Add files using upload-large-folder tool

preprint/preprint__7e12039b3edcb5d3b25d34c354cf8a8eb7554255f59d07a479f069122448bcfc/preprint__7e12039b3edcb5d3b25d34c354cf8a8eb7554255f59d07a479f069122448bcfc.mmd

@@ -0,0 +1,341 @@
+
+# First Demonstration of In-Memory Computing Crossbar using Multi-level Cell FeFET  
+
+Taha Soliman Robert Bosch GmbH https://orcid.org/0000- 0002- 9421- 9489  
+
+Swetaki Chatterjee University of Stuttgart https://orcid.org/0000- 0002- 2550- 9626  
+
+Nellie Laleni Fraunhofer IPMS  
+
+Franz Müller Fraunhofer IPMS https://orcid.org/0000- 0002- 6564- 9121  
+
+Tobias Kirchner Robert Bosch GmbH  
+
+Norbert Wehn Rheinland- Pfälzischen Technischen Universität Kaiserslautern- Landau (RPTU)  
+
+Thomas Kämpfe Fraunhofer Institute for Photonic Microsystems  
+
+Yogesh Singh Chauhan Indian Institute of Technology Kanpur https://orcid.org/0000- 0002- 3356- 8917  
+
+Hussam Amrouch ( amrouch@tum.de ) Technical University of Munich (TUM)  
+
+Article  
+
+Keywords:  
+
+Posted Date: May 26th, 2023  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 2948718/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.
+
+<--- Page Split --->
+
+Version of Record: A version of this preprint was published at Nature Communications on October 10th, 2023. See the published version at https://doi.org/10.1038/s41467-023-42110-y.
+
+<--- Page Split --->
+
+# First Demonstration of In-Memory Computing Crossbar using Multi-level Cell FeFET  
+
+Taha Soliman \(^{1 + *}\) , Swetaki Chatterjee \(^{2,5 + }\) , Nellie Lalemi \(^{3}\) , Franz Müller \(^{3}\) , Tobias Kirchner \(^{1}\) , Norbert Wehn \(^{4}\) , Thomas Kämpfe \(^{3*}\) , Yogesh Singh Chauhan \(^{5}\) , Hussam Amrouch \(^{6,7*}\)  
+
+\(^{1}\) Robert Bosch GmbH, Renningen, Germany \(^{2}\) University of Stuttgart, Stuttgart, Germany \(^{3}\) Fraunhofer IPMS, Dresden, Germany \(^{4}\) RPTU Kaiserslautern- Landau, Kaiserlautern, Germany \(^{5}\) Indian Institute of Technology Kanpur, Kanpur, India \(^{6}\) Munich Institute of Robotics and Machine Intelligence, Munich, Germany \(^{7}\) AI Processor Design, Technical University of Munich (TUM), Munich, Germany  
+
+\(^{+}\) Equal Contribution \(^{*}\) To whom correspondence should be addressed Email: taha.soliman@bosch.de, thomas.kaempfe@ipms.fraunhofer.de, amrouch@tum.de  
+
+Advancements in AI led to the emergence of in- memory- computing architectures as a promising solution for the associated computing and memory challenges. This study introduces a novel in- memory- computing (IMC) crossbar macro utilizing a multi- level FeFET cell for multi- bit multiply and accumulate (MAC) operations. The proposed 1FeFET- 1R cell design stores multi- bit information while minimizing device variability effects on accuracy. Experimental validation was performed using 28 nm HKMG technology- based ferroelectric field- effect transistor (FeFET) devices. Unlike traditional resistive memory- based analog computing, our approach leverages the electrical characteristics of stored data within the memory cell to derive MAC operation results encoded in activation time and accumulated current. Remarkably, our design achieves
+
+<--- Page Split --->
+
+96.6% accuracy for handwriting recognition and 91.5% accuracy for image classification without extra training. Furthermore, it demonstrates exceptional performance, achieving 885.4 TOPS/W- nearly double that of existing designs. This study represents the first successful implementation of an in- memory macro using a multi- state FeFET cell for complete MAC operations, preserving crossbar density without additional structural overhead.  
+
+## Introduction  
+
+Although the demand for data transmission is increasing globally, it is expected that the future will prioritize data- centric local intelligence at the edge node. This, in turn, will inevitably necessitate devices, including wearables, sensors, smartphones, and cars, to locally analyze data and make autonomous decisions.  
+
+Edge- AI devices have great potential to enable new applications with higher performance and support local embedded intelligence, real- time learning, and autonomy. This could propel the semiconductor industry's next growth phase. Energy- efficient local computing is crucial to enable smart connected Internet of Things (IoT) devices. Further, the rise of memory- intensive computational tasks has led to a significant increase in the amount of data that needs to be accessed compared to local computation inside the algorithmic logic units. This issue is commonly known as the von- Neumann bottleneck \(^{1}\) . One of the most notable examples of such storage- bound tasks is related to AI, specifically
+
+<--- Page Split --->
+
+when it comes to deep learning applications. In recent years, the research community has demonstrated the potential of various architectural improvements in computing systems, such as near- memory or IMC, to meet these energy, compute and memory requirements \(^{2}\) .  
+
+Similarly, deep neural networks (DNNs) have gained popularity due to their remarkable performance, especially in applications such as speech recognition and image processing. However, in order to design an efficient DNN, various metrics such as throughput, latency and energy efficiency must be jointly optimized. Therefore, researchers have turned their attention to IMC architectures for deploying such networks. Such architectures perform MAC operations by utilizing the memory array without the need for data movement, resulting in improved system performance and large energy savings, thus, overcoming the fundamental bottleneck of von- Neumann architecture.
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1: Overview of the proposed IMC macro for MAC operations a. The material stack of FeFETs b. The multi-bit FeFET can be programmed to different states to store the weight of the synapse. c. Previous works \(^{3 - 5}\) only considered binary AND or XNOR operation to compute a single-bit multiplication operation. d. Our proposed 2-bit multiplication operation with input encoding and 2-bit storage is shown. The corresponding output activates at different instances of time. e. An encoder provides the gate voltage depending on the input value which changes between three levels at different instances of </center>
+
+<--- Page Split --->
+
+time. f. The multiplication output of the input and stored state in the cell depends on the time at which one cell is activated which is accumulated and sampled using the decoder. g. Depending on the activation time and the number of cells activated at a given time, the voltage across the capacitor connected to a column of cells is accumulated which corresponds linearly with the MAC output and has a minimal impact of the underlying device variation. h. The corresponding MAC operation is performed in the crossbar, accumulating the output in the capacitor voltage. i. IMC accelerators facilitate MAC operations for AI workloads where our proposed design can be utilized.  
+
+Advanced IMC architecture design relies on the usage of emerging memory cells, particularly non- volatile memory (NVM) cells. These NVMs are used to store weights in neural network inference architectures without needing a steady power supply, and to perform MAC operations using their analog properties. Various emerging memories have been presented, such as resistive RAM (ReRAM), magnetic RAM, and FeFETs \(^{6 - 8}\) . In this study, we focus on using FeFETs as memory cells due to their superior performance compared to other emerging memories \(^{8}\) . Our FeFET is co- integrated into a 28nm high- \(\kappa\) - metal- gate (HKMG) technology, resulting in a low footprint. The FeFET exhibits low read latency ( \(\sim 1\) ns), current source capability, and extremely high write power efficiency ( \(< 1\) fJ) with a short write duration ( \(\sim 1\mu s\) ), making it a highly suitable memory cell for IMC \(^{8}\) .  
+
+State- of- the- art FeFET based IMC architectures have been limited to binary logical
+
+<--- Page Split --->
+
+operations, specifically logical AND and XNOR \(^{3 - 5,9,10}\) . These operations are restricted to storing only two states within the FeFET memory cell. Even the use of multi-level cell (MLC) FeFETs will not offer any additional advantages in these architectures as the computations are limited to only binary operations. However, MLC FeFET have been utilized for other tasks such as matching in hyperdimensional computing \(^{11}\) . We demonstrate for the first time a multi-bit MAC operation with variation- affected FeFET cells.  
+
+Contrary to analog computing based implementations of the MAC macro \(^{12,13}\) , we do not perform direct analog multiplication of the input and weight, which is highly prone to variations and require a high degree of linearity in the stored states. Instead, we operate using the current- limited cell such that each cell that is activated has the same current contribution, which limits the impact of variation and improves operation accuracy. This is concisely what enabled us to overcome the variation and utilize MLC FeFET for the first time to demonstrate full MAC operations.  
+
+We designed and demonstrated a MAC circuit macro consisting of cells connected in the crossbar structure using the 1FeFET- 1R configuration, which includes a single FeFET and a single resistor. In this macro, the FeFET cell acts as a memory for the entire weight value. We explore in this work three dimensions (time, stored threshold voltage (Vth) state of the FeFET, and output current) to perform a complete MAC operation using the single FeFET per weight, detailed in Fig. 1. The input is encoded in applied voltage duration and magnitude, the multi- bit weight is stored in the FeFET, and the output is accumulated as
+
+<--- Page Split --->
+
+the capacitor voltage that depends on the activation time and number of FeFETs activated as shown in Fig. 1.  
+
+Depending on the stored and input values, the multiplication operation result is encoded in the FeFET cell activation time. Additionally, the number of simultaneously activated FeFET cells is represented by the drain current over time which reflects the accumulation of the performed multiplications. Based on the voltage of the capacitor in the decoder block connected to an operating column of cells as shown in Fig. 1h, the final MAC operation results can be determined and quantized to the required bit precision using a suitable analog- to- digital (ADC) converter. To prove the validity of our macro design, the stored weights, input, as well as output, are quantized to 2 bits. However, the presented macro and computational idea can be extended to higher precision as long as the memory cell can store the targeted precision while maintaining the clear separation between the memory states.  
+
+We demonstrate in this paper the multiplication operation by a single FeFET cell and extend to complete MAC using the IMC macro of 32x32 FeFETs. This macro is tested against the LeNet network for MNIST handwritten digit database and several layers from VGG- 19 for the CIFAR- 10 dataset. Our presented work maintained the network accuracy at 97% with less than 2% accuracy loss. We highlight how IMC combined with emerging technologies and temporal- based computation can fulfil the increasing AI field computational and memory requirements and overcome the von- Neumann bottleneck.
+
+<--- Page Split --->
+
+## Our Proposed Architecture  
+
+Single- Cell Multiply operation The polarization of the ferroelectric layer can be modulated by the application of voltage pulses which in turn changes the \(\mathrm{V_{th}}\) of the underlying transistor and hence the conductance. The different \(\mathrm{V_{th}}\) states are achieved with respect to the value of the voltage pulse and the pulse duration. Fig. 1b illustrates the MLC FeFET where it is programmed to four different states to store 2- bit information.  
+
+In this work, we use the 1FeFET- 1R concept (to limit the drain current variability and operate in the saturation region) for a memory cell \(^{4,11}\) . This concept is constructed by adding a drain resistance to the FeFET. The resistance is in the range of \(1M\Omega\) for a drain current of approximately \(100\mathrm{nA}\) and can be implemented by a transistor biased at its gate or a physical resistance \(^{11,13}\) . In this way, the current variability is reduced, as well as the power consumption of the memory unit cell due to the limiting of the drain current to a fixed value, at the expense of the total area.  
+
+We propose a novel methodology to use the MLC FeFET cell to perform the complete MAC operation. As shown in Fig. 2a, the 2- bit weight is encoded in the stored state inside the FeFET state such that the smallest value is represented by the highest \(\mathrm{V_{th}}\) state and the largest by lowest \(\mathrm{V_{th}}\) state. On the other hand, the input value will be encoded as the time stamp at which the different gate voltages are applied according to Eq. (1)
+
+<--- Page Split --->
+
+\[2\mathrm{bit} - V_{g}(\mathrm{Input}, \mathrm{time}) = \left\{ \begin{array}{ll}0, & \mathrm{if} \mathrm{Input.time} < \epsilon_{1}, \\ V\mathrm{th}1, & \mathrm{if} \epsilon_{1} < = \mathrm{Input.time} < \epsilon_{2}, \\ V\mathrm{th}2, & \mathrm{if} \epsilon_{2} < = \mathrm{Input.time} < \epsilon_{3}, \\ V\mathrm{th}3, & \mathrm{if} \epsilon_{3} < = \mathrm{Input.time}. \end{array} \right. \quad (1)\]  
+
+According to Eq. (1), the value of the gate voltage applied to the FeFET cell over time is dependent on the 2- bit input value. Consequently, the timestamp at which the FeFET saturates can encode the results of the multiplication of the stored state inside the FeFET as well as the input voltage. However, one of the main issues will be ensuring the commutative property for multiplication. In Eq. (1)), the values \(\epsilon_{1},\epsilon_{2},\epsilon_{3}\) are used to tune the ramping voltage speed to ensure the commutative property. As shown in Fig. 2, we can demonstrate a 2- bit/2- bit multiply operation performed by our memory cell in both simulations and measurements.  
+
+Multiple- Cell Accumulation Operation To read out the multiple- cell accumulation operation, we utilized the charging of a capacitor due to the total accumulated drain currents in a column over a period of time. The drain current from multiple FeFETs are collected and directed to the accumulation capacitor. The saturation time of the FeFET encodes the multiplication output, and the number of FeFETs activated at a given time represents the accumulation. The capacitor voltage at the end of the period or the sampling time( \(\mathrm{t}_{\mathrm{s}}\) ) is given by
+
+<--- Page Split --->
+
+\[V_{sampling} = \frac{1}{C}\times \int_{0}^{t_{s}}\sum_{n = 1}^{nFeFETS}I_{d,n}(t)dt \quad (2)\]  
+
+where \(C\) is the capacitance of the capacitor attached to the column, \(I_{d,n}\) is the drain current of the \(n^{th}\) FeFET, \(nFeFETS\) is the total number of FeFETs connected. At that point, the FeFETs are in the saturation region and limited by the resistance according to Eq. (3).  
+
+\[I_{d,n}(t) = \frac{V_{dd} - V_{s}(t)}{R_{out}}\cdot h(t - t_{o}) \quad (3)\]  
+
+\(V_{s}(t)\) is following the charging voltage function over a capacitance \(C\) and time constant \(\tau = R_{out}\times C\) where \(R_{out}\) is the total resistance seen by the source of the FeFET including the current limiter resistance \(R\) . The time- constant \(\tau\) is sufficiently high to not allow a steep fall in the current. \(h(t - t_{o})\) is the step- function at time \(t_{o}\) . It represents the time at which the FeFET is activated and it depends on the input applied and the stored state. The value of \(t_{o}\) is such that it maintains the commutative property of multiplication. Since the current follows a step- jump at \(t_{o}\) , the voltage across the capacitor follows a ramp after time \(t_{o}\) .  
+
+To quantize the MAC output, we employ StrongArm voltage input comparators \(^{14}\) followed by latches and encoder for a complete 2- bit output. We explored the influence of the value of \(nFeFETS\) and the induced cell variation on the accumulation results and the resulting loss in inference accuracy.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2: Direct 2-bit multiply operation in a single cell. a. The measured \(\mathrm{I_{ds} - V_{gs}}\) characteristics of the 1FeFET-1R cell corresponding to the four stored states. b. Input voltage against time applied to the FeFETs for measurement, and c. Output current against time on applying the input pulse. \(\mathrm{I_{ds}}\) rises at different instants of time corresponding to different outputs, which are used to get the product of input and stored states. d.-f. Stored, input, and output from a single cell multiplication operation verified through simulations and determines the maximum speed of operation possible. </center>
+
+<--- Page Split --->
+
+## Results  
+
+Experimental measurements The FeFET test structures are fabricated in the GlobalFoundries \(28\mathrm{nm}\) high- k/metal gate technology node, for which co- integration of FeFETs with CMOS devices has been demonstrated 15. The FeFETs consist of a \(\mathrm{SiO_2}\) interfacial oxide layer, followed by an 8 to \(10\mathrm{nm}\) thick, ferroelectric doped \(\mathrm{HfO_2}\) layer as illustrated in Fig. 1a and Fig. S1a. The gate is capped with a TiN metal cap, and silicided poly silicon 16.  
+
+A 1FeFET- 1R cell is constructed by externally connecting a \(1\mathrm{M}\Omega\) resistor to a FeFET with an area of \(450\times 450\mathrm{nm}^2\) . This is necessary to limit the ON current and control variations. The 1FeFET- 1R cell is written to 4 distinct states, as shown in Fig. 2a. For details of the methods of writing the FeFET into the desired state, refer to the Methods section.  
+
+As stated earlier, for a single cell, the input is encoded in duration, and the magnitude of the voltage applied and shown in Fig. 2b. For an input of '0', the gate voltage is kept constant at 0 V. This voltage is less than the \(\mathrm{V_{th}}\) of any stored state in the FeFET, and no FeFET is turned ON. For the other inputs, the magnitude of the voltage is changed to levels \(\mathrm{V_1}\) , \(\mathrm{V_2}\) , and \(\mathrm{V_3}\) at a certain point in time, as stated in Eq. (1). The voltage level corresponds to the read voltage of the FeFETs storing '3', '2', and '1', respectively. Consequently, the FeFET storing '3' is turned on earliest, and a FeFET storing '0' is never turned ON. For input '3', voltage \(\mathrm{V_1}\) is applied at 0, and for input '1', voltage \(\mathrm{V_1}\) is applied later at \(300\mu \mathrm{s}\) . Correspondingly, the FeFET storing '3' would turn on immediately for input '3' and turn on later for input '1'. Hence, the output of the single multiplication operation
+
+<--- Page Split --->
+
+between the stored weight and the applied input is encoded as the time when the FeFET turns on, i.e.- the activation time of the FeFET.  
+
+Evaluation at different input conditions is performed from the resulting \(I_{ds} - V_{gs}\) transfer characteristics. The corresponding output is shown in Fig. 2c. The verification is done for a drain current of 75 nA. The timing marked for each output state is distinguishable. The current rises for an output of 9 first and at last for an output of 1. For intermediate output states, the activation time is in between. The commutative property is also maintained as seen for output of '2', '3', and '6'. The instance of time when the current rises can distinguish between the different output states, as shown in Fig. 2c. This forms the basis for the MAC operation.  
