th>Model</th>
+    <th>Numerical simulation</th>
+    <th>TPU</th>
+    <th>PA12</th>
+  </tr>
+  <tr>
+    <td>P-1</td>
+    <td>100.0%</td>
+    <td style="background-color: yellow">100.0%</td>
+    <td style="background-color: yellow">100.0%</td>
+  </tr>
+  <tr>
+    <td>P-2</td>
+    <td>57.2%</td>
+    <td style="background-color: yellow">45.9%</td>
+    <td style="background-color: yellow">51.1%</td>
+  </tr>
+  <tr>
+    <td>P-4</td>
+    <td>994.5%</td>
+    <td style="background-color: yellow">1152.9%</td>
+    <td style="background-color: yellow">994.3%</td>
+  </tr>
+  <tr>
+    <td>P-5</td>
+    <td>841.4%</td>
+    <td style="background-color: yellow">811.8%</td>
+    <td style="background-color: yellow">900.0%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-1</td>
+    <td>231.2%</td>
+    <td style="background-color: yellow">247.1%</td>
+    <td style="background-color: yellow">252.9%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-2</td>
+    <td>112.8%</td>
+    <td style="background-color: yellow">100.0%</td>
+    <td style="background-color: yellow">104.3%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-4</td>
+    <td>1550.2%</td>
+    <td style="background-color: yellow">1941.2%</td>
+    <td style="background-color: yellow">1695.7%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-5</td>
+    <td>1263.3%</td>
+    <td style="background-color: yellow">1282.4%</td>
+    <td style="background-color: yellow">1308.6%</td>
+  </tr>
+</table>
+
+Table R4 Normalized Young’s moduli comparison for fabricated 4X4X4 models
+
+<table>
+  <tr>
+    <th rowspan="2">Model</th>
+    <th colspan="3">Numerical simulation</th>
+    <th colspan="3">Numerical simulation</th>
+  </tr>
+  <tr>
+    <th>PA12-FBC</th>
+    <th>AISi10Mg-FBC</th>
+    <th>PA12-PBC</th>
+    <th>AISi10Mg-PBC</th>
+  </tr>
+  <tr>
+    <td>P-1</td>
+    <td>100.0%</td>
+    <td>100.0%</td>
+    <td>100.0%</td>
+    <td>100.0%</td>
+    <td>100.0%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-1</td>
+    <td>61.6%</td>
+    <td>64.0%</td>
+    <td>95.8%</td>
+    <td>95.2%</td>
+    <td>112.5%</td>
+    <td>94.0%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-2</td>
+    <td>40.7%</td>
+    <td style="background-color: yellow">17.6%</td>
+    <td>58.8%</td>
+    <td>83.6%</td>
+    <td style="background-color: red">65.9%</td>
+    <td>92.0%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-4</td>
+    <td>383.7%</td>
+    <td style="background-color: yellow">366.3%</td>
+    <td>198.7%</td>
+    <td>385.9%</td>
+    <td style="background-color: red">378.9%</td>
+    <td>218.4%</td>
+  </tr>
+  <tr>
+    <td>Opt-P-5</td>
+    <td>309.3%</td>
+    <td style="background-color: yellow">295.8%</td>
+    <td>200.7%</td>
+    <td>334.5%</td>
+    <td style="background-color: red">327.7%</td>
+    <td>186.1%</td>
+  </tr>
+</table>
+Fig. R4 | Printing quality comparison for 4X4X4 array models fabricated with PA12 and AlSi10Mg. a, printed models with PA12. b, printed models with AlSi10Mg.
+
+Reference
+[1] Al-Ketan, O. and Abu Al-Rub, R.K. (2019), Multifunctional Mechanical Metamaterials Based on Triply Periodic Minimal Surface Lattices. Adv. Eng. Mater., 21: 1900524. https://doi.org/10.1002/adem.201900524
+
+Comment (2): The authors have delved into the topic of energy absorption in the proposed design. However, it has been observed that the structures were compressed by less than 10%, which might be insufficient for a comprehensive evaluation of energy absorption. An elaboration on this limitation is necessary.
+
+Response: We are sorry for the confusion. The size for the model was 60mmX60mmX60mm. The compressing speed in our simulation was given at 2mm/ms, and the time duration was 20ms, leading to a compression of 66.67% of the whole sample (40mm, 2/3 of the z-direction length). We think this distance is sufficient as some samples have been compressed into compact states, such as the P-4, -5, and Opt-P-4, -5 (Fig. R5, the forces keep increasing).
+
+We have revised the 'Results' part in the main text as follows,
+• '… The size for all the models was 60mmX60mmX60mm. The tests were carried out using uniaxial dynamic compression at a constant speed of 2mm/ms, and the time duration was 20ms, leading to a compression of 66.67% of the whole model (40mm, 2/3 of the z-direction length) ….'