+
+Simulation The functionality of the proposed in- memory macro is exemplified through simulations. The FeFET is simulated using a Preisach- based model \(^{17}\) of the Ferroelectric capacitor and industry- standard compact model \(^{18}\) of the underlying transistor (for details, see SI). We use 2- bit storage for the FeFET as in measurements, which corresponds to four different \(\mathrm{V_{th}}\) states. The simulation characteristics are matched to the experiments as closely as possible, and the resulting \(I_{ds} - V_{gs}\) characteristics are shown in Fig. 2d.  
+
+Correspondingly, we simulate the single- cell multiply operation as in measurements. The timing of the input pulse is modified such as to have almost linear characteristics of the output voltage against the desired output. Also, to determine the maximum speed of operation, the input pulse width is greatly reduced. The values are as per the fast read
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3: Accumulation of the output as the voltage across the capacitor a. Structure of the single cell connected to the capacitor. b. The voltage across the capacitor \(\left(V_{\mathrm{cap}}\right)\) against time. As the current for output 9 (inp-3 x stored-3) is turned on first, the voltage across the capacitor at \(t_{\mathrm{sampling}}\) of 14 ns is highest. Similarly, for output 1 (inp-1 x stored-1), the voltage is lowest. c. The sampling voltage across the capacitor \(\left(V_{\mathrm{sampling}}\right)\) at \(t_{\mathrm{sampling}}\) vs the output MAC value. It is linear and perfectly aligned for the same output. </center>
+
+<--- Page Split --->
+
+out proposed in FeFETs \(^{8}\) . The input and the corresponding output are shown in Fig. 2e. and Fig. 2f., respectively. Each output is distinguishable with no overlap, and also, the commutative property of the multiplication is maintained. We further simulated and evaluated the proposed MAC macro for a complete array of connected cells for neural network inference.  
+
+## Evaluation  
+
+Crossbar Level The memory cells are arranged in the crossbar structure. In our recent work \(^{19}\) , we demonstrated that such an array can be programmed with MLC cells for up to 3b precision employing inhibit voltage levels and target erase schemes. A capacitor of 64 fF is connected at the bit- line for each column which is charged and discharged after every cycle. The value of the capacitor is so chosen to allow the charging without being saturated. The voltage to which it is charged depends on the total current flowing into it and the time for which it flows, as given by equation Eq. (2). The time is determined by the input and stored state at which a particular FeFET activates. The number of FeFETs activated at a given time determines the total current that flows into the capacitor. The voltage across the capacitor ( \(\mathrm{V}_{\mathrm{cap}}\) ) is sampled at a particular time ( \(\mathrm{t}_{\mathrm{sampling}}\) ) using a 2- bit ADC to get the final output.  
+
+The single cell connected to the capacitor is shown in Fig. 3a. The capacitor is charged with an almost constant current of 100 nA, and thus, \(\mathrm{V}_{\mathrm{cap}}\) rises linearly with
+
+<--- Page Split --->
+
+time as shown in Fig. 3b. The sampled voltage is maximum for the case of output 9 because \(\mathrm{I_d}\) turns on earliest in this case (at 1 ns) Fig. 3c. For the case of output 1, voltage is minimum because \(\mathrm{I_d}\) turns on the last (at 13 ns).  
+
+Similarly, in the case of 2 cells in the array, for output 18, \(\mathrm{V_{cap}}\) is maximum. Here, both the cells are activated at 1 ns, and the current is double that of one cell. Hence, the sampled voltage is also approximately double that of output 9. For output 1, only one of the FeFETs is turned on at 13 ns, and the sampled voltage is minimum. For intermediate values of output ('2', '3', '4', and '6') the sampled voltage at the capacitor lies in between the maximum and minimum as shown in Fig. 3c.  
+
+The reasoning can be extended to a higher number of cells in the column. The sampled voltage across the capacitor progressively increases with the number of cells activated. Fig. 4 shows the sampled voltage \((\mathrm{V_{sampling}})\) against the MAC output for up to 32 cells in the array. We considered all the possible input and stored combinations to generate the MAC output. We maintained a clear distinction between each output level and the numerical MAC output. However, with an increase in the number of cells, the linearity for higher output values is lost, and the sampling voltage starts to saturate. This is also expected from equation Eq. (2).  
+
+The final output is converted into 2 bits using strong- ARM comparators \(^{14}\) connected across the capacitor. The comparison threshold levels are selected based on the quantization required. For the final simulations of the DNN, 32 cells in a column are considered.
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4: Complete MAC output for increasing number of cells connected in the column. a.- f. 1 cell to 32 cells sampled voltage across the capacitor against the MAC output connected in a single column. High degree of linearity is observed, which is desired for the final neural network inference. </center>
+
+<--- Page Split --->
+
+The crossbar is divided into \(32 \times 32\) cells in a segment. Each column in the segment is connected to the 2- bit ADC. The quantized weights are directly written to the FeFET crossbar (For details on the quantization, see SI). The quantized input is applied using a digital- to- analog converter (DAC) connected to the word line for each row, which selects a particular voltage based on the applied input.  
+
+Afterwards, the variability in the FeFET devices are incorporated in the FeFET model to calculate the loss in inference accuracy of the neural network. The variation in \(\mathrm{V}_{\mathrm{th}}\) is measured from real fabricated FeFET devices. Fig. 5 shows the distribution for three different target condition sets of \(\mathrm{V}_{\mathrm{th}}\) . In the first set, the target levels are chosen as \(0.2 \mathrm{~V}, 0.6 \mathrm{~V}, 1.0 \mathrm{~V}\) and \(1.4 \mathrm{~V}\) . In the next set, the target levels are \(0.3 \mathrm{~V}, 0.633 \mathrm{~V}, 0.967 \mathrm{~V}\) and \(1.3 \mathrm{~V}\) . In the third set, the target levels are \(0.3 \mathrm{~V}, 0.6 \mathrm{~V}, 0.9 \mathrm{~V}\) and \(1.2 \mathrm{~V}\) . The maximum standard deviation of \(38 \mathrm{mV}\) is obtained for the difference in actual and target \(\mathrm{V}_{\mathrm{th}}\) .  
+
+For simulations, a standard deviation of \(40 \mathrm{mV}\) for \(\mathrm{V}_{\mathrm{th}}\) of the FeFET is assumed for each state. 1000 Monte Carlo samples for each stored, and the input value is simulated for up to 4 cells in the array. For a higher number of cells, the total variability of each state is calculated algebraically from the lower number of cells (for details, see SI). A maximum standard deviation of less than \(4 \mathrm{mV}\) for any given output state is observed in the case of 32 cells in the array. Quantization of the output into 4 levels (2 bits) further reduces the error probability. Finally, the neural network models are simulated to calculate the loss in inference accuracy and derive the performance metrics of the proposed MAC macro.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Figure 5: Experimentally measured device-to-device variation with respect to \(\mathbf{V}_{\mathrm{th}}\) . a. Target condition 1. b. Target condition 2. c. Target condition 3. Different target conditions were set for measuring the standard deviation in target and measured \(\mathbf{V}_{\mathrm{th}}\) . A maximum standard deviation of \(38\mathrm{mV}\) is observed. </center>  
+
+Architecture level We used the experimental and simulation data to estimate the performance of the demonstrated macro on two neural network models. We perform the inference of these networks on the target macro and at limited variability of \(40\mathrm{mV}\) based on our measurements. As shown in Fig. 6a,b, we fully quantize the LeNet model \(^{20}\) for MNIST \(^{21}\) into 2- bit activation and 2- bit weights to fit our macro capabilities of MAC precision. The network consists of three convolutional layers and two dense layers. The network requires 397920 MAC operations using 61610 parameters. Considering a max of 40 mv device variation, we achieve \(96.64\%\) network accuracy compared to the original model accuracy of \(99.11\%\) achieved at full floating point precision.  
+
+Additionally, we also quantize two layers from the VGG19 \(^{26}\) network for the CIFAR- \(^{10}\) \(^{27}\) dataset as shown in Fig. 6c and d. The network consists of 19 layers. However, we
+
+<--- Page Split --->
+![](images/Figure_6.jpg)
+
+<center>Figure 6: Neural network simulation utilizing the proposed IMC MAC macro. a. LeNet neural network is tested for handwritten digit recognition MNIST dataset. All the MAC layers are quantized to 2 bits. b. The weight distribution shows the trained weights when they are re-quantized to 2 bits for the second convolutional layer. An accuracy of \(96.64\%\) is achieved considering device variations. c. VGG19 neural network is tested for object classification CIFAR-10 dataset where only two convolutional layers are quantized for testing. d. The weight distribution shows the trained weights when they are re-quantized to 2 bits for a convolutional layer in VGG19. An accuracy of \(91.55\%\) is achieved under the effects of device-to-device variations. </center>
+
+<--- Page Split --->
+
+
+Table 1: Comparison of in-memory crossbars with this work   
+
+<table><tr><td>Crossbar</td><td>Memory</td><td>Tech. [nm]</td><td>Freq. [Mhz]</td><td>Power [μW]</td><td>Throughput [GOPS]</td><td>Efficiency [TOPS/W]</td></tr><tr><td>ISSCC&#x27;22a</td><td>SRAM</td><td>28</td><td>333</td><td>-</td><td>-</td><td>438</td></tr><tr><td>SLC-MLC 23</td><td>PCM</td><td>40</td><td>307</td><td>14900</td><td>3900</td><td>261</td></tr><tr><td>nvCIM 24</td><td>RRAM</td><td>55</td><td>1</td><td>0.0322</td><td>0.002</td><td>62.11</td></tr><tr><td>Samsung 25</td><td>MRAM</td><td>28</td><td>11.1</td><td>225</td><td>91.125</td><td>405</td></tr><tr><td>This work</td><td>FeFET</td><td>28</td><td>66</td><td>153.6</td><td>136</td><td>885.4</td></tr></table>
+
+a Performance scaled to 2-bit/2-bit precision  
+
+tested and quantized only two convolutional layers. The networks require 38947914 parameters out of which we quantized only 1179648 parameters to 2- bit, where the rest are quantized to 8- bit. The 2- bit quantization layers use 2- bit quantized activations. We consider a max of 40 mv device variation for those two layers. We achieved \(91.55\%\) network accuracy compared to the original model accuracy of \(93.22\%\) .  
+
+As demonstrated, the variation has a very limited influence on the accuracy of the final network resulting in less than \(4\%\) in both networks over the testing set. This can be reasoned by the cell architecture of 1FeFET- 1R, which limits the current and the quantization of the final output into 2 bits, which limits the impact of the variation on the sampling voltage thresholds.
+
+<--- Page Split --->
+
+In our experiments, the crossbar and the ADCs consume \(153.6 \mu \mathrm{W}\) measured directly for the given frequency (66 MHz) and accumulation capacitance (64 fF). The energy efficiency of the presented crossbar accordingly is \(885.4\) tera- operations per second power watt (TOPS/W), where each operation refers to 2- bit/2- bit multiply or accumulate. We compared our design against existing in- memory crossbars which are shown in Table 1.  
+
+## Discussion  
+
+In this work, we have demonstrated the multi- bit MAC operation exploiting for the first time a multi- bit FeFET cell and a novel encoding and decoding scheme. The variability was controlled with the help of the external current- limiting resistor approach. Compared to other works in the literature with different memory architectures, we show a higher throughput and efficiency. The results portray that FeFET can be a strong contender for DNN acceleration with high efficiency.  
+
+We see an important potential for our computational innovation to allow for a dense IMC macro that can use simple memory cell 1FeFET- 1R while performing multi- bit MAC operation per each cell. Additionally, the combination of several computational dimensions time/voltage/current encodings allow for combating the variability and maintaining the macro efficiency in an unprecedented way.
+
+<--- Page Split --->
+
+## Methods  
+
+1FeFET- 1R Cell characterization To write the FeFET, into 4 distinct states (i.e 2- bit), we employed the write- verify scheme. A FeFET is written into a fully programmed state by applying \(+4.5\mathrm{V}\) for \(500~\mathrm{ns}\) and into a fully erased state by applying - 5 V for \(500~\mathrm{ns}\) while keeping the source and drain terminals grounded. Before evaluation, each FeFET is cycled 50 times with these conditions for preconditioning. A FeFET is then written to 4 distinct states using a write- verify- scheme. Therefore, the FeFET is initially fully erased. Starting with a write voltage of \(1.4\mathrm{V}\) for \(200~\mathrm{ns}\) , the FeFET is gradually programmed. Write voltage is incremented in steps of \(40\mathrm{mV}\) . After each write pulse, a delay of \(500~\mathrm{ms}\) is applied for charge detrapping, and a read operation verifies the state. This scheme is continued until the target value is reached. Target levels are selected at \(0.3\mathrm{V}\) , \(0.7\mathrm{V}\) , \(1.1\mathrm{V}\) , and \(1.5\mathrm{V}\) at a constant current condition of \(80\mathrm{nA}\) . After setting the target state, a final readout is performed. Before reading, a sufficiently large time is waited for any charge detrapping. In this case, a delay time of 2 seconds is chosen. For reading a voltage ramp, \(V_{\mathrm{G}}\) from - 0.2 V to \(1.7\mathrm{V}\) in steps of \(10\mathrm{mV}\) is applied to the gate. Current \(I_{D}\) is measured at the drain side while biasing the drain- terminal at \(0.1\mathrm{V}\) , obtaining the 4 distinct \(I_{ds} - V_{gs}\) curves.  
+
+Simulation methodology All the simulations are performed in the commercial SPICE simulator Cadence Spectre. For simulating the FeFET, a Preisach- based model of the FeFET is considered along with the BSIM- IMG model of the transistor \(^{17,18}\) . To simulate
+
+<--- Page Split --->
+
+the FeFET crossbar array, a single column in the crossbar is first selected. Each cell in the crossbar is a 1FeFET- 1R structure. The ADC connected to the column is simulated using well- calibrated BSIM- IMG transistors based on measured data \(^{28}\) . For all possible input and stored combinations, netlists for the column are generated using a Python script. The netlists are then run in SPICE, and the results are extracted. To include the variability of the FeFET on the MAC output, Monte Carlo simulations are performed following a normal distribution with \(3\sigma\) truncation. 1000 sample Monte- Carlo runs is simulated for each possible input and stored combination. The mean, 5th, and 95th percentile of the sampling voltage are extracted corresponding to each MAC output. Finally, the output is sampled using the ADC. The threshold voltages of the comparators are chosen according to the 2- bit quantization of the neural network. Finally, the output levels are converted to binary values and are used for the neural network simulation.  
+
+To assess the impact of FeFET variability and the DNN quantization on the performed task accuracy, a bit- accurate simulation of the multi- level FeFET macro was implemented using the simulation framework ProxSim \(^{29}\) based on Tensorflow \(^{30}\) . We implemented a custom CUDA operator to simulate the results from different simulations and measurements.
+
+<--- Page Split --->
+
+## Data availability  
+
+The data that support the plots within this paper and other findings of this study are available from the corresponding authors on reasonable request.  
+
+## References  
+
+1. Talati, N. et al. mmpu—a real processing-in-memory architecture to combat the von neumann bottleneck. Applications of Emerging Memory Technology: Beyond Storage 191–213 (2020).  
+
+2. Chang, L. Process-in-memory (pim), near-data-processing (ndp). https://github.com/miglopst/PIM.NDP_papers (2019).  
+
+3. De, S. et al. Demonstration of multiply-accumulate operation with 28 nm fefet cross-bar array. IEEE Electron Device Letters 43, 2081–2084 (2022).  
+
+4. Soliman, T. et al. Ultra-low power flexible precision fefet based analog in-memory computing. In 2020 IEEE International Electron Devices Meeting (IEDM), 29.2.1–29.2.4 (2020).  
+
+5. Soliman, T. et al. Felix: A ferroelectric fet based low power mixed-signal in-memory architecture for dnn acceleration. ACM Transactions on Embedded Computing Systems 21, 1–25 (2022).
+
+<--- Page Split --->
+
+6. Chen, X., Song, T. & Han, Y. Rram-based analog in-memory computing : Invited paper. In 2021 IEEE/ACM International Symposium on Nanoscale Architectures (NANOARCH), 1-6 (2021).  
+
+7. Cai, H. et al. Proposal of analog in-memory computing with magnified tunnel magnetoresistance ratio and universal stt-mram cell. IEEE Transactions on Circuits and Systems I: Regular Papers 69, 1519-1531 (2022).  
+
+8. Mulaosmanovic, H. et al. Ferroelectric field-effect transistors based on hfo2: a review. Nanotechnology 32, 502002 (2021).  
+
+9. Reis, D., Niemier, M. & Hu, X. S. Computing in memory with fefets. In Proceedings of the International Symposium on Low Power Electronics and Design, ISLPED '18 (Association for Computing Machinery, New York, NY, USA, 2018). URL https://doi.org/10.1145/3218603.3218640.  
+
+10. Thomann, S., Nguyen, H. L., Genssler, P. R. & Amrouch, H. All-in-memory brain-inspired computing using fefet synapses. Frontiers in Electronics 3, 3 (2022).  
+
+11. Kazemi, A. et al. Achieving software-equivalent accuracy for hyperdimensional computing with ferroelectric-based in-memory computing. Scientific reports 12, 19201 (2022).  
+
+12. Jerry, M. et al. Ferroelectric fet analog synapse for acceleration of deep neural network training. In 2017 IEEE International Electron Devices Meeting (IEDM), 6.2.1-6.2.4 (2017).
+
+<--- Page Split --->
+
+13. Saito, D. et al. Analog in-memory computing in fefet-based 1t1r array for edge ai applications. In 2021 Symposium on VLSI Technology, 1–2 (2021).  
+
+14. Razavi, B. The strongarm latch [a circuit for all seasons]. IEEE Solid-State Circuits Magazine 7, 12–17 (2015).  
+
+15. Trentzsch, M. et al. A 28nm hkmg super low power embedded nvm technology based on ferroelectric fets. In 2016 IEEE International Electron Devices Meeting (IEDM), 11.5.1–11.5.4 (2016).  
+
+16. Beyer, S. et al. Fefet: A versatile cmos compatible device with game-changing potential. In IMW (2020).  
+
+17. Chatterjee, S. et al. Ferroelectric fdsoi fet modeling for memory and logic applications. Solid-State Electronics 200, 108554 (2023). URL https://www.sciencedirect.com/science/article/pii/S0038110122003264.  
+
+18. Agarwal, H. et al. Bsim-img: Advanced model for fdsoi transistors with back channel inversion. In 2020 4th IEEE Electron Devices Technology & Manufacturing Conference (EDTM), 1–4 (2020).  
+
+19. Müller, F. et al. Multilevel operation of ferroelectric fet memory arrays considering current percolation paths impacting switching behavior. In IEEE Electron Device Letters, vol. 44, 757–760 (2016).  
+
+20. Lecun, Y., Bottou, L., Bengio, Y. & Haffner, P. Gradient-based learning applied to document recognition. Proceedings of the IEEE 86, 2278–2324 (1998).
+
+<--- Page Split --->
+
+21. LECUN, Y. The mnist database of handwritten digits. http://yann.lecun.com/exdb/mnist/ URL https://cir.nii.ac.jp/crid/1571417126193283840.  
+
+22. Yan, B. et al. A 1.041-mb/mm2 27.38-tops/w signed-int8 dynamic-logic-based adc-less sram compute-in-memory macro in 28nm with reconfigurable bitwise operation for ai and embedded applications. In 2022 IEEE International Solid-State Circuits Conference (ISSCC), vol. 65, 188-190 (2022).  
+
+23. Khwa, W.-S. et al. A 40-nm, 2m-cell, 8b-precision, hybrid slc-mlc pcm computing-in-memory macro with 20.5 - 65.0tops/w for tiny-al edge devices. In 2022 IEEE International Solid-State Circuits Conference (ISSCC), vol. 65, 1-3 (2022).  
+
+24. Huo, Q. et al. A computing-in-memory macro based on three-dimensional resistive random-access memory. Nature Electronics 5, 469-477 (2022).  
+
+25. Jung, S. et al. A crossbar array of magnetoresistive memory devices for in-memory computing. Nature 601, 211-216 (2022).  
+
+26. Simonyan, K. et al. Very deep convolutional networks for large-scale image recognition. In Proc. Int. Conf. Learn. Repr. (2015).  
+
+27. Krizhevsky, A. et al. Cifar-10 (2009). URL https://www.cs.toronto.edu/~kriz/cifar.html.  
+
+28. Kumar, S., Chatterjee, S., Dabhi, C. K., Amrouch, H. & Chauhan, Y. S. Novel fdsol-based dynamic xnor logic for ultra-dense highly-efficient computing. In 2022 IEEE International Symposium on Circuits and Systems (ISCAS), 3373-3377 (2022).
+
+<--- Page Split --->
+
+29. Parra, C. D. L. et al. Proxim: Simulation framework for cross-layer approximate dnn optimization. In DATE (2020).  
+
+30. Abadi, M. et al. TensorFlow: Large-scale machine learning on heterogeneous systems (2015). URL https://www.tensorflow.org/.  
+
+31. Mulaosmanovic, H. et al. Investigation of Accumulative Switching in Ferroelectric FETs: Enabling Universal Modeling of the Switching Behavior. IEEE Transactions on Electron Devices 67, 5804-5809 (2020).  
+
+32. Muller, F. et al. Microstructural implications for neuromorphic synapses based on ferroelectric hafnium oxide. In 2021 IEEE International Symposium on Applications of Ferroelectrics (ISAF), 1-4 (IEEE, 5/16/2021 - 5/21/2021).  
+
+## Acknowledgements  
+
+This work has received funding from the ECSEL Joint Undertaking (JU) under grant agreement No 826655 and No 876925. The JU receives support from the European Union's Horizon 2020 research and innovation programme and Belgium, France, Germany, Portugal, Spain, The Netherlands, Switzerland.
+
+<--- Page Split --->
+
+## Author contributions  
+
+T.S., S.C., and H.A. conceived the idea. T.S. and S.C. contributed equally to the work. H.A., Y.C. supervised the analysis at the device and circuit levels. N.W. supervised the analysis at the architecture level. T.K. supervised the experimental demonstration. F.M. performed the circuit/device measurements. S.C., T.S. and N.L. conducted the device and circuit simulations, variation analysis, and architectural- level benchmarking. All authors contributed to the manuscript writing and provided feedback.  
+
+## Competing interests  
+
+The authors declare no competing interests.
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+Supplementarymaterial.pdf
+
+<--- Page Split --->

preprint/preprint__7e12039b3edcb5d3b25d34c354cf8a8eb7554255f59d07a479f069122448bcfc/preprint__7e12039b3edcb5d3b25d34c354cf8a8eb7554255f59d07a479f069122448bcfc_det.mmd

@@ -0,0 +1,441 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 839, 175]]<|/det|>
+# First Demonstration of In-Memory Computing Crossbar using Multi-level Cell FeFET  
+
+<|ref|>text<|/ref|><|det|>[[44, 196, 604, 238]]<|/det|>
+Taha Soliman Robert Bosch GmbH https://orcid.org/0000- 0002- 9421- 9489  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 604, 284]]<|/det|>
+Swetaki Chatterjee University of Stuttgart https://orcid.org/0000- 0002- 2550- 9626  
+
+<|ref|>text<|/ref|><|det|>[[44, 290, 207, 330]]<|/det|>
+Nellie Laleni Fraunhofer IPMS  
+
+<|ref|>text<|/ref|><|det|>[[44, 336, 562, 377]]<|/det|>
+Franz Müller Fraunhofer IPMS https://orcid.org/0000- 0002- 6564- 9121  
+
+<|ref|>text<|/ref|><|det|>[[44, 382, 237, 422]]<|/det|>
+Tobias Kirchner Robert Bosch GmbH  
+
+<|ref|>text<|/ref|><|det|>[[44, 428, 732, 469]]<|/det|>
+Norbert Wehn Rheinland- Pfälzischen Technischen Universität Kaiserslautern- Landau (RPTU)  
+
+<|ref|>text<|/ref|><|det|>[[44, 474, 468, 515]]<|/det|>
+Thomas Kämpfe Fraunhofer Institute for Photonic Microsystems  
+
+<|ref|>text<|/ref|><|det|>[[44, 521, 741, 561]]<|/det|>
+Yogesh Singh Chauhan Indian Institute of Technology Kanpur https://orcid.org/0000- 0002- 3356- 8917  
+
+<|ref|>text<|/ref|><|det|>[[44, 566, 415, 607]]<|/det|>
+Hussam Amrouch ( amrouch@tum.de ) Technical University of Munich (TUM)  
+
+<|ref|>text<|/ref|><|det|>[[44, 650, 102, 667]]<|/det|>
+Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 688, 135, 706]]<|/det|>
+Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 725, 297, 744]]<|/det|>
+Posted Date: May 26th, 2023  
+
+<|ref|>text<|/ref|><|det|>[[44, 764, 473, 783]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 2948718/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 801, 909, 844]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 862, 531, 881]]<|/det|>
+Additional Declarations: There is NO Competing Interest.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 936, 88]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on October 10th, 2023. See the published version at https://doi.org/10.1038/s41467-023-42110-y.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 136, 870, 191]]<|/det|>
+# First Demonstration of In-Memory Computing Crossbar using Multi-level Cell FeFET  
+
+<|ref|>text<|/ref|><|det|>[[141, 213, 863, 283]]<|/det|>
+Taha Soliman \(^{1 + *}\) , Swetaki Chatterjee \(^{2,5 + }\) , Nellie Lalemi \(^{3}\) , Franz Müller \(^{3}\) , Tobias Kirchner \(^{1}\) , Norbert Wehn \(^{4}\) , Thomas Kämpfe \(^{3*}\) , Yogesh Singh Chauhan \(^{5}\) , Hussam Amrouch \(^{6,7*}\)  
+
+<|ref|>text<|/ref|><|det|>[[140, 305, 850, 435]]<|/det|>
+\(^{1}\) Robert Bosch GmbH, Renningen, Germany \(^{2}\) University of Stuttgart, Stuttgart, Germany \(^{3}\) Fraunhofer IPMS, Dresden, Germany \(^{4}\) RPTU Kaiserslautern- Landau, Kaiserlautern, Germany \(^{5}\) Indian Institute of Technology Kanpur, Kanpur, India \(^{6}\) Munich Institute of Robotics and Machine Intelligence, Munich, Germany \(^{7}\) AI Processor Design, Technical University of Munich (TUM), Munich, Germany  
+
+<|ref|>text<|/ref|><|det|>[[113, 444, 881, 504]]<|/det|>
+\(^{+}\) Equal Contribution \(^{*}\) To whom correspondence should be addressed Email: taha.soliman@bosch.de, thomas.kaempfe@ipms.fraunhofer.de, amrouch@tum.de  
+
+<|ref|>text<|/ref|><|det|>[[111, 525, 886, 876]]<|/det|>
+Advancements in AI led to the emergence of in- memory- computing architectures as a promising solution for the associated computing and memory challenges. This study introduces a novel in- memory- computing (IMC) crossbar macro utilizing a multi- level FeFET cell for multi- bit multiply and accumulate (MAC) operations. The proposed 1FeFET- 1R cell design stores multi- bit information while minimizing device variability effects on accuracy. Experimental validation was performed using 28 nm HKMG technology- based ferroelectric field- effect transistor (FeFET) devices. Unlike traditional resistive memory- based analog computing, our approach leverages the electrical characteristics of stored data within the memory cell to derive MAC operation results encoded in activation time and accumulated current. Remarkably, our design achieves
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 88, 886, 294]]<|/det|>
+96.6% accuracy for handwriting recognition and 91.5% accuracy for image classification without extra training. Furthermore, it demonstrates exceptional performance, achieving 885.4 TOPS/W- nearly double that of existing designs. This study represents the first successful implementation of an in- memory macro using a multi- state FeFET cell for complete MAC operations, preserving crossbar density without additional structural overhead.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 338, 253, 360]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[113, 399, 886, 531]]<|/det|>
+Although the demand for data transmission is increasing globally, it is expected that the future will prioritize data- centric local intelligence at the edge node. This, in turn, will inevitably necessitate devices, including wearables, sensors, smartphones, and cars, to locally analyze data and make autonomous decisions.  
+
+<|ref|>text<|/ref|><|det|>[[112, 568, 886, 846]]<|/det|>
+Edge- AI devices have great potential to enable new applications with higher performance and support local embedded intelligence, real- time learning, and autonomy. This could propel the semiconductor industry's next growth phase. Energy- efficient local computing is crucial to enable smart connected Internet of Things (IoT) devices. Further, the rise of memory- intensive computational tasks has led to a significant increase in the amount of data that needs to be accessed compared to local computation inside the algorithmic logic units. This issue is commonly known as the von- Neumann bottleneck \(^{1}\) . One of the most notable examples of such storage- bound tasks is related to AI, specifically
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 88, 884, 185]]<|/det|>
+when it comes to deep learning applications. In recent years, the research community has demonstrated the potential of various architectural improvements in computing systems, such as near- memory or IMC, to meet these energy, compute and memory requirements \(^{2}\) .  
+
+<|ref|>text<|/ref|><|det|>[[111, 220, 885, 499]]<|/det|>
+Similarly, deep neural networks (DNNs) have gained popularity due to their remarkable performance, especially in applications such as speech recognition and image processing. However, in order to design an efficient DNN, various metrics such as throughput, latency and energy efficiency must be jointly optimized. Therefore, researchers have turned their attention to IMC architectures for deploying such networks. Such architectures perform MAC operations by utilizing the memory array without the need for data movement, resulting in improved system performance and large energy savings, thus, overcoming the fundamental bottleneck of von- Neumann architecture.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 90, 886, 595]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 613, 884, 856]]<|/det|>
+<center>Figure 1: Overview of the proposed IMC macro for MAC operations a. The material stack of FeFETs b. The multi-bit FeFET can be programmed to different states to store the weight of the synapse. c. Previous works \(^{3 - 5}\) only considered binary AND or XNOR operation to compute a single-bit multiplication operation. d. Our proposed 2-bit multiplication operation with input encoding and 2-bit storage is shown. The corresponding output activates at different instances of time. e. An encoder provides the gate voltage depending on the input value which changes between three levels at different instances of </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 886, 367]]<|/det|>
+time. f. The multiplication output of the input and stored state in the cell depends on the time at which one cell is activated which is accumulated and sampled using the decoder. g. Depending on the activation time and the number of cells activated at a given time, the voltage across the capacitor connected to a column of cells is accumulated which corresponds linearly with the MAC output and has a minimal impact of the underlying device variation. h. The corresponding MAC operation is performed in the crossbar, accumulating the output in the capacitor voltage. i. IMC accelerators facilitate MAC operations for AI workloads where our proposed design can be utilized.  
+
+<|ref|>text<|/ref|><|det|>[[112, 440, 886, 792]]<|/det|>
+Advanced IMC architecture design relies on the usage of emerging memory cells, particularly non- volatile memory (NVM) cells. These NVMs are used to store weights in neural network inference architectures without needing a steady power supply, and to perform MAC operations using their analog properties. Various emerging memories have been presented, such as resistive RAM (ReRAM), magnetic RAM, and FeFETs \(^{6 - 8}\) . In this study, we focus on using FeFETs as memory cells due to their superior performance compared to other emerging memories \(^{8}\) . Our FeFET is co- integrated into a 28nm high- \(\kappa\) - metal- gate (HKMG) technology, resulting in a low footprint. The FeFET exhibits low read latency ( \(\sim 1\) ns), current source capability, and extremely high write power efficiency ( \(< 1\) fJ) with a short write duration ( \(\sim 1\mu s\) ), making it a highly suitable memory cell for IMC \(^{8}\) .  
+
+<|ref|>text<|/ref|><|det|>[[157, 828, 883, 850]]<|/det|>
+State- of- the- art FeFET based IMC architectures have been limited to binary logical
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 87, 885, 294]]<|/det|>
+operations, specifically logical AND and XNOR \(^{3 - 5,9,10}\) . These operations are restricted to storing only two states within the FeFET memory cell. Even the use of multi-level cell (MLC) FeFETs will not offer any additional advantages in these architectures as the computations are limited to only binary operations. However, MLC FeFET have been utilized for other tasks such as matching in hyperdimensional computing \(^{11}\) . We demonstrate for the first time a multi-bit MAC operation with variation- affected FeFET cells.  
+
+<|ref|>text<|/ref|><|det|>[[111, 330, 886, 570]]<|/det|>
+Contrary to analog computing based implementations of the MAC macro \(^{12,13}\) , we do not perform direct analog multiplication of the input and weight, which is highly prone to variations and require a high degree of linearity in the stored states. Instead, we operate using the current- limited cell such that each cell that is activated has the same current contribution, which limits the impact of variation and improves operation accuracy. This is concisely what enabled us to overcome the variation and utilize MLC FeFET for the first time to demonstrate full MAC operations.  
+
+<|ref|>text<|/ref|><|det|>[[111, 606, 886, 850]]<|/det|>
+We designed and demonstrated a MAC circuit macro consisting of cells connected in the crossbar structure using the 1FeFET- 1R configuration, which includes a single FeFET and a single resistor. In this macro, the FeFET cell acts as a memory for the entire weight value. We explore in this work three dimensions (time, stored threshold voltage (Vth) state of the FeFET, and output current) to perform a complete MAC operation using the single FeFET per weight, detailed in Fig. 1. The input is encoded in applied voltage duration and magnitude, the multi- bit weight is stored in the FeFET, and the output is accumulated as
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 89, 883, 147]]<|/det|>
+the capacitor voltage that depends on the activation time and number of FeFETs activated as shown in Fig. 1.  
+
+<|ref|>text<|/ref|><|det|>[[112, 184, 886, 572]]<|/det|>
+Depending on the stored and input values, the multiplication operation result is encoded in the FeFET cell activation time. Additionally, the number of simultaneously activated FeFET cells is represented by the drain current over time which reflects the accumulation of the performed multiplications. Based on the voltage of the capacitor in the decoder block connected to an operating column of cells as shown in Fig. 1h, the final MAC operation results can be determined and quantized to the required bit precision using a suitable analog- to- digital (ADC) converter. To prove the validity of our macro design, the stored weights, input, as well as output, are quantized to 2 bits. However, the presented macro and computational idea can be extended to higher precision as long as the memory cell can store the targeted precision while maintaining the clear separation between the memory states.  
+
+<|ref|>text<|/ref|><|det|>[[112, 607, 886, 849]]<|/det|>
+We demonstrate in this paper the multiplication operation by a single FeFET cell and extend to complete MAC using the IMC macro of 32x32 FeFETs. This macro is tested against the LeNet network for MNIST handwritten digit database and several layers from VGG- 19 for the CIFAR- 10 dataset. Our presented work maintained the network accuracy at 97% with less than 2% accuracy loss. We highlight how IMC combined with emerging technologies and temporal- based computation can fulfil the increasing AI field computational and memory requirements and overcome the von- Neumann bottleneck.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 88, 405, 111]]<|/det|>
+## Our Proposed Architecture  
+
+<|ref|>text<|/ref|><|det|>[[113, 148, 885, 318]]<|/det|>
+Single- Cell Multiply operation The polarization of the ferroelectric layer can be modulated by the application of voltage pulses which in turn changes the \(\mathrm{V_{th}}\) of the underlying transistor and hence the conductance. The different \(\mathrm{V_{th}}\) states are achieved with respect to the value of the voltage pulse and the pulse duration. Fig. 1b illustrates the MLC FeFET where it is programmed to four different states to store 2- bit information.  
+
+<|ref|>text<|/ref|><|det|>[[112, 353, 886, 594]]<|/det|>
+In this work, we use the 1FeFET- 1R concept (to limit the drain current variability and operate in the saturation region) for a memory cell \(^{4,11}\) . This concept is constructed by adding a drain resistance to the FeFET. The resistance is in the range of \(1M\Omega\) for a drain current of approximately \(100\mathrm{nA}\) and can be implemented by a transistor biased at its gate or a physical resistance \(^{11,13}\) . In this way, the current variability is reduced, as well as the power consumption of the memory unit cell due to the limiting of the drain current to a fixed value, at the expense of the total area.  