+
+![Force-displacement curves for the P set and Opt-P set.](page_1012_1342_388_388.png)
+
+Fig. R5 | Force-displacement curves for the P set and Opt-P set.
+Comment (3): Figure 5 presents data on energy absorption in the proposed designs. To enhance the comprehensibility of the findings, the authors are encouraged to include a discussion on the mechanisms responsible for energy dissipation, as well as the evolution of deformation patterns. This additional insight will provide a more thorough understanding of the results.
+
+Response: Thanks for the suggestion. Energy absorption is a dynamic process. As can be seen from Fig. R5, energy absorption is the integral of external force and displacement. By increasing the stiffness of the cellular metamaterial, thereby increasing the resisting force, the energy absorption performance can be enhanced. Also, the stability of cellular metamaterial can make a difference in the impact of dynamic progress. As shown in Fig. R5, the force for P-4, and -5 were relatively stable during the impact. For Opt-P-4 and -5, their stiffnesses show evident improvement, as a result, the maximum forces are at a higher level compared with P-4, and -5. However, their forces during the impact demonstrate clear fluctuations compared with P-4 and -5, indicating severe structural buckling is induced with the compression (Fig. R5). In other words, the energy absorption performance can be further improved with enhanced structural stability.
+
+We have added a discussion in the ‘Results’ part of the main text, and the compression movies for the discussed models are provided in the supplementary files, where the deformation patterns for these models can be observed,
+• ‘… (also see deformation pattern evolutions from the Supplementary Movies 1-8 in SI) …’
+• ‘… In this paper, we focus on improving the stiffness of the cellular metamaterial, thereby increasing the resisting force and enhancing the energy absorption performance. However, energy absorption is a dynamic process, and the stability of the cellular model during the impact needs to be taken into consideration. By boosting stability, structural buckling can be alleviated under compression, as a result, the material can be fully utilized and the energy absorption performance can be further improved …’
+Reviewer #2 (Remarks to the Author):
+
+The authors propose a novel topology optimization method for designing multilayered mechanical structures. Some experimental results suggest that the proposed multilayered designs are superior to non-optimized structures in terms of static and dynamic mechanical properties. I find the manuscript to be well-written and interesting to a wide audience. However, the authors should address the following comments before I recommend the manuscript for publication in Nature Communications:
+
+Response: The authors appreciate the reviewer’s positive feedback and constructive comments. Below, we carefully addressed the comments from the reviewer.
+
+Comment (1): Although I understand that the manuscript focuses on cellular lattice structures, I think that the authors should explain the advantages of the proposed multilayer strategy compared to standard optimized designs. Specifically, why do the authors need the multilayer strategy to increase the design flexibility? For example, to the best of my knowledge, the standard and conventional SIMP method provides maximum design flexibility.
+
+Response: Thanks for the comment. The reviewer is correct that the cubic design domain discretized with standard density element enjoys maximum design freedom. It is certain that the conventional SIMP method can theoretically produce optimized results. However, as the SIMP method is typically performed on the standard cubic density element, it is usually difficult to represent or design surface structures, unless a massive amount of fine elements are used. Apparently, this can lead to computational inefficiency. Unlike the standard SIMP method, the proposed multilayer strategy here regularizes and tailors the design space to better accommodate the surface design while ensuring enough flexibility due to the stacking of multiple layers. It also enables the use of shell elements for plane and surface geometries, and the optimization can be efficient. Another advantage of the multilayer strategy is that each layer itself and the cavities separated by those layers can be set as independent design spaces (Fig. R6). In this way, these sub-design spaces
+
+![Multilayer strategy enables for independent design spaces.](page_374_1092_1002_563.png)
+
+Fig. R6 | Multilayer strategy enables for independent design spaces.
+can be assigned the same or different material properties, of the same or different boundary conditions, and the optimization can be performed on those sub-design spaces with the same or different objectives, which can be flexible in dealing with a series of 'multi-' problems (Fig. R6). In addition, the standard SIMP method does not necessarily guarantee the optimized result is open-cell, which might not be printing-friendly. For the multilayer model, the optimized result can always be open-cell as topology optimization creates holes on the surface of each layer. In summary, the multilayer strategy introduced in this paper can be useful in many perspectives.
+
+We have added a sentence in the 'Discussion' part of the revised manuscript,
+• ‘… Also, the multilayer strategy regularizes and tailors the design space. Another advantage is that each layer itself and the cavities separated by those layers can be defined as independent sub-design spaces, which can be flexible in extensions to various problems …’
+
+Comment (2): In Figure 2a, the left-most Opt-IWP performs worse than the non-optimized IWP. Please add a discussion on this anomaly.