+
+<|ref|>text<|/ref|><|det|>[[113, 631, 886, 800]]<|/det|>
+We propose a novel methodology to use the MLC FeFET cell to perform the complete MAC operation. As shown in Fig. 2a, the 2- bit weight is encoded in the stored state inside the FeFET state such that the smallest value is represented by the highest \(\mathrm{V_{th}}\) state and the largest by lowest \(\mathrm{V_{th}}\) state. On the other hand, the input value will be encoded as the time stamp at which the different gate voltages are applied according to Eq. (1)
+
+<--- Page Split --->
+<|ref|>equation<|/ref|><|det|>[[233, 108, 882, 285]]<|/det|>
+\[2\mathrm{bit} - V_{g}(\mathrm{Input}, \mathrm{time}) = \left\{ \begin{array}{ll}0, & \mathrm{if} \mathrm{Input.time} < \epsilon_{1}, \\ V\mathrm{th}1, & \mathrm{if} \epsilon_{1} < = \mathrm{Input.time} < \epsilon_{2}, \\ V\mathrm{th}2, & \mathrm{if} \epsilon_{2} < = \mathrm{Input.time} < \epsilon_{3}, \\ V\mathrm{th}3, & \mathrm{if} \epsilon_{3} < = \mathrm{Input.time}. \end{array} \right. \quad (1)\]  
+
+<|ref|>text<|/ref|><|det|>[[112, 320, 886, 597]]<|/det|>
+According to Eq. (1), the value of the gate voltage applied to the FeFET cell over time is dependent on the 2- bit input value. Consequently, the timestamp at which the FeFET saturates can encode the results of the multiplication of the stored state inside the FeFET as well as the input voltage. However, one of the main issues will be ensuring the commutative property for multiplication. In Eq. (1)), the values \(\epsilon_{1},\epsilon_{2},\epsilon_{3}\) are used to tune the ramping voltage speed to ensure the commutative property. As shown in Fig. 2, we can demonstrate a 2- bit/2- bit multiply operation performed by our memory cell in both simulations and measurements.  
+
+<|ref|>text<|/ref|><|det|>[[112, 634, 886, 876]]<|/det|>
+Multiple- Cell Accumulation Operation To read out the multiple- cell accumulation operation, we utilized the charging of a capacitor due to the total accumulated drain currents in a column over a period of time. The drain current from multiple FeFETs are collected and directed to the accumulation capacitor. The saturation time of the FeFET encodes the multiplication output, and the number of FeFETs activated at a given time represents the accumulation. The capacitor voltage at the end of the period or the sampling time( \(\mathrm{t}_{\mathrm{s}}\) ) is given by
+
+<--- Page Split --->
+<|ref|>equation<|/ref|><|det|>[[343, 117, 881, 166]]<|/det|>
+\[V_{sampling} = \frac{1}{C}\times \int_{0}^{t_{s}}\sum_{n = 1}^{nFeFETS}I_{d,n}(t)dt \quad (2)\]  
+
+<|ref|>text<|/ref|><|det|>[[113, 204, 884, 300]]<|/det|>
+where \(C\) is the capacitance of the capacitor attached to the column, \(I_{d,n}\) is the drain current of the \(n^{th}\) FeFET, \(nFeFETS\) is the total number of FeFETs connected. At that point, the FeFETs are in the saturation region and limited by the resistance according to Eq. (3).  
+
+<|ref|>equation<|/ref|><|det|>[[363, 369, 881, 410]]<|/det|>
+\[I_{d,n}(t) = \frac{V_{dd} - V_{s}(t)}{R_{out}}\cdot h(t - t_{o}) \quad (3)\]  
+
+<|ref|>text<|/ref|><|det|>[[112, 429, 885, 707]]<|/det|>
+\(V_{s}(t)\) is following the charging voltage function over a capacitance \(C\) and time constant \(\tau = R_{out}\times C\) where \(R_{out}\) is the total resistance seen by the source of the FeFET including the current limiter resistance \(R\) . The time- constant \(\tau\) is sufficiently high to not allow a steep fall in the current. \(h(t - t_{o})\) is the step- function at time \(t_{o}\) . It represents the time at which the FeFET is activated and it depends on the input applied and the stored state. The value of \(t_{o}\) is such that it maintains the commutative property of multiplication. Since the current follows a step- jump at \(t_{o}\) , the voltage across the capacitor follows a ramp after time \(t_{o}\) .  
+
+<|ref|>text<|/ref|><|det|>[[113, 744, 884, 875]]<|/det|>
+To quantize the MAC output, we employ StrongArm voltage input comparators \(^{14}\) followed by latches and encoder for a complete 2- bit output. We explored the influence of the value of \(nFeFETS\) and the induced cell variation on the accumulation results and the resulting loss in inference accuracy.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[121, 179, 880, 515]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 538, 884, 779]]<|/det|>
+<center>Figure 2: Direct 2-bit multiply operation in a single cell. a. The measured \(\mathrm{I_{ds} - V_{gs}}\) characteristics of the 1FeFET-1R cell corresponding to the four stored states. b. Input voltage against time applied to the FeFETs for measurement, and c. Output current against time on applying the input pulse. \(\mathrm{I_{ds}}\) rises at different instants of time corresponding to different outputs, which are used to get the product of input and stored states. d.-f. Stored, input, and output from a single cell multiplication operation verified through simulations and determines the maximum speed of operation possible. </center>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 89, 198, 110]]<|/det|>
+## Results  
+
+<|ref|>text<|/ref|><|det|>[[113, 150, 918, 318]]<|/det|>
+Experimental measurements The FeFET test structures are fabricated in the GlobalFoundries \(28\mathrm{nm}\) high- k/metal gate technology node, for which co- integration of FeFETs with CMOS devices has been demonstrated 15. The FeFETs consist of a \(\mathrm{SiO_2}\) interfacial oxide layer, followed by an 8 to \(10\mathrm{nm}\) thick, ferroelectric doped \(\mathrm{HfO_2}\) layer as illustrated in Fig. 1a and Fig. S1a. The gate is capped with a TiN metal cap, and silicided poly silicon 16.  
+
+<|ref|>text<|/ref|><|det|>[[113, 353, 885, 488]]<|/det|>
+A 1FeFET- 1R cell is constructed by externally connecting a \(1\mathrm{M}\Omega\) resistor to a FeFET with an area of \(450\times 450\mathrm{nm}^2\) . This is necessary to limit the ON current and control variations. The 1FeFET- 1R cell is written to 4 distinct states, as shown in Fig. 2a. For details of the methods of writing the FeFET into the desired state, refer to the Methods section.  
+
+<|ref|>text<|/ref|><|det|>[[112, 521, 886, 875]]<|/det|>
+As stated earlier, for a single cell, the input is encoded in duration, and the magnitude of the voltage applied and shown in Fig. 2b. For an input of '0', the gate voltage is kept constant at 0 V. This voltage is less than the \(\mathrm{V_{th}}\) of any stored state in the FeFET, and no FeFET is turned ON. For the other inputs, the magnitude of the voltage is changed to levels \(\mathrm{V_1}\) , \(\mathrm{V_2}\) , and \(\mathrm{V_3}\) at a certain point in time, as stated in Eq. (1). The voltage level corresponds to the read voltage of the FeFETs storing '3', '2', and '1', respectively. Consequently, the FeFET storing '3' is turned on earliest, and a FeFET storing '0' is never turned ON. For input '3', voltage \(\mathrm{V_1}\) is applied at 0, and for input '1', voltage \(\mathrm{V_1}\) is applied later at \(300\mu \mathrm{s}\) . Correspondingly, the FeFET storing '3' would turn on immediately for input '3' and turn on later for input '1'. Hence, the output of the single multiplication operation
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 883, 147]]<|/det|>
+between the stored weight and the applied input is encoded as the time when the FeFET turns on, i.e.- the activation time of the FeFET.  
+
+<|ref|>text<|/ref|><|det|>[[112, 183, 885, 463]]<|/det|>
+Evaluation at different input conditions is performed from the resulting \(I_{ds} - V_{gs}\) transfer characteristics. The corresponding output is shown in Fig. 2c. The verification is done for a drain current of 75 nA. The timing marked for each output state is distinguishable. The current rises for an output of 9 first and at last for an output of 1. For intermediate output states, the activation time is in between. The commutative property is also maintained as seen for output of '2', '3', and '6'. The instance of time when the current rises can distinguish between the different output states, as shown in Fig. 2c. This forms the basis for the MAC operation.  
+
+<|ref|>text<|/ref|><|det|>[[112, 498, 885, 703]]<|/det|>
+Simulation The functionality of the proposed in- memory macro is exemplified through simulations. The FeFET is simulated using a Preisach- based model \(^{17}\) of the Ferroelectric capacitor and industry- standard compact model \(^{18}\) of the underlying transistor (for details, see SI). We use 2- bit storage for the FeFET as in measurements, which corresponds to four different \(\mathrm{V_{th}}\) states. The simulation characteristics are matched to the experiments as closely as possible, and the resulting \(I_{ds} - V_{gs}\) characteristics are shown in Fig. 2d.  
+
+<|ref|>text<|/ref|><|det|>[[113, 739, 886, 871]]<|/det|>
+Correspondingly, we simulate the single- cell multiply operation as in measurements. The timing of the input pulse is modified such as to have almost linear characteristics of the output voltage against the desired output. Also, to determine the maximum speed of operation, the input pulse width is greatly reduced. The values are as per the fast read
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 272, 880, 461]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 483, 884, 686]]<|/det|>
+<center>Figure 3: Accumulation of the output as the voltage across the capacitor a. Structure of the single cell connected to the capacitor. b. The voltage across the capacitor \(\left(V_{\mathrm{cap}}\right)\) against time. As the current for output 9 (inp-3 x stored-3) is turned on first, the voltage across the capacitor at \(t_{\mathrm{sampling}}\) of 14 ns is highest. Similarly, for output 1 (inp-1 x stored-1), the voltage is lowest. c. The sampling voltage across the capacitor \(\left(V_{\mathrm{sampling}}\right)\) at \(t_{\mathrm{sampling}}\) vs the output MAC value. It is linear and perfectly aligned for the same output. </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 884, 255]]<|/det|>
+out proposed in FeFETs \(^{8}\) . The input and the corresponding output are shown in Fig. 2e. and Fig. 2f., respectively. Each output is distinguishable with no overlap, and also, the commutative property of the multiplication is maintained. We further simulated and evaluated the proposed MAC macro for a complete array of connected cells for neural network inference.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 310, 234, 330]]<|/det|>
+## Evaluation  
+
+<|ref|>text<|/ref|><|det|>[[112, 370, 886, 757]]<|/det|>
+Crossbar Level The memory cells are arranged in the crossbar structure. In our recent work \(^{19}\) , we demonstrated that such an array can be programmed with MLC cells for up to 3b precision employing inhibit voltage levels and target erase schemes. A capacitor of 64 fF is connected at the bit- line for each column which is charged and discharged after every cycle. The value of the capacitor is so chosen to allow the charging without being saturated. The voltage to which it is charged depends on the total current flowing into it and the time for which it flows, as given by equation Eq. (2). The time is determined by the input and stored state at which a particular FeFET activates. The number of FeFETs activated at a given time determines the total current that flows into the capacitor. The voltage across the capacitor ( \(\mathrm{V}_{\mathrm{cap}}\) ) is sampled at a particular time ( \(\mathrm{t}_{\mathrm{sampling}}\) ) using a 2- bit ADC to get the final output.  
+
+<|ref|>text<|/ref|><|det|>[[113, 795, 884, 852]]<|/det|>
+The single cell connected to the capacitor is shown in Fig. 3a. The capacitor is charged with an almost constant current of 100 nA, and thus, \(\mathrm{V}_{\mathrm{cap}}\) rises linearly with
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 184]]<|/det|>
+time as shown in Fig. 3b. The sampled voltage is maximum for the case of output 9 because \(\mathrm{I_d}\) turns on earliest in this case (at 1 ns) Fig. 3c. For the case of output 1, voltage is minimum because \(\mathrm{I_d}\) turns on the last (at 13 ns).  
+
+<|ref|>text<|/ref|><|det|>[[112, 220, 885, 426]]<|/det|>
+Similarly, in the case of 2 cells in the array, for output 18, \(\mathrm{V_{cap}}\) is maximum. Here, both the cells are activated at 1 ns, and the current is double that of one cell. Hence, the sampled voltage is also approximately double that of output 9. For output 1, only one of the FeFETs is turned on at 13 ns, and the sampled voltage is minimum. For intermediate values of output ('2', '3', '4', and '6') the sampled voltage at the capacitor lies in between the maximum and minimum as shown in Fig. 3c.  
+
+<|ref|>text<|/ref|><|det|>[[112, 461, 885, 740]]<|/det|>
+The reasoning can be extended to a higher number of cells in the column. The sampled voltage across the capacitor progressively increases with the number of cells activated. Fig. 4 shows the sampled voltage \((\mathrm{V_{sampling}})\) against the MAC output for up to 32 cells in the array. We considered all the possible input and stored combinations to generate the MAC output. We maintained a clear distinction between each output level and the numerical MAC output. However, with an increase in the number of cells, the linearity for higher output values is lost, and the sampling voltage starts to saturate. This is also expected from equation Eq. (2).  
+
+<|ref|>text<|/ref|><|det|>[[113, 775, 884, 871]]<|/det|>
+The final output is converted into 2 bits using strong- ARM comparators \(^{14}\) connected across the capacitor. The comparison threshold levels are selected based on the quantization required. For the final simulations of the DNN, 32 cells in a column are considered.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[123, 250, 880, 550]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 579, 884, 710]]<|/det|>
+<center>Figure 4: Complete MAC output for increasing number of cells connected in the column. a.- f. 1 cell to 32 cells sampled voltage across the capacitor against the MAC output connected in a single column. High degree of linearity is observed, which is desired for the final neural network inference. </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 885, 259]]<|/det|>
+The crossbar is divided into \(32 \times 32\) cells in a segment. Each column in the segment is connected to the 2- bit ADC. The quantized weights are directly written to the FeFET crossbar (For details on the quantization, see SI). The quantized input is applied using a digital- to- analog converter (DAC) connected to the word line for each row, which selects a particular voltage based on the applied input.  
+
+<|ref|>text<|/ref|><|det|>[[112, 294, 886, 535]]<|/det|>
+Afterwards, the variability in the FeFET devices are incorporated in the FeFET model to calculate the loss in inference accuracy of the neural network. The variation in \(\mathrm{V}_{\mathrm{th}}\) is measured from real fabricated FeFET devices. Fig. 5 shows the distribution for three different target condition sets of \(\mathrm{V}_{\mathrm{th}}\) . In the first set, the target levels are chosen as \(0.2 \mathrm{~V}, 0.6 \mathrm{~V}, 1.0 \mathrm{~V}\) and \(1.4 \mathrm{~V}\) . In the next set, the target levels are \(0.3 \mathrm{~V}, 0.633 \mathrm{~V}, 0.967 \mathrm{~V}\) and \(1.3 \mathrm{~V}\) . In the third set, the target levels are \(0.3 \mathrm{~V}, 0.6 \mathrm{~V}, 0.9 \mathrm{~V}\) and \(1.2 \mathrm{~V}\) . The maximum standard deviation of \(38 \mathrm{mV}\) is obtained for the difference in actual and target \(\mathrm{V}_{\mathrm{th}}\) .  
+
+<|ref|>text<|/ref|><|det|>[[112, 571, 886, 850]]<|/det|>
+For simulations, a standard deviation of \(40 \mathrm{mV}\) for \(\mathrm{V}_{\mathrm{th}}\) of the FeFET is assumed for each state. 1000 Monte Carlo samples for each stored, and the input value is simulated for up to 4 cells in the array. For a higher number of cells, the total variability of each state is calculated algebraically from the lower number of cells (for details, see SI). A maximum standard deviation of less than \(4 \mathrm{mV}\) for any given output state is observed in the case of 32 cells in the array. Quantization of the output into 4 levels (2 bits) further reduces the error probability. Finally, the neural network models are simulated to calculate the loss in inference accuracy and derive the performance metrics of the proposed MAC macro.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[131, 97, 867, 264]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 288, 884, 417]]<|/det|>
+<center>Figure 5: Experimentally measured device-to-device variation with respect to \(\mathbf{V}_{\mathrm{th}}\) . a. Target condition 1. b. Target condition 2. c. Target condition 3. Different target conditions were set for measuring the standard deviation in target and measured \(\mathbf{V}_{\mathrm{th}}\) . A maximum standard deviation of \(38\mathrm{mV}\) is observed. </center>  
+
+<|ref|>text<|/ref|><|det|>[[112, 466, 886, 781]]<|/det|>
+Architecture level We used the experimental and simulation data to estimate the performance of the demonstrated macro on two neural network models. We perform the inference of these networks on the target macro and at limited variability of \(40\mathrm{mV}\) based on our measurements. As shown in Fig. 6a,b, we fully quantize the LeNet model \(^{20}\) for MNIST \(^{21}\) into 2- bit activation and 2- bit weights to fit our macro capabilities of MAC precision. The network consists of three convolutional layers and two dense layers. The network requires 397920 MAC operations using 61610 parameters. Considering a max of 40 mv device variation, we achieve \(96.64\%\) network accuracy compared to the original model accuracy of \(99.11\%\) achieved at full floating point precision.  
+
+<|ref|>text<|/ref|><|det|>[[115, 817, 884, 875]]<|/det|>
+Additionally, we also quantize two layers from the VGG19 \(^{26}\) network for the CIFAR- \(^{10}\) \(^{27}\) dataset as shown in Fig. 6c and d. The network consists of 19 layers. However, we
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[118, 143, 880, 480]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 496, 886, 812]]<|/det|>
+<center>Figure 6: Neural network simulation utilizing the proposed IMC MAC macro. a. LeNet neural network is tested for handwritten digit recognition MNIST dataset. All the MAC layers are quantized to 2 bits. b. The weight distribution shows the trained weights when they are re-quantized to 2 bits for the second convolutional layer. An accuracy of \(96.64\%\) is achieved considering device variations. c. VGG19 neural network is tested for object classification CIFAR-10 dataset where only two convolutional layers are quantized for testing. d. The weight distribution shows the trained weights when they are re-quantized to 2 bits for a convolutional layer in VGG19. An accuracy of \(91.55\%\) is achieved under the effects of device-to-device variations. </center>
+
+<--- Page Split --->
+<|ref|>table<|/ref|><|det|>[[156, 134, 840, 400]]<|/det|>
+<|ref|>table_caption<|/ref|><|det|>[[235, 100, 760, 120]]<|/det|>
+Table 1: Comparison of in-memory crossbars with this work   
+
+<table><tr><td>Crossbar</td><td>Memory</td><td>Tech. [nm]</td><td>Freq. [Mhz]</td><td>Power [μW]</td><td>Throughput [GOPS]</td><td>Efficiency [TOPS/W]</td></tr><tr><td>ISSCC&#x27;22a</td><td>SRAM</td><td>28</td><td>333</td><td>-</td><td>-</td><td>438</td></tr><tr><td>SLC-MLC 23</td><td>PCM</td><td>40</td><td>307</td><td>14900</td><td>3900</td><td>261</td></tr><tr><td>nvCIM 24</td><td>RRAM</td><td>55</td><td>1</td><td>0.0322</td><td>0.002</td><td>62.11</td></tr><tr><td>Samsung 25</td><td>MRAM</td><td>28</td><td>11.1</td><td>225</td><td>91.125</td><td>405</td></tr><tr><td>This work</td><td>FeFET</td><td>28</td><td>66</td><td>153.6</td><td>136</td><td>885.4</td></tr></table>
+
+<|ref|>table_footnote<|/ref|><|det|>[[173, 415, 499, 432]]<|/det|>
+a Performance scaled to 2-bit/2-bit precision  
+
+<|ref|>text<|/ref|><|det|>[[112, 502, 886, 671]]<|/det|>
+tested and quantized only two convolutional layers. The networks require 38947914 parameters out of which we quantized only 1179648 parameters to 2- bit, where the rest are quantized to 8- bit. The 2- bit quantization layers use 2- bit quantized activations. We consider a max of 40 mv device variation for those two layers. We achieved \(91.55\%\) network accuracy compared to the original model accuracy of \(93.22\%\) .  
+
+<|ref|>text<|/ref|><|det|>[[112, 708, 886, 876]]<|/det|>
+As demonstrated, the variation has a very limited influence on the accuracy of the final network resulting in less than \(4\%\) in both networks over the testing set. This can be reasoned by the cell architecture of 1FeFET- 1R, which limits the current and the quantization of the final output into 2 bits, which limits the impact of the variation on the sampling voltage thresholds.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 885, 259]]<|/det|>
+In our experiments, the crossbar and the ADCs consume \(153.6 \mu \mathrm{W}\) measured directly for the given frequency (66 MHz) and accumulation capacitance (64 fF). The energy efficiency of the presented crossbar accordingly is \(885.4\) tera- operations per second power watt (TOPS/W), where each operation refers to 2- bit/2- bit multiply or accumulate. We compared our design against existing in- memory crossbars which are shown in Table 1.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 310, 236, 331]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[112, 370, 886, 575]]<|/det|>
+In this work, we have demonstrated the multi- bit MAC operation exploiting for the first time a multi- bit FeFET cell and a novel encoding and decoding scheme. The variability was controlled with the help of the external current- limiting resistor approach. Compared to other works in the literature with different memory architectures, we show a higher throughput and efficiency. The results portray that FeFET can be a strong contender for DNN acceleration with high efficiency.  
+
+<|ref|>text<|/ref|><|det|>[[112, 612, 885, 780]]<|/det|>
+We see an important potential for our computational innovation to allow for a dense IMC macro that can use simple memory cell 1FeFET- 1R while performing multi- bit MAC operation per each cell. Additionally, the combination of several computational dimensions time/voltage/current encodings allow for combating the variability and maintaining the macro efficiency in an unprecedented way.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[114, 88, 213, 110]]<|/det|>
+## Methods  
+
+<|ref|>text<|/ref|><|det|>[[111, 150, 886, 720]]<|/det|>
+1FeFET- 1R Cell characterization To write the FeFET, into 4 distinct states (i.e 2- bit), we employed the write- verify scheme. A FeFET is written into a fully programmed state by applying \(+4.5\mathrm{V}\) for \(500~\mathrm{ns}\) and into a fully erased state by applying - 5 V for \(500~\mathrm{ns}\) while keeping the source and drain terminals grounded. Before evaluation, each FeFET is cycled 50 times with these conditions for preconditioning. A FeFET is then written to 4 distinct states using a write- verify- scheme. Therefore, the FeFET is initially fully erased. Starting with a write voltage of \(1.4\mathrm{V}\) for \(200~\mathrm{ns}\) , the FeFET is gradually programmed. Write voltage is incremented in steps of \(40\mathrm{mV}\) . After each write pulse, a delay of \(500~\mathrm{ms}\) is applied for charge detrapping, and a read operation verifies the state. This scheme is continued until the target value is reached. Target levels are selected at \(0.3\mathrm{V}\) , \(0.7\mathrm{V}\) , \(1.1\mathrm{V}\) , and \(1.5\mathrm{V}\) at a constant current condition of \(80\mathrm{nA}\) . After setting the target state, a final readout is performed. Before reading, a sufficiently large time is waited for any charge detrapping. In this case, a delay time of 2 seconds is chosen. For reading a voltage ramp, \(V_{\mathrm{G}}\) from - 0.2 V to \(1.7\mathrm{V}\) in steps of \(10\mathrm{mV}\) is applied to the gate. Current \(I_{D}\) is measured at the drain side while biasing the drain- terminal at \(0.1\mathrm{V}\) , obtaining the 4 distinct \(I_{ds} - V_{gs}\) curves.  
+
+<|ref|>text<|/ref|><|det|>[[113, 755, 884, 848]]<|/det|>
+Simulation methodology All the simulations are performed in the commercial SPICE simulator Cadence Spectre. For simulating the FeFET, a Preisach- based model of the FeFET is considered along with the BSIM- IMG model of the transistor \(^{17,18}\) . To simulate
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 88, 886, 514]]<|/det|>
+the FeFET crossbar array, a single column in the crossbar is first selected. Each cell in the crossbar is a 1FeFET- 1R structure. The ADC connected to the column is simulated using well- calibrated BSIM- IMG transistors based on measured data \(^{28}\) . For all possible input and stored combinations, netlists for the column are generated using a Python script. The netlists are then run in SPICE, and the results are extracted. To include the variability of the FeFET on the MAC output, Monte Carlo simulations are performed following a normal distribution with \(3\sigma\) truncation. 1000 sample Monte- Carlo runs is simulated for each possible input and stored combination. The mean, 5th, and 95th percentile of the sampling voltage are extracted corresponding to each MAC output. Finally, the output is sampled using the ADC. The threshold voltages of the comparators are chosen according to the 2- bit quantization of the neural network. Finally, the output levels are converted to binary values and are used for the neural network simulation.  
+
+<|ref|>text<|/ref|><|det|>[[112, 549, 885, 715]]<|/det|>
+To assess the impact of FeFET variability and the DNN quantization on the performed task accuracy, a bit- accurate simulation of the multi- level FeFET macro was implemented using the simulation framework ProxSim \(^{29}\) based on Tensorflow \(^{30}\) . We implemented a custom CUDA operator to simulate the results from different simulations and measurements.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 88, 297, 111]]<|/det|>
+## Data availability  
+
+<|ref|>text<|/ref|><|det|>[[114, 150, 884, 208]]<|/det|>
+The data that support the plots within this paper and other findings of this study are available from the corresponding authors on reasonable request.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 260, 235, 281]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[123, 320, 884, 415]]<|/det|>
+1. Talati, N. et al. mmpu—a real processing-in-memory architecture to combat the von neumann bottleneck. Applications of Emerging Memory Technology: Beyond Storage 191–213 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[123, 454, 882, 512]]<|/det|>
+2. Chang, L. Process-in-memory (pim), near-data-processing (ndp). https://github.com/miglopst/PIM.NDP_papers (2019).  
+
+<|ref|>text<|/ref|><|det|>[[123, 551, 882, 609]]<|/det|>
+3. De, S. et al. Demonstration of multiply-accumulate operation with 28 nm fefet cross-bar array. IEEE Electron Device Letters 43, 2081–2084 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[123, 647, 884, 743]]<|/det|>
+4. Soliman, T. et al. Ultra-low power flexible precision fefet based analog in-memory computing. In 2020 IEEE International Electron Devices Meeting (IEDM), 29.2.1–29.2.4 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[123, 781, 884, 876]]<|/det|>
+5. Soliman, T. et al. Felix: A ferroelectric fet based low power mixed-signal in-memory architecture for dnn acceleration. ACM Transactions on Embedded Computing Systems 21, 1–25 (2022).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[121, 88, 884, 185]]<|/det|>
+6. Chen, X., Song, T. & Han, Y. Rram-based analog in-memory computing : Invited paper. In 2021 IEEE/ACM International Symposium on Nanoscale Architectures (NANOARCH), 1-6 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[121, 216, 884, 313]]<|/det|>
+7. Cai, H. et al. Proposal of analog in-memory computing with magnified tunnel magnetoresistance ratio and universal stt-mram cell. IEEE Transactions on Circuits and Systems I: Regular Papers 69, 1519-1531 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[121, 344, 883, 404]]<|/det|>
+8. Mulaosmanovic, H. et al. Ferroelectric field-effect transistors based on hfo2: a review. Nanotechnology 32, 502002 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[121, 434, 884, 567]]<|/det|>
+9. Reis, D., Niemier, M. & Hu, X. S. Computing in memory with fefets. In Proceedings of the International Symposium on Low Power Electronics and Design, ISLPED '18 (Association for Computing Machinery, New York, NY, USA, 2018). URL https://doi.org/10.1145/3218603.3218640.  
+
+<|ref|>text<|/ref|><|det|>[[114, 598, 883, 659]]<|/det|>
+10. Thomann, S., Nguyen, H. L., Genssler, P. R. & Amrouch, H. All-in-memory brain-inspired computing using fefet synapses. Frontiers in Electronics 3, 3 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[114, 689, 883, 787]]<|/det|>
+11. Kazemi, A. et al. Achieving software-equivalent accuracy for hyperdimensional computing with ferroelectric-based in-memory computing. Scientific reports 12, 19201 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[114, 818, 883, 877]]<|/det|>
+12. Jerry, M. et al. Ferroelectric fet analog synapse for acceleration of deep neural network training. In 2017 IEEE International Electron Devices Meeting (IEDM), 6.2.1-6.2.4 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 884, 150]]<|/det|>
+13. Saito, D. et al. Analog in-memory computing in fefet-based 1t1r array for edge ai applications. In 2021 Symposium on VLSI Technology, 1–2 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[112, 172, 884, 233]]<|/det|>
+14. Razavi, B. The strongarm latch [a circuit for all seasons]. IEEE Solid-State Circuits Magazine 7, 12–17 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[112, 255, 884, 350]]<|/det|>
+15. Trentzsch, M. et al. A 28nm hkmg super low power embedded nvm technology based on ferroelectric fets. In 2016 IEEE International Electron Devices Meeting (IEDM), 11.5.1–11.5.4 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[112, 373, 883, 434]]<|/det|>
+16. Beyer, S. et al. Fefet: A versatile cmos compatible device with game-changing potential. In IMW (2020).  
+
+<|ref|>text<|/ref|><|det|>[[112, 456, 884, 555]]<|/det|>
+17. Chatterjee, S. et al. Ferroelectric fdsoi fet modeling for memory and logic applications. Solid-State Electronics 200, 108554 (2023). URL https://www.sciencedirect.com/science/article/pii/S0038110122003264.  
+
+<|ref|>text<|/ref|><|det|>[[112, 577, 884, 675]]<|/det|>
+18. Agarwal, H. et al. Bsim-img: Advanced model for fdsoi transistors with back channel inversion. In 2020 4th IEEE Electron Devices Technology & Manufacturing Conference (EDTM), 1–4 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[112, 697, 884, 793]]<|/det|>
+19. Müller, F. et al. Multilevel operation of ferroelectric fet memory arrays considering current percolation paths impacting switching behavior. In IEEE Electron Device Letters, vol. 44, 757–760 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[112, 816, 884, 876]]<|/det|>
+20. Lecun, Y., Bottou, L., Bengio, Y. & Haffner, P. Gradient-based learning applied to document recognition. Proceedings of the IEEE 86, 2278–2324 (1998).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 884, 150]]<|/det|>
+21. LECUN, Y. The mnist database of handwritten digits. http://yann.lecun.com/exdb/mnist/ URL https://cir.nii.ac.jp/crid/1571417126193283840.  
+
+<|ref|>text<|/ref|><|det|>[[112, 172, 884, 303]]<|/det|>
+22. Yan, B. et al. A 1.041-mb/mm2 27.38-tops/w signed-int8 dynamic-logic-based adc-less sram compute-in-memory macro in 28nm with reconfigurable bitwise operation for ai and embedded applications. In 2022 IEEE International Solid-State Circuits Conference (ISSCC), vol. 65, 188-190 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[112, 328, 884, 424]]<|/det|>
+23. Khwa, W.-S. et al. A 40-nm, 2m-cell, 8b-precision, hybrid slc-mlc pcm computing-in-memory macro with 20.5 - 65.0tops/w for tiny-al edge devices. In 2022 IEEE International Solid-State Circuits Conference (ISSCC), vol. 65, 1-3 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[112, 448, 884, 509]]<|/det|>
+24. Huo, Q. et al. A computing-in-memory macro based on three-dimensional resistive random-access memory. Nature Electronics 5, 469-477 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[112, 532, 883, 593]]<|/det|>
+25. Jung, S. et al. A crossbar array of magnetoresistive memory devices for in-memory computing. Nature 601, 211-216 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[112, 616, 883, 677]]<|/det|>
+26. Simonyan, K. et al. Very deep convolutional networks for large-scale image recognition. In Proc. Int. Conf. Learn. Repr. (2015).  
+
+<|ref|>text<|/ref|><|det|>[[112, 699, 883, 758]]<|/det|>
+27. Krizhevsky, A. et al. Cifar-10 (2009). URL https://www.cs.toronto.edu/~kriz/cifar.html.  
+
+<|ref|>text<|/ref|><|det|>[[112, 781, 884, 876]]<|/det|>
+28. Kumar, S., Chatterjee, S., Dabhi, C. K., Amrouch, H. & Chauhan, Y. S. Novel fdsol-based dynamic xnor logic for ultra-dense highly-efficient computing. In 2022 IEEE International Symposium on Circuits and Systems (ISCAS), 3373-3377 (2022).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 90, 884, 148]]<|/det|>
+29. Parra, C. D. L. et al. Proxim: Simulation framework for cross-layer approximate dnn optimization. In DATE (2020).  
+
+<|ref|>text<|/ref|><|det|>[[112, 209, 884, 268]]<|/det|>
+30. Abadi, M. et al. TensorFlow: Large-scale machine learning on heterogeneous systems (2015). URL https://www.tensorflow.org/.  