+
+Response: Thanks for the comment. In comparison to the original model without optimization, we use the thickness-compensation scheme to ensure they are at the same relative density for fair comparison. The thickness compensation, however, does not necessarily guarantee that the thickened optimized results are better than the original models. Firstly, the optimization of shell structures is thickness-dependent. For example, as can be seen in Fig. R7a, the optimized results show quite different topological configurations as the thickness changes for the Neovius model. Thus, the optimized result with compensated thickness can be different from the optimized result with the original thickness. On the other hand, topology optimization reconfigures the material distribution, which may lead to a shift of deformation mode, such as the change from stretching-dominated to bending-dominated. This may weaken the mechanical performance of the original model. For example, as can be seen in Fig. R7b, the optimized IWP shows clear bending behavior while the Neovius is still stretching-dominated. As a result, the thickness-compensation works well for the Neovius and its stiffness demonstrates remarkable improvement, while the performance of IWP may suffer deterioration. In addition, topology optimization inevitably creates holes on the surface of the designable region (though we can control the area fraction to limit the area of those holes), as we use shell element to optimize and there is only one layer of the shell element, which can induce stress concentration and local deformation around those holes (Fig. R7c), thereby significantly impairing the mechanical performance.
+
+To counter the above-mentioned limitations, one way is to use the multilayer solid element to perform the optimization (Fig. R7d). Among the multilayer solid elements, each layer can be assigned as designable or non-designable. In this way, the optimization can be implemented without creating holes by deliberately prescribing a non-designable region, thereby maintaining the smooth surface and avoiding stress concentration. Although this may render close-cell of the cellular material, leading to difficulty in printing postprocess, the setting for non-designable regions can always be optional.
+
+We have added a sentence in the 'Results' part of the revised main text, and a discussion in the 'Supplementary Note 5' in SI,
+• ‘… However, digging holes on the surface with topology optimization does not necessarily guarantee better performance. Several factors, such as the thickness, deformation mode, stress concentration and local deformation, may result in structural deterioration (detailed discussion in Supplementary Note 5 in SI) …’
+• ‘… Supplementary Note 5: Optimization anomaly …’
+
+Comment (3): I do not comprehend how to interpret Figure 3b and Figure 4c, 4d. The authors present the experimental results for each model (IP set and Opt-P set) with four different printing materials. However, the numerical results are presented for only one case. In this numerical simulation, which printing material is assumed? From these results, I do not observe any consistency between the experimental and numerical results, despite the authors claiming that “... shows much consistency with corresponding numerical results, yet …”.
+
+…
+Fig. R7 | Design and optimization strategy for multilayer shell-based lattice. a, optimized results with different thickness. b, deformation modes comparison for the optimized IWP and Neovius. c, deformation modes comparison for the Neovius without and with holes. d, a multilayer strategy for a single-layer model with independent designable or non-designable regions.
+
+Response: We are sorry for the confusion. We only give one simulation result because we study the relative proportional ratios of the effective Young’s moduli among the P set and Opt-P set and they should be material-independent. Thus, we chose four different printing materials to verify the simulation results. In our numerical simulation, the material iron is used (Young’s modulus: 210GPa, yield strength: 400MPa). The first reviewer raised the same concern, please see the
+detailed response to comment (1) from Reviewer 1.
+
+Comment (4): In Supplementary information, please add some references or more detailed descriptions on the material characterization. For instance, how the effective elastic tensor, Zener's ratio, etc. are derived or determined.
+
+Response: Thanks for the suggestion. We have added related discussion in ‘Supplementary Note 1’, and references in the ‘Supplementary References’ in SI,
+• ‘… Supplementary Note 1: Material characterization …’
+• ‘… Supplementary References …’
+
+Comment (5): There are typos in the section title and Eq.(8) in Supplementary information.
+
+Response: Thanks for the correction. We have revised those typos in the manuscript.
+REVIEWERS' COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+Accept
+
+Reviewer #2 (Remarks to the Author):
+I am satisfied with the authors' response to my review comments. I recommend the revised manuscript for the publication in Nature Communications.
+Response to the comments of the reviewers
+
+REVIEWER COMMENTS
+
+Reviewer #1 (Remarks to the Author):
+
+Accept
+
+Reviewer #2 (Remarks to the Author):
+
+I am satisfied with the authors' response to my review comments. I recommend the revised manuscript for the publication in Nature Communications.
+
+Detailed response to the Reviewers
+
+The authors appreciate very much the editors and reviewers’ time and effort spent on reviewing our manuscript. Thanks to your constructive and insightful comments, our manuscript has improved a lot.
+
+Reviewer #1 (Remarks to the Author):
+
+Accept
+
+Response: The authors would like to thank the reviewer for the comment.
+
+Reviewer #2 (Remarks to the Author):
+
+I am satisfied with the authors' response to my review comments. I recommend the revised manuscript for the publication in Nature Communications.
+
+Response: Many thanks to the reviewer’s recommendation.