+
+<|ref|>text<|/ref|><|det|>[[112, 330, 884, 425]]<|/det|>
+31. Mulaosmanovic, H. et al. Investigation of Accumulative Switching in Ferroelectric FETs: Enabling Universal Modeling of the Switching Behavior. IEEE Transactions on Electron Devices 67, 5804-5809 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[112, 486, 886, 581]]<|/det|>
+32. Muller, F. et al. Microstructural implications for neuromorphic synapses based on ferroelectric hafnium oxide. In 2021 IEEE International Symposium on Applications of Ferroelectrics (ISAF), 1-4 (IEEE, 5/16/2021 - 5/21/2021).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 684, 332, 706]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[112, 744, 884, 876]]<|/det|>
+This work has received funding from the ECSEL Joint Undertaking (JU) under grant agreement No 826655 and No 876925. The JU receives support from the European Union's Horizon 2020 research and innovation programme and Belgium, France, Germany, Portugal, Spain, The Netherlands, Switzerland.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 88, 345, 110]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[112, 148, 886, 355]]<|/det|>
+T.S., S.C., and H.A. conceived the idea. T.S. and S.C. contributed equally to the work. H.A., Y.C. supervised the analysis at the device and circuit levels. N.W. supervised the analysis at the architecture level. T.K. supervised the experimental demonstration. F.M. performed the circuit/device measurements. S.C., T.S. and N.L. conducted the device and circuit simulations, variation analysis, and architectural- level benchmarking. All authors contributed to the manuscript writing and provided feedback.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 405, 335, 428]]<|/det|>
+## Competing interests  
+
+<|ref|>text<|/ref|><|det|>[[115, 468, 494, 488]]<|/det|>
+The authors declare no competing interests.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 323, 150]]<|/det|>
+Supplementarymaterial.pdf
+
+<--- Page Split --->

preprint/preprint__7e1438f59389b4c978bcf7a8264b8f610836b4ff4052a498f0dc86d20496081f/images_list.json

@@ -0,0 +1,77 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        85,
+        130,
+        884,
+        737
+      ]
+    ],
+    "page_idx": 43
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        75,
+        70,
+        827,
+        788
+      ]
+    ],
+    "page_idx": 45
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        80,
+        177,
+        910,
+        875
+      ]
+    ],
+    "page_idx": 46
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        70,
+        60,
+        820,
+        789
+      ]
+    ],
+    "page_idx": 48
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5",
+    "footnote": [],
+    "bbox": [
+      [
+        62,
+        203,
+        920,
+        920
+      ]
+    ],
+    "page_idx": 49
+  }
+]

preprint/preprint__7e1c473154ced3f4c26c8b6bc6a1ccd24c3e92a096ba53571746bb619bcedd31/preprint__7e1c473154ced3f4c26c8b6bc6a1ccd24c3e92a096ba53571746bb619bcedd31_det.mmd

@@ -0,0 +1,793 @@
+<|ref|>title<|/ref|><|det|>[[44, 106, 904, 175]]<|/det|>
+# Non-canonical functions of UHRF1 maintain DNA methylation homeostasis in cancer cells  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 245, 214]]<|/det|>
+Pierre Antoine Defossez  
+
+<|ref|>text<|/ref|><|det|>[[55, 222, 433, 241]]<|/det|>
+pierre- antoine.defossez@u- paris.fr  
+
+<|ref|>text<|/ref|><|det|>[[44, 268, 600, 288]]<|/det|>
+CNRS/Univ Paris Cite https://orcid.org/0000- 0002- 6463- 9263  
+
+<|ref|>text<|/ref|><|det|>[[44, 293, 600, 335]]<|/det|>
+Kosuke Yamaguchi CNRS/Univ Paris Cite https://orcid.org/0000- 0003- 2926- 9444  
+
+<|ref|>text<|/ref|><|det|>[[44, 340, 245, 380]]<|/det|>
+Xiaoying Chen CNRS/Univ Paris Cite  
+
+<|ref|>text<|/ref|><|det|>[[44, 386, 245, 427]]<|/det|>
+Brianna Rodgers CNRS/Univ Paris Cite  
+
+<|ref|>text<|/ref|><|det|>[[44, 432, 572, 473]]<|/det|>
+Fumihito Miura Kyushu University https://orcid.org/0000- 0003- 2656- 486X  
+
+<|ref|>text<|/ref|><|det|>[[44, 478, 234, 519]]<|/det|>
+Pavel Bashtrykov Universitat Stuttgart  
+
+<|ref|>text<|/ref|><|det|>[[44, 525, 260, 565]]<|/det|>
+Laure Ferry Universite Paris Diderot  
+
+<|ref|>text<|/ref|><|det|>[[44, 571, 245, 611]]<|/det|>
+Olivier Kirsh CNRS/Univ Paris Cite  
+
+<|ref|>text<|/ref|><|det|>[[44, 617, 245, 657]]<|/det|>
+Marthe Laisne CNRS/Univ Paris Cite  
+
+<|ref|>text<|/ref|><|det|>[[44, 663, 222, 703]]<|/det|>
+Frederic Bonhomme Institut Pasteur  
+
+<|ref|>text<|/ref|><|det|>[[44, 709, 855, 772]]<|/det|>
+Andrea Scelfo Institut Curie, PSL Research University, CNRS, UMR 144, 26 rue d'Ulm, F- 75005, Paris, France. https://orcid.org/0000- 0001- 8767- 0688  
+
+<|ref|>text<|/ref|><|det|>[[44, 778, 270, 819]]<|/det|>
+Catalina Salinas- Luypaert UMR 144/Institut Curie  
+
+<|ref|>text<|/ref|><|det|>[[44, 825, 408, 866]]<|/det|>
+Enes Ugur Ludwig Maximilian University of Munich  
+
+<|ref|>text<|/ref|><|det|>[[44, 871, 216, 911]]<|/det|>
+Paola Arimondo Institut Pasteur  
+
+<|ref|>text<|/ref|><|det|>[[44, 917, 212, 936]]<|/det|>
+Heinrich Leonhardt
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[52, 45, 763, 65]]<|/det|>
+Ludwig Maximilian University of Munich https://orcid.org/0000- 0002- 5086- 6449  
+
+<|ref|>text<|/ref|><|det|>[[44, 70, 668, 110]]<|/det|>
+Masato Kanemaki National Institute of Genetics https://orcid.org/0000- 0002- 7657- 1649  
+
+<|ref|>text<|/ref|><|det|>[[44, 116, 525, 157]]<|/det|>
+Daniele Fachinetti Institut Curie https://orcid.org/0000- 0002- 8795- 6771  
+
+<|ref|>text<|/ref|><|det|>[[44, 163, 589, 204]]<|/det|>
+Albert Jeltsch Universitat Stuttgart https://orcid.org/0000- 0001- 6113- 9290  
+
+<|ref|>text<|/ref|><|det|>[[44, 210, 794, 273]]<|/det|>
+Takashi Ito Department of Biochemistry, Kyushu University Graduate School of Medical Sciences https://orcid.org/0000- 0001- 6097- 2803  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 314, 102, 332]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 352, 135, 370]]<|/det|>
+Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 389, 296, 409]]<|/det|>
+Posted Date: July 26th, 2023  
+
+<|ref|>text<|/ref|><|det|>[[44, 428, 473, 448]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 3154646/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 465, 910, 508]]<|/det|>
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 526, 530, 546]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 581, 950, 625]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on April 5th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 47314- 4.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[133, 85, 866, 124]]<|/det|>
+# Non-canonical functions of UHRF1 maintain DNA methylation homeostasis in cancer cells  
+
+<|ref|>text<|/ref|><|det|>[[118, 158, 882, 228]]<|/det|>
+Kosuke Yamaguchi \(^{1,*}\) , Xiaoying Chen \(^{1}\) , Brianna Rodgers \(^{1}\) , Fumihito Miura \(^{2}\) , Pavel Bashtrykov \(^{3}\) , Laure Ferry \(^{1}\) , Olivier Kirsh \(^{1}\) , Marthe Laisné \(^{1}\) , Frédéric Bonhomme \(^{4}\) , Catalina Salinas- Luypaert \(^{5}\) , Andrea Scelfo \(^{5}\) , Enes Ugur \(^{6}\) , Paola B. Arimondo \(^{4}\) , Heinrich Leonhardt \(^{6}\) , Masato T. Kanemaki \(^{7}\) , Daniele Fachinetti \(^{5}\) , Albert Jeltsch \(^{3}\) , Takashi Ito \(^{2}\) , Pierre- Antoine Defossez \(^{1,*}\)  
+
+<|ref|>text<|/ref|><|det|>[[117, 245, 855, 472]]<|/det|>
+1: Université Paris Cité, CNRS, Epigenetics and Cell Fate, F- 75013 Paris, France.  
+2: Department of Biochemistry, Kyushu University Graduate School of Medical Sciences, Fukuoka, Fukuoka, 812- 8582, Japan.  
+3: Institute of Biochemistry and Technical Biochemistry, Department of Biochemistry, University of Stuttgart, Allmandring 31, 70569 Stuttgart, Germany.  
+4: Institut Pasteur, Université Paris Cité, Epigenetic Chemical Biology, UMR 3523, CNRS, 75724 Paris, France  
+5: Institut Curie, PSL Research University, CNRS, UMR 144, Paris, France.  
+6: Faculty of Biology and Center for Molecular Biosystems (BioSysM), Human Biology and Biolmaging, Ludwig- Maximilians- Universität München, Munich 81377, Germany.  
+7: Department of Chromosome Science, National Institute of Genetics, Research Organization of Information and Systems (ROIS), Yata 1111, Mishima, Shizuoka, 411- 8540, Japan.  
+
+<|ref|>text<|/ref|><|det|>[[118, 507, 745, 542]]<|/det|>
+\* Authors for correspondence: yamako0801@icloud.com, pierre- antoine.defossez@cnrs.fr
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 231, 100]]<|/det|>
+## Abstract new  
+
+<|ref|>text<|/ref|><|det|>[[118, 118, 883, 310]]<|/det|>
+DNA methylation is an essential epigenetic chromatin modification, and its maintenance in mammals requires the protein UHRF1. It is yet unclear if UHRF1 functions solely by stimulating DNA methylation maintenance by DNMT1, or if it has important additional functions. Using degron alleles, we show that UHRF1 depletion causes a much greater loss of DNA methylation than DNMT1 depletion. This is not caused by passive demethylation as UHRF1- depleted cells proliferate more slowly than DNMT1- depleted cells. Instead, bioinformatics, proteomics and genetics experiments establish that UHRF1, besides activating DNMT1, interacts with DNMT3A and DNMT3B and promotes their activity. In addition, we show that UHRF1 antagonizes active DNA demethylation by TET2. Therefore, UHRF1 has non- canonical roles that contribute importantly to DNA methylation homeostasis; these findings have practical implications for epigenetics in health and disease.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 225, 100]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[118, 102, 881, 170]]<|/det|>
+DNA methylation is an essential epigenetic mark in mammals. The methylation of cytosines, mostly in the CpG context, ensures the proper regulation of imprinted and tissue- specific genes, silences repeated elements, and contributes to the function of key functional elements of the genome such as centromeres \(^{1,2}\) .  
+
+<|ref|>text<|/ref|><|det|>[[118, 172, 882, 276]]<|/det|>
+The DNA methylation pattern observed in mammalian tissues is the result of a dynamic process. First, most of the cytosine methylation brought by the gametes is erased in early development, in a process that involves active demethylation by the TET enzymes \(^{3}\) . Then, the proper tissue- and cell- specific methyl marks are re- established in the embryo starting at the time of implantation. This re- establishment of DNA methylation depends on "de novo" methyltransferases, of which two exist in humans: DNMT3A and DNMT3B \(^{4}\) .  
+
+<|ref|>text<|/ref|><|det|>[[118, 277, 882, 466]]<|/det|>
+Even after cells have acquired their proper DNA methylation pattern, the overall stability of this pattern depends on a dynamic equilibrium of gains and losses of cytosine methylation. There can be local losses of DNA methylation due to TET activity, compensated by de novo DNA methylation, in the cell types that do express DNMT3A or DNMT3B. In addition, there is a global remodeling of DNA methylation at the time of DNA replication. Indeed, at this point, the two parental strands of DNA carrying cytosine methylation are separated, and each is used as a template for the synthesis of a daughter strand, which is initially totally devoid of cytosine methylation. It follows that every CpG that was symmetrically methylated before replication becomes hemimethylated. The process whereby the hemimethylated sites return to a fully methylated state is called "maintenance DNA methylation", and it involves two key actors: DNMT1 and UHRF1 \(^{5}\) .  
+
+<|ref|>text<|/ref|><|det|>[[118, 467, 882, 588]]<|/det|>
+The first crucial participant in maintenance DNA methylation is the enzyme DNMT1 \(^{6}\) . Unlike the de novo methyltransferases, DNMT1 is expressed in every replicating cell, and it has higher DNA methyltransferase activity on hemimethylated than on unmethylated sites. This specificity of DNMT1 comes in part from intramolecular inhibitions, which have to be lifted for the enzyme to come into action \(^{7}\) . Some of the molecular mechanisms contributing to lifting this inhibition after DNA replication have been uncovered, and they involve the protein UHRF1 \(^{8 - 10}\) .  
+
+<|ref|>text<|/ref|><|det|>[[118, 589, 882, 797]]<|/det|>
+UHRF1 has an SRA domain that binds DNA with a preference for hemimethylated CpGs \(^{11}\) . It also has a Tandem Tudor Domain (TTD) which, together with the adjoining PHD domain, binds histone H3K9me3 \(^{12}\) . In addition, the TTD domain binds an H3K9me3- like motif within DNA Ligase 1 (LIG1), which ligates Okazaki fragments on the lagging strand \(^{13,14}\) . These different interactions contribute to the recruitment of UHRF1 to replicating chromatin, where it can then modify histones. Its Ubiquitin- Like (Ubl) domain cooperates with its RING finger \(^{15,16}\) , which then targets histone H3 for mono- ubiquitination at two positions, H3K14 and H3K18 \(^{17}\) . The H3K14Ub/K18Ub then binds with high affinity to the RFTs domain of DNMT1, relieving the auto- inhibition \(^{18}\) . In a similar fashion, UHRF1 also mono- ubiquitinates the PCNA- associated factor PAF15, which can then bind the RFTs, freeing the catalytic domain of DNMT1 \(^{19}\) . To summarize, there is incontrovertible evidence that UHRF1 is an upstream activator of DNMT1, yet these advances leave some important questions open.  
+
+<|ref|>text<|/ref|><|det|>[[118, 798, 882, 902]]<|/det|>
+One such question is whether UHRF1 controls DNA methylation only by acting on DNMT1, or whether it also impinges on other epigenetic actors. Besides its importance for the biology of normal cells, this question is especially relevant for cancer. Indeed, the DNA methylation pattern of cancer cells has characteristic abnormalities, marked by global hypomethylation and focal hypermethylation \(^{20}\) , and these abnormalities are likely caused, at least in part, by imperfect DNA methylation maintenance \(^{21}\) . In parallel, most tumors express
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 881, 153]]<|/det|>
+high levels of UHRF1<sup>22</sup>, overexpression of UHRF1 is oncogenic<sup>22</sup>, and UHRF1 is necessary for colon cancer cells to maintain their DNA methylation pattern and survive<sup>23,24</sup>. Therefore, UHRF1 is a key regulator of the cancer epigenome, and it is important to elucidate its role, both for basic research and for medical purposes.  
+
+<|ref|>text<|/ref|><|det|>[[118, 154, 881, 205]]<|/det|>
+Therefore, the questions we address in this paper are: how does UHRF1 control DNA methylation in human cancer cells? Does it only stimulate DNMT1 or does it have other functions? If yes, which one(s)?  
+
+<|ref|>text<|/ref|><|det|>[[118, 207, 882, 361]]<|/det|>
+The model we chose to investigate the question in is colorectal cancer, a prevalent disease in which the contribution of epigenetic is solidly established. Earlier studies have yielded valuable information<sup>23,25</sup>, but some of their conclusions have suffered from technical limitations. In particular, the loss- of- function approaches have been imperfect: siRNA has effects that are asynchronous, limited in time, and sometimes partial; shRNA can be partial or select for cells with the least depletion; constitutional knock- outs can lead to adaptation; whereas inducible knock- outs have delayed kinetics. In contrast, degron alleles have emerged as very powerful tools for loss- of- function studies, permitting rapid, total, and synchronous depletion of proteins of interest in cells<sup>26</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 362, 882, 500]]<|/det|>
+We have generated and validated degron alleles of UHRF1 and/or DNMT1 in human colorectal cancer cell lines. We then used genomics and bioinformatics to precisely describe the DNA demethylation dynamics in these cells, leading to the conclusion that UHRF1 maintains DNA methylation in cancer cells not only by stimulating DNMT1. Proteomics and genetics lead us to conclude that UHRF1 regulates DNMT3A, DNMT3B and TET2 activity in addition to regulating DNMT1. The tools we have developed will be valuable for future research efforts, and our results advance our understanding of cancer epigenetics, with potentially important therapeutic applications.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 180, 100]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 118, 820, 135]]<|/det|>
+## Establishment of degron alleles for UHRF1 and DNMT1 in colorectal cancer cell lines  
+
+<|ref|>text<|/ref|><|det|>[[118, 136, 882, 222]]<|/det|>
+To investigate the respective roles of DNMT1 and UHRF1 in cancer cells, we chose as a model the human colorectal cell lines HCT116 and DLD1, as they have been widely used to study the genetic and epigenetic events that cause and sustain transformation. Both lines have an activated Kras and microsatellite instability but maintain a near- diploid karyotype. HCT116 cells have functional p53, whereas DLD1 cells have mutated p53<sup>27</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 224, 882, 310]]<|/det|>
+In these cells, we utilized the Auxin- Inducible Degron (AID) system to perform precisely controlled, rapid, and synchronous loss- of- function experiments<sup>26</sup>. To prevent unwanted degradation of the target proteins in basal conditions, we employed HCT116 with a doxycycline- inducible OsTIR1<sup>28</sup>, while we used the recently optimized F74G variant of OsTIR1 in the DLD1 background<sup>29,30</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 311, 882, 450]]<|/det|>
+Using Cas9- mediated knock- in, we introduced the tags into the endogenous UHRF1 and/or DNMT1 genes in the HCT116 and DLD1 cell lines, and both genes simultaneously in HCT116 (Fig. 1A- B, Fig. S1A). As UHRF1 can be inactivated by N- terminal modifications<sup>15,16</sup>, we inserted the AID tag at the C- terminus along with the green fluorescent protein, mClover (Fig. 1A). In contrast, N- terminal tagging of DNMT1 can be used to generate a degron allele<sup>31</sup>. For this reason, we placed the AID tag at the N- terminus of DNMT1, accompanied by the red fluorescent protein mRuby2 (Fig. 1A). Three independent clones were generated for each construct and used in further experiments (Fig. 1C, Fig. S1B).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 468, 575, 485]]<|/det|>
+## Characterization and validation of the tagged cell lines  
+
+<|ref|>text<|/ref|><|det|>[[118, 486, 882, 606]]<|/det|>
+Having obtained the lines of the desired genotypes, we then characterized them by growth assays, microscopy, and DNA methylation measurements. In the absence of auxin, the UHRF1- AID, DNMT1- AID, or UHRF1- AID/DNMT1- AID cells grew indistinguishably from the parental HCT116 or DLD1 cells (Fig. S1C- E). We next examined the localization of tagged UHRF1 and DNMT1. In fixed cells, both proteins were nuclear with some colocalizing foci (Fig. 1D). In live- cell microscopy, we found, as expected, that DNMT1 and UHRF1 had a dynamic nuclear distribution and formed colocalizing foci during S phase (Supplemental Movies 1- 3).  
+
+<|ref|>text<|/ref|><|det|>[[118, 607, 882, 764]]<|/det|>
+We further verified the functionality of the tagged proteins by measuring DNA methylation levels in HCT116 derivatives by 3 independent methods: a restriction- enzyme- based assay (LUMA), liquid chromatography followed by tandem mass spectrometry (LC- MS/MS), and whole genome bisulfite sequencing (WGBS). These data showed no significant difference between parental and single AID- tagged cells in HCT116, yet the compound UHRF1- AID/DNMT1- AID line showed \(\sim 10\%\) less DNA methylation than its wild- type counterpart (Fig. 1E). We also carried out LUMA in the DLD1 derivatives and found that the UHRF1- AID and DNMT1- AID cells had a small but significant reduction of DNA methylation (6% less than in the WT, Fig. S1F).  
+
+<|ref|>text<|/ref|><|det|>[[118, 765, 881, 815]]<|/det|>
+Collectively these results confirm that the tags added to UHRF1 and DNMT1 do not measurably affect cell viability, growth, or nuclear localization, and have minimal effects on DNA methylation, therefore validating their use for functional analyses.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 833, 757, 850]]<|/det|>
+## The depletion of UHRF1 and/or DNMT1 is efficient and causes growth arrest  
+
+<|ref|>text<|/ref|><|det|>[[118, 851, 881, 902]]<|/det|>
+After validating these basal conditions, we next tested the effects of triggering the degradation of UHRF1 and/or DNMT1 in the AID- tagged cell lines. Western blotting revealed that, as early as two hours after treatment with auxin, UHRF1 and/or DNMT1 protein levels in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 882, 170]]<|/det|>
+HCT116 and DLD1 cells became undetectable, and that this depletion persisted as long as auxin was present (Fig. 2A; Fig. S2A). We have noted in 3 independent clones that the degradation of DNMT1 and UHRF1 in the compound mutant cells is equally rapid but incomplete by \(\sim 8\) hour after treatment with auxin, for reasons that are yet undetermined (Fig. 2A).  
+
+<|ref|>text<|/ref|><|det|>[[118, 172, 882, 362]]<|/det|>
+We then measured cell proliferation after auxin addition, using Incuyte videomicroscopy. The control cells (expressing OsTIR1 but having no AID- tagged protein) grew vigorously in the presence of auxin, as expected. However, cells depleted for UHRF1 and/or DNMT1 proliferated significantly slower than the control cells (Fig. 2B). This decrease in cell proliferation was markedly more pronounced after UHRF1 depletion than after DNMT1 depletion, and the compound UHRF1/DNMT1 depletion had a slightly stronger effect than the single UHRF1 depletion (Fig. 2B). Incuyte measurements detect confluency, which depends not only on the number of cells but on their size as well, so we also performed standard cell counting; these data confirmed the slower proliferation in UHRF1- depleted compared to DNMT1- depleted HCT116 cells (Fig. 2C). A similar trend was seen in DLD1 cells, where UHRF1 depletion led to a stronger inhibition of proliferation than DNMT1 depletion (Fig. S2B).  
+
+<|ref|>text<|/ref|><|det|>[[118, 363, 882, 500]]<|/det|>
+A previous study has reported that inducible DNMT1- KO in HCT116 cells caused mitotic catastrophe and apoptosis within 4 days<sup>32</sup>, so we sought to determine whether the decrease in cell proliferation may result from cell death. For this, we measured cell viability with trypan blue staining every four days after auxin treatment, but we did not detect any significant cell viability loss (Fig. S2C). Together these results indicate: that UHRF1 and/or DNMT1 depletion occurs effectively in the AID- tagged cell lines; that this depletion leads to profound growth retardation without detectable cell death; and that UHRF1 depletion has a more severe effect than DNMT1 depletion.  
+
+<|ref|>text<|/ref|><|det|>[[118, 519, 882, 604]]<|/det|>
+Genetic rescues identify the domains of UHRF1 and DNMT1 critical for supporting growth We next investigated the mechanism underlying the growth retardation. For this, we used genetic rescue of the AID- tagged HCT116 cell lines with DNMT1 and UHRF1 variants bearing point mutations in their critical domains (Fig. 2D). All the mutant proteins were expressed at levels similar to, or slightly higher than, the corresponding endogenous protein (Fig. S2D- E).  
+
+<|ref|>text<|/ref|><|det|>[[118, 606, 881, 674]]<|/det|>
+For the UHRF1 rescue constructs, we observed that the exogenously expressed WT and TTD mutant rescued cell proliferation to a similar extent (Fig. 2E). In contrast, inactivating the UBL, PHD, SRA, or RING domain rendered UHRF1 non- functional for supporting growth (Fig. 2E).  
+
+<|ref|>text<|/ref|><|det|>[[118, 676, 881, 727]]<|/det|>
+The WT DNMT1 construct and its PBD mutant derivatives both rescued the cell proliferation (Fig. 2F). In contrast, the UIM mutant, H3K9me3 binding motif mutant, or catalytically inactive form of DNMT1 were all unable to rescue the slow growth phenotype.  
+
+<|ref|>text<|/ref|><|det|>[[118, 729, 881, 779]]<|/det|>
+To summarize, some but not all of the domains of UHRF1 and DNMT1 are required to support cell proliferation in HCT116 cells. The links between the proliferation defect and DNA methylation loss are explored in the following sections.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 797, 836, 814]]<|/det|>
+## UHRF1 depletion induces a more severe DNA methylation loss than DNMT1 depletion  
+
+<|ref|>text<|/ref|><|det|>[[118, 815, 881, 884]]<|/det|>
+We then examined the dynamics of DNA methylation loss upon removal of UHRF1 and/or DNMT1. As above, we started our experiments with the HCT116 cells and used 3 independent methods that measure DNA methylation levels: LUMA, LC- MS/MS, and shallow- coverage WGBS (Fig. 3A).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 883, 206]]<|/det|>
+LUMA showed that the parental cells (WT) displayed no change in DNA methylation over the course of a 12- day auxin treatment. In contrast, cells depleted of UHRF1 and/or DNMT1 progressively lost DNA methylation, as expected (Fig. 3A). Strikingly, UHRF1 depletion caused a markedly stronger loss than DNMT1 depletion; for instance, 6 days after treatment, the percentage of restriction- resistant sites was \(\sim 75\%\) in WT cells, \(\sim 55\%\) in DNMT1- depleted cells, and \(\sim 40\%\) in UHRF1- depleted cells (Fig. 3A). The cells depleted for both UHRF1 and DNMT1 had a slightly stronger loss than the cells lacking UHRF1 only.  
+
+<|ref|>text<|/ref|><|det|>[[118, 206, 882, 258]]<|/det|>
+LC- MS/MS and WGBS results were fully consistent with the LUMA data (Fig. 3B- C). In addition, LUMA on DLD1 degron cells showed that UHRF1 depletion caused a more severe loss of methylation than DNMT1 depletion in this cellular background as well (Fig. S3A).  
+
+<|ref|>text<|/ref|><|det|>[[118, 260, 882, 379]]<|/det|>
+Lastly, we used LUMA after auxin treatment to verify which of the rescue constructs can maintain DNA methylation levels following degradation of the endogenous UHRF1 or DNMT1 proteins (Fig. S3B- C). The only mutant form of UHRF1 supporting DNA methylation maintenance was the TTD mutant (Fig S3B), while the only mutant form of DNMT1 that retained activity towards DNA methylation was the PBD mutant (Fig. S3C). Therefore, for the 9 variants of UHRF1 and DNMT1 that we have tested, there is a one- to- one correspondence between the ability to support growth, and the ability to maintain DNA methylation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 380, 882, 484]]<|/det|>
+Together these results further suggest that loss of DNA methylation underpins the growth retardation of the various degron lines treated with auxin. In addition, they show that UHRF1 depletion causes a more severe loss of DNA methylation than DNMT1 depletion, in parallel with a more severe growth retardation. Importantly, the slower growth of UHRF1- depleted cells rules out passive dilution of DNA methylation as an explanation for the greater loss of methylation they experience, when compared to DNMT1- depleted cells.  
+
+<|ref|>text<|/ref|><|det|>[[118, 502, 882, 605]]<|/det|>
+Our previous data clearly suggested that the role of UHRF1 in DNA methylation homeostasis goes beyond its canonical function of promoting DNMT1 activity. To get deeper insight into the mechanism(s) underlying this phenomenon, we performed deep- coverage WGBS, focusing on the early time points after auxin addition (days 0, 2, 4), which showed interesting dynamics yet minimize secondary effects due to growth differences.  
+
+<|ref|>text<|/ref|><|det|>[[118, 606, 882, 761]]<|/det|>
+For our analysis, we segmented the genome into 1- kb bins. Four days after auxin addition, cells lacking DNMT1 showed \(\sim 600,000\) tiles that had lost \(25\%\) or more DNA methylation relative to day 0. However, that number was over twice as great in the UHRF1- depleted cells, which showed more than 1.3 million demethylated tiles (Fig. 4A). The joint depletion of UHRF1 and DNMT1 had an effect similar to, but slightly stronger than, UHRF1 depletion alone. A similar analysis performed only 2 days after auxin addition yielded similar results, albeit with smaller numbers of demethylated tiles (Fig. S4A). The Venn diagrams of Fig. 4B and S4B illustrate that most of the tiles demethylated after DNMT1 depletion were also demethylated after UHRF1 depletion.  
+
+<|ref|>text<|/ref|><|det|>[[118, 763, 882, 849]]<|/det|>
+We then refined this analysis by looking at distinct genomic regions (Fig. S4C). The loss of DNA methylation in UHRF1 and/or DNMT1- depleted cells is pervasive and affects promoters, gene bodies, and intergenic regions. However, we noticed that gene bodies in particular experience greater loss of DNA methylation upon UHRF1 depletion than upon DNMT1 depletion (Fig. S4C).  
+
+<|ref|>text<|/ref|><|det|>[[118, 851, 882, 902]]<|/det|>
+Gene- body methylation involves the de novo methyltransferases DNMT3A and DNMT3B33- 35, so the results prompted us to examine whether UHRF1 might have an effect on the targets of DNMT3A and DNMT3B, which are expressed in HCT116 cells.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 883, 275]]<|/det|>
+In previous studies, kinetic DNA methylation studies performed with randomized oligonucleotides have determined systematically which flanking sequences are favored by DNMT1, DNMT3A, and DNMT3B in vitro, and the in vitro preferences are reflected in the cellular DNA methylation patterns<sup>36- 40</sup>. We have exploited these data in the following manner (Fig. 4C): for each of the enzymes, we created a table in which the 256 possible NNCGNN sequences are ranked by order of preference in vitro. In parallel, we ranked the 256 possible NNCGNN sequences by average methylation level in each point of our WGBS dataset. Then we calculated pairwise Pearson r- correlation coefficients between the in vitro preferences and the actual WGBS values. This bioinformatic approach quantifies how much the flanking sequence preferences of a particular enzyme match to the actual genome- wide methylation in cells.  
+
+<|ref|>text<|/ref|><|det|>[[117, 276, 883, 467]]<|/det|>
+Fig. 4C shows the results for 4 conditions: DNMT1- AID and UHRF1- AID cells, each before and 4 days after auxin addition. The data show that, before auxin is added, there is high correlation between the in vitro DNMT1 and DNMT3A preferences, and the actual average methylation levels in NNCGNN bins observed in cells, suggesting that these two enzymes have a strong contribution in shaping the methylome of HCT116 cells under our experimental conditions, which is not the case for DNMT3B (correlation score close to zero). When DNMT1 was depleted by auxin addition, its most preferred target sites were no longer among the most methylated ones, as the correlation coefficient dropped from 0.423 to 0.183. In contrast, the sites favored by DNMT3A were less affected, as the coefficient only marginally declined from 0.443 to 0.338. Therefore, DNMT1 depletion seems to affect preferentially DNMT1 target sites, as expected, providing a validation of our analysis.  
+
+<|ref|>text<|/ref|><|det|>[[117, 468, 883, 640]]<|/det|>
+After UHRF1 depletion, the preferred DNMT1 sites lost methylation as well, which was also expected. Notably, the drop was more profound after UHRF1 depletion (from 0.436 to - 0.078) than after DNMT1 depletion. As DNMT1 is already completely depleted in the DNMT1- AID cells, this means that another activity contributing to methylation of the DNMT1 sites is also decreased in the UHRF1- AID cells. Interestingly, UHRF1 depletion also had a very strong effect on the DNMT3A sites, for which the correlation score went from 0.430 to 0.070, suggesting that the enzyme was no longer a major contributor to the DNA methylation pattern. The values for DNMT3B went from 0.040 to - 0.451, indicative of a strong anticorrelation, and meaning that the best DNMT3 sites actually fell among the least methylated sites when UHRF1 was removed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 642, 882, 710]]<|/det|>
+To summarize, this rich dataset shows that UHRF1 depletion leads to profound decreases of DNA methylation not just at the best DNMT1 target sites, but also at the best DNMT3A and DNMT3B target sites suggesting that UHRF1 also has a role in DNMT3A and DNMT3B mediated DNA methylation.  
+
+<|ref|>text<|/ref|><|det|>[[117, 711, 883, 902]]<|/det|>
+We obtained further support for this scenario by examining DNA methylation losses at H3K36me3- marked CpG islands, which are a well- described target of de novo methyltransferases in HCT116 cells<sup>34</sup>. We extracted from our WGBS data the methylation values for CpG islands and ranked them in 10 bins according to their H3K36me3 content (Fig. 4D). CpG islands with low levels of H3K36me3 lost the same amount of DNA methylation after UHRF1 depletion or after DNMT1 depletion: the methylation difference between these two conditions was close to zero. In contrast, CpG islands with higher levels of H3K36me3 lost significantly more methylation when UHRF1 was removed than when DNMT1 was removed (Fig. 4D). As a control, we carried out the same analysis with H3K79me2, another histone mark that is also found in gene bodies but is not associated with de novo DNA methyltransferases (Fig. S4D). In that case we found no correlation between H3K79me2 levels and reliance on
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 881, 137]]<|/det|>
+UHRF1. This analysis shows that regions of the genome that are especially reliant on de novo methyltransferases to gain DNA methylation are also especially reliant on UHRF1 to maintain their DNA methylation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 154, 880, 188]]<|/det|>
+## Physical, functional, and genetic interactions between UHRF1 and the de novo methyltransferases  
+
+<|ref|>text<|/ref|><|det|>[[118, 189, 881, 293]]<|/det|>
+To test a possible physical association between UHRF1 and DNMT3A or DNMT3B, we performed a series of co- immunoprecipitation (co- IP) experiments. These experiments showed that UHRF1 indeed interacts with both DNMT3A and DNMT3B (Fig. 5A); furthermore the TTD domain was sufficient for interaction (Fig. 5A). We repeated these co- IP with full- length UHRF1 in the presence of Ethidium Bromide and obtained identical results, indicating that the interactions are not bridged by chromatin (Fig. S5A- B).  
+
+<|ref|>text<|/ref|><|det|>[[118, 293, 882, 448]]<|/det|>
+Work with UHRF1 deletion mutants showed that the TTD and PHD domain were necessary for interaction with DNMT3A and DNMT3B, whereas the UBL, SRA, and RING finger were not (Fig. 5B). As the experiments pointed to an important role of the TTD, we performed a last series of co- IP experiments, with a mutant form of UHRF1 that is full- length but has two mutations (Y188A/Y191A) that inactivate the hydrophobic pocket of the TTD. The mutations significantly reduced the capacity of UHRF1 to interact with both DNMT3A and DNMT3B (Fig. 5C). To summarize, we detect a physical interaction between UHRF1 and the two de novo methyltransferases in HCT116 cells, this interaction involves the TTD, and it is not indirectly mediated by chromatin.  
+
+<|ref|>text<|/ref|><|det|>[[118, 450, 882, 606]]<|/det|>
+We identified a further mechanistic link between UHRF1 and the de novo methyltransferases by a fully independent approach. We used a recently developed proteomic approach<sup>41</sup> to characterize the "chromatome" of our cell lines at various time points after DNMT1 or UHRF1 depletion (Fig. S5C). One of the proteins that was less abundant in chromatin after UHRF1 removal than after DNMT1 removal was DNMT3B (Fig. S5D). We carried out western blotting on whole- cell lysates and found that UHRF1 depletion had no discernible effect on the amount of DNMT1 or DNMT3A, but that it led to a decrease of DNMT3B abundance, while DNMT1 depletion had no such effect (Fig. 5D). The decrease of DNMT3B on chromatin in the absence of UHRF1 is therefore mirrored in whole- cell extracts.  
+
+<|ref|>text<|/ref|><|det|>[[118, 607, 882, 745]]<|/det|>
+We then explored the genetic interactions between UHRF1, DNMT1, DNMT3A, and DNMT3B. For this, we generated CRISPR knockouts of DNMT3A and DNMT3B in the DNMT1- AID and UHRF1- AID lines (Fig. S5E), and observed their effects on DNA methylation levels. As expected, removing DNMT3A and DNMT3B from the DNMT1- AID line (D3AB DKO derivative) led to a greater loss of DNA methylation upon auxin treatment (Fig. 5E). In contrast, the D3AB DKO mutations did not make the loss of methylation more severe in the UHRF1- AID line (Fig. 5E). This important result suggests that UHRF1 does not act in parallel to, but instead upstream of, DNMT3A and DNMT3B, which is consistent with our co- IP results.  
+
+<|ref|>text<|/ref|><|det|>[[118, 746, 881, 832]]<|/det|>
+Lastly, these genetic experiments brought another crucial conclusion: the DNMT1- AID/DNMT3A KO/DNMT3B KO, which are completely devoid of DNMT activity upon auxin addition, still lose DNA methylation more slowly than the UHRF1- AID line treated with auxin (Fig 5E). Therefore, besides stimulating the activity of the DNA methyltransferases, UHRF1 must be preserving DNA methylation homeostasis by at least one other mechanism.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 850, 504, 867]]<|/det|>
+## UHRF1 opposes active demethylation by TET2  
+
+<|ref|>text<|/ref|><|det|>[[118, 868, 881, 901]]<|/det|>
+To guide the next set of experiments, we went back to our WGBS data. The sequence preferences of TET1 and TET2 have been identified in vitro<sup>42</sup>, and we asked whether the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 881, 154]]<|/det|>
+optimal target sites of these enzymes were particularly likely to lose methylation in the absence of DNMT1 or UHRF1. We used the same workflow described earlier in Fig. 4C, and calculated correlation coefficients between WGBS- derived methylation data and in vitro data for the TET enzymes (Fig. 6A).  
+
+<|ref|>text<|/ref|><|det|>[[118, 154, 881, 242]]<|/det|>
+We found that the optimal TET1 and TET2 sites became strongly hypomethylated upon DNMT1 removal (correlation coefficients of - 0.330 and - 0.451 respectively). However, the demethylation at these sites became even more marked after UHRF1 was removed (coefficients of - 0.451 and - 0.579 respectively). This result is compatible with heightened TET action upon UHRF1 removal, suggesting that UHRF1 might oppose TET activity.  
+
+<|ref|>text<|/ref|><|det|>[[118, 242, 881, 327]]<|/det|>
+We tested this possibility genetically, focusing on TET2, which is the more expressed enzyme in HCT116 cells. For, this, we generated stable shTET2 derivatives of our UHRF1- AID, DNMT1- AID, and UHRF1- AID/DNMT1- AID HCT116 lines. The knockdown efficiency was \(\sim 80\%\) at the mRNA level (Fig. 6B). We then measured DNA methylation by the LUMA assay in the various shCtrl and shTET2 lines, before and after auxin addition.  
+
+<|ref|>text<|/ref|><|det|>[[118, 328, 881, 431]]<|/det|>
+In the absence of auxin treatment, shTET2 led to a small but significant increase of DNA methylation, only in the DNMT1- AID and UHRF1- AID/DNMT1- AID lines (Fig. 5A). Upon 4 days of auxin treatment, the UHRF1- AID, DNMT1- AID, and compound UHRF1- AID/DNMT1- AID lines expressing non- targeting shRNA lost DNA methylation to various extents, with the cells lacking UHRF1 losing more DNA methylation than the cells lacking DNMT1 (Fig. 6C), which agrees with all of our previously presented data.  
+
+<|ref|>text<|/ref|><|det|>[[118, 431, 881, 622]]<|/det|>
+We then examined the effects of shTET2 combined with auxin treatment. In the DNMT1- AID line, the shTET2 did not rescue the DNA methylation loss, suggesting that active demethylation by TET2 is not the main contributor in this situation. In contrast, the shTET2 did significantly alleviate the DNA methylation loss experienced by UHRF1- AID or UHRF1- AID/DNMT1- AID cells (Fig. 6C). This key result establishes that TET2 activity contributes to DNA methylation loss when UHRF1 is absent, but not when DNMT1 is absent. Similar results were obtained after 8 days of auxin depletion (Fig. 5B). In addition, we measured cell proliferation in all the cell lines to eliminate possible confounding factors (Fig. 5C). In all cases, the shTET2 derivatives grew faster than the matched shControl line. Therefore, shTET2 does not preserve DNA methylation in UHRF1- depleted cells by preventing passive DNA methylation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 623, 881, 694]]<|/det|>
+We therefore conclude that UHRF1 protects the genome against TET2 activity, which contributes to the more severe DNA hypomethylation seen in UHRF1- depleted cells, as compared to DNMT1- depleted cells, or even cells lacking all three DNMTs (See model in Fig. 7).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 207, 100]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[118, 102, 882, 206]]<|/det|>
+Using degron tools, we have carried out a precise time- resolved analysis of DNA methylation loss upon removal of UHRF1, DNMT1, or both. Our genomics data coupled to genetic experiments show that, in addition to its well- described role as an activator of DNMT1, UHRF1 also interacts functionally with DNMT3A and DNMT3B. In addition, we show that UHRF1 opposes the DNA demethylating activity of TET2. Besides their conceptual importance, these findings may be relevant for developing novel cancer therapies.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 225, 565, 240]]<|/det|>
+## A powerful tool to study UHRF1 and DNMT1 function  
+
+<|ref|>text<|/ref|><|det|>[[118, 242, 882, 309]]<|/det|>
+We generated colorectal cancer cell lines in which the endogenous copies of UHRF1 and/or DNMT1 are tagged with fluorescent markers as well as degron tags, allowing for their rapid and controlled depletion. These cell lines constitute a valuable resource for research into the dynamics and functions of these two essential epigenetic regulators.  
+
+<|ref|>text<|/ref|><|det|>[[118, 310, 882, 414]]<|/det|>
+In the absence of auxin, the fluorescently labeled UHRF1 and DNMT1 proteins appear fully functional (Fig. 1D and Supplemental Movies). This provides an ideal system with which to study the abundance, localization and dynamics of these two key epigenetic actors. For instance, a chemical screen to identify regulators of UHRF1 protein stability has previously been carried out with an exogenous UHRF1- GFP protein \(^{43}\) , and it may be worthwhile to repeat it on the endogenously tagged protein.  
+
+<|ref|>text<|/ref|><|det|>[[118, 415, 882, 483]]<|/det|>
+In addition, we chose the mClover/mRuby fluorescent protein pair because it can be used for FRET analysis \(^{44}\) . This opens up opportunities for future work, including high- content microscopy screens \(^{45,46}\) to identify chemical compounds or genes that regulate the colocalization of endogenous UHRF1 and DNMT1.  
+
+<|ref|>text<|/ref|><|det|>[[118, 484, 882, 570]]<|/det|>
+The proteins are rapidly, fully, and synchronously degraded upon auxin addition, allowing us to examine DNA demethylation dynamics upon removal of the key regulators. This question has been addressed in the past, for instance by using shRNA \(^{23}\) or by transfecting the Cre protein into conditional KO cells \(^{47}\) . However our degron approach has unprecedented temporal resolution and population homogeneity, permitting more precise analyses.  
+
+<|ref|>text<|/ref|><|det|>[[118, 572, 882, 692]]<|/det|>
+Our system also lends itself to rescue experiments, allowing us to examine which domains of UHRF1 and DNMT1 are essential for their function. The results we obtained confirmed earlier results obtained with other systems, such as shRNA \(^{23}\) . However, the better kinetics and homogeneity of the degron system make it possible to consider more systematic screens, such as alanine scanning mutagenesis of the entire proteins in order to reveal new critical positions. This would be a useful complement to other approaches addressing the same question, such as high- density CRISPR scanning \(^{48}\) .  
+
+<|ref|>text<|/ref|><|det|>[[118, 694, 882, 744]]<|/det|>
+Lastly, the degron system is reversible upon auxin removal, and can be coupled to cell synchronization. These features will be valuable in designing future experiments addressing the roles of UHRF1 and DNMT1 in the different phases of the cell cycle.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 763, 671, 779]]<|/det|>
+## Roles of UHRF1 and DNMT1 in cancer cell proliferation or viability  
+
+<|ref|>text<|/ref|><|det|>[[118, 781, 882, 831]]<|/det|>
+The addition of auxin to our AID- tagged cells leads to a rapid and extensive decrease in UHRF1 and/or DNMT1 protein abundance. This leads to a severe impairment of cell growth both in HCT116 and DLD- 1 cells, yet the cells maintain viability.  
+
+<|ref|>text<|/ref|><|det|>[[118, 833, 882, 901]]<|/det|>
+These results are consistent with a recent report describing DNMT1- degron cells \(^{31}\) , yet they contrast with earlier publications: most notably, the inducible deletion of the DNMT1 gene in HCT116 cells has been reported to cause a G2 arrest, eventually followed by escape and mitotic catastrophe \(^{32}\) . Possible causes for this discrepancy with our observations might
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 882, 170]]<|/det|>
+include the removal of an uncharacterized important genetic element along with the targeted DNMT1 genomic sequence and/or the expression at low levels of a truncated DNMT1 protein that has negative consequences in the knockout cells. However, we cannot rule out the possibility that minute amounts of DNMT1 escaping degradation in our system are sufficient to promote survival.  
+
+<|ref|>text<|/ref|><|det|>[[118, 172, 882, 310]]<|/det|>
+Similarly, previous reports in which UHRF1 was depleted by siRNA or shRNA reached various conclusions as to the effects of the depletion<sup>23,25</sup>. Removal of the protein by a CRISPR KO has been attempted, but only yielded hypomorphs<sup>49</sup>, suggesting that the protein might be essential. In our study, we observed a strong cell proliferation defect after UHRF1 depletion compared with WT cells (Fig. 2B, C). This likely explains why UHRF1 KO have not yet been reported in cancer cells. It also suggests that caution should be exercised when carrying out and interpreting siRNA or shRNA experiments on UHRF1, as the least depleted cells will have a growth advantage over the most depleted ones.  
+
+<|ref|>text<|/ref|><|det|>[[118, 310, 882, 466]]<|/det|>
+The mechanisms underpinning the essentiality of UHRF1 and DNMT1 for long- term cancer cell proliferation have been suggested to be linked with their role in DNA methylation homeostasis<sup>23</sup>. Our rescue experiments are compatible with this hypothesis, as mutants that rescue DNA methylation also rescue growth, and vice versa. However, the number of mutants we and others have examined is still limited, and the mutations studied, such as the RING finger inactivation, may affect other important functions in addition to DNA methylation maintenance. The tools we have developed may help reveal if the functions of DNMT1 and UHRF1 in cell proliferation and DNA methylation maintenance are indeed fully linked, or whether they can be dissociated.  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 484, 754, 500]]<|/det|>
+## Functional and physical interaction between UHRF1, DNMT3A and DNMT3B  
+
+<|ref|>text<|/ref|><|det|>[[118, 502, 882, 554]]<|/det|>
+There have been some indications in the past that UHRF1 might be connected to the de novo DNA methylation machinery<sup>50- 52</sup> but our results now rigorously establish this connection, ground it in molecular detail, and determine its effects on DNA methylation genome- wide.  
+
+<|ref|>text<|/ref|><|det|>[[118, 555, 882, 710]]<|/det|>
+The physical interaction between proteins involves the TTD of UHRF1 and, more precisely still, its hydrophobic pocket. Our co- immunoprecipitations in the presence of Ethidium Bromide eliminate the possibility that the interaction is bridged by chromatin, however we cannot presently conclude whether the interaction is direct, or involves other unknown factors. We note that DNMT3A contains a histone- like TARK motif that is methylated on the lysine by G9A and GLP<sup>53</sup>. This situation is reminiscent of other proteins directly bound by the TTD, namely histone H3 and DNA Ligase 1<sup>13,14</sup>. Thus, one possibility for future exploration will be to test the possibility that UHRF1 interacts directly with the TARK motif of DNMT3A.  
+
+<|ref|>text<|/ref|><|det|>[[118, 711, 882, 780]]<|/det|>
+We find that depleting UHRF1 leads to decreased abundance of the DNMT3B protein, without affecting DNMT1 or DNMT3A (Fig. 5D). Additional experiments could be carried out in the future to identify the underlying mechanism which could be direct or indirect, for example depending on the fact that methylated nucleosomes appear to stabilize DNMT3B<sup>54</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 781, 882, 832]]<|/det|>
+Lastly, we have carried out our experiments in human cancer cells, but it will be worthwhile in the future to clarify whether UHRF1 also promotes DNMT3A/DNMT3B activity in other systems, such as mouse embryonic stem cells.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 850, 355, 866]]<|/det|>
+## UHRF1 inhibits TET2 activity  
+
+<|ref|>text<|/ref|><|det|>[[118, 868, 882, 901]]<|/det|>
+Our epistasis studies reveal that TET2 contributes to DNA demethylation more actively when UHRF1 is absent. This finding may at first sight appear discordant with a recent report, which
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 883, 170]]<|/det|>
+found that UHRF1 actually recruits the short form of TET1 to heterochromatin, where it catalyzes DNA hydroxymethylation<sup>55</sup>. However, disparities in cellular systems, coupled to dissimilarities between TET1 and TET2, could contribute to the contrast between our results. Also, we note that the recruitment of TET1 by UHRF1 appears to be limited to the late S- phase, and could be counterbalanced by other processes in other phases of the cell cycle.  
+
+<|ref|>text<|/ref|><|det|>[[118, 172, 883, 258]]<|/det|>
+At this stage, we cannot say if the decreased TET2 action is due to an inhibition at the level of transcription, translation, stability, or activity of the protein. However, an interesting parallel might possibly be drawn with results obtained in mouse ES cells, where UHRF1 has been proposed to inhibit SETDB1 activity by binding hemimethylated DNA<sup>56</sup>. A similar regulation might occur between UHRF1 and the TETs.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 277, 451, 293]]<|/det|>
+## UHRF1 as a therapeutic target in cancer  
+
+<|ref|>text<|/ref|><|det|>[[117, 294, 883, 536]]<|/det|>
+Cancer cells have an aberrant epigenome, and this creates opportunities for anti- tumoral therapies<sup>57</sup>. Among the various epigenetic marks, DNA methylation has been validated as a valuable target<sup>20</sup>. The DNMT1 inhibitor 5- aza- cytidine is successfully used in the clinic against Myelodysplasia and Acute Myeloid Leukemia but has limitations such as high toxicity, rapid degradation, and emergence of resistance<sup>58</sup>. The new generation of selective DNMT1 inhibitors that has been developed<sup>59</sup> may alleviate some of those issues, yet these molecules still trigger DNMT1 degradation<sup>60</sup>, which might have unwanted side effects. Our data point out that an altogether different strategy may be viable, by targeting UHRF1 instead of DNMT1, which justifies drug design efforts currently ongoing in the community<sup>61- 64</sup>. As with any essential protein, one of the challenges will be to identify a therapeutic dosage window and/or appropriate delivery methods such that cancer cells are harmed while healthy cells are spared. It is possible that the high expression levels of UHRF1 in tumors<sup>22,23</sup> will provide such a window. Altogether, our work reveals new, non- canonical functions of UHRF1, and open up avenues for further exploration of this key epigenetic regulator in normal cells and in disease.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 315, 100]]<|/det|>
+## Materials and Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 120, 298, 135]]<|/det|>
+## Plasmid Construction  
+
+<|ref|>text<|/ref|><|det|>[[118, 136, 882, 222]]<|/det|>
+We utilized the pX330- U6- Chimeric_BB- CBh- hSpCas9 plasmid, obtained from Feng Zhang (Addgene #42230), as the basis for constructing CRISPR/Cas vectors. The construction process followed the protocol outlined by Ran et al.65 . To generate the mAlD donor plasmids, we modified constructs of the Kanemaki lab (Addgene #72827 and #121180). In order to incorporate mRuby2, we replaced mCherry2 in the donor plasmid (Addgene #121180).  
+
+<|ref|>text<|/ref|><|det|>[[117, 224, 883, 396]]<|/det|>
+For the rescue experiments, wild- type (WT) UHRF1 and each of the point mutants (M8R/F46V, Y188A, DAE, G448D, and H741A) were cloned into pLenti6.2/V5- DEST (invitrogen). Likewise, WT DNMT1 and each of the point mutants (H170V, D381A/E382A/S392A, W464A/W465A, C1226W) were cloned into pSBbi- Bla (Addgene: #60526). To target DNMT3A and DNMT3B, we cloned the oligonucleotide sequences for gRNA into the lenticRISPR v2- Blast vector (Addgene #83480). Additionally, we cloned the shRNA targeting TET2 into the pLKO.1- blast vector (Addgene #26655). Plasmids were generated using PCR, restriction enzymes, or Gibson Assembly Cloning techniques. All plasmids underwent sequencing prior to their utilization. The oligonucleotide sequences inserted into the LenticRISPR v2- Blast vector and pLKO.1- blast vector are available in Supplementary File 1.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 415, 512, 431]]<|/det|>
+## Cell Culture, Transfection, and Colony Isolation  
+
+<|ref|>text<|/ref|><|det|>[[117, 432, 882, 586]]<|/det|>
+The HCT116 cell line, which conditionally expresses OsTIR1 under the control of a tetracycline (Tet)- inducible promoter, was obtained from the RIKEN BRC Cell Bank (http://cell.brc.riken.jp/en/) and genotyped by Eurofins. HCT116 cell lines were cultured in McCoy's 5A medium (Sigma- Aldrich), supplemented with \(10\%\) FBS (Gibco), \(2\mathrm{mM}\) L- glutamine, \(100~\mathrm{U / mL}\) penicillin, and \(100~\mu \mathrm{g / mL}\) streptomycin. The DLD1 cell line, which constitutively expresses OsTIR1 (F74A), was provided by the Kanemaki Lab. DLD1 cell lines were cultured in RPMI- 1640 medium (Sigma- Aldrich), supplemented with \(10\%\) FBS (Gibco), \(2\mathrm{mM}\) L- glutamine, \(100~\mathrm{U / mL}\) penicillin, and \(100~\mu \mathrm{g / mL}\) streptomycin. Both cell lines were maintained in a \(37^{\circ}C\) humid incubator with \(5\%\) CO2.  
+
+<|ref|>text<|/ref|><|det|>[[117, 588, 882, 673]]<|/det|>
+To establish stable cell lines, cells were seeded in a 24- well plate and transfected with CRISPR/Cas and donor plasmids using Lipofectamine 2000 (Thermo Fisher Scientific). Two days post- transfection, cells were transferred and diluted into \(10 - cm\) dishes, followed by selection in the presence of \(700~\mu \mathrm{g / ml}\) G418 or \(100~\mu \mathrm{g / ml}\) Hygromycin B. After a period of 10- 12 days, colonies were individually picked for further selection in a 96- well plate.  
+
+<|ref|>text<|/ref|><|det|>[[117, 675, 882, 778]]<|/det|>
+For the induction of AID- fused protein degradation in HCT116 cell lines, cells were seeded and incubated with \(0.2\mu \mathrm{g / mL}\) doxycycline (Dox) and \(20\mu \mathrm{M}\) auxinole for one day. Subsequently, the medium was replaced with fresh medium containing \(0.2\mu \mathrm{g / mL}\) Dox and \(500~\mu \mathrm{M}\) indole- 3- acetic acid (IAA), a natural auxin. Similarly, to induce AID- fused protein degradation in DLD1 cell lines, cells were seeded and incubated with regular medium for one day, followed by medium replacement with \(1\mu \mathrm{M}5\mathrm{- Ph - IAA}\)  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 798, 364, 813]]<|/det|>
+## Confocal microscopy analysis  
+
+<|ref|>text<|/ref|><|det|>[[118, 815, 882, 883]]<|/det|>
+Cells were fixed in \(2\%\) paraformaldehyde at room temperature for \(10\mathrm{min}\) . After fixation, cells were permeabilized with \(0.5\%\) Triton X- 100 in PBS for \(10\mathrm{min}\) at \(4^{\circ}C\) , then washed with PBS. Cells were mounted with ProLong Diamond Antifade Mountant with DAPI (P36961, Thermo Fisher Scientific). Images were obtained using a Leica DMi6000 (Leica Microsystems).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 335, 101]]<|/det|>
+## Videomicroscopy analysis  
+
+<|ref|>text<|/ref|><|det|>[[117, 102, 883, 259]]<|/det|>
+Videomicroscopy analysisFor live cell imaging, cells were grown on 35 mm FluoroDish (World Precision Instruments) with 0.17 mm thick optical quality glass bottom and fitted with a 4- well silicone insert (Ibid). Timelapse images were taken every 10 minutes for 20 hours using an inverted Eclipse Ti- E microscope (Nikon) equipped with a CSU- X1 (Yokogawa) spinning disk integrated in Metamorph software, and a 4- laser bench (Gataka systems). \(\sim 45 \mu \mathrm{m} Z\) stacks were acquired (Z- step size: \(3 \mu \mathrm{m}\) ) with a \(60 \times\) CFI Plan Apo VC oil- immersion objective (numerical aperture 1.4). The microscope has a motorized Nano z100 piezo stage (Mad City Lab), a stage top incubator (Tokai Hit) and an EMCCD camera (Evolve, Photometrics). The images were 3D deconvolved using the NIS Elements software (Nikon).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 277, 507, 293]]<|/det|>
+## Infection/transfection for rescue experiments  
+
+<|ref|>text<|/ref|><|det|>[[118, 294, 882, 379]]<|/det|>
+The generation of lentiviral or Sleeping Beauty transposon vectors followed the methodology of "Plasmid Construction." Subsequently, the cell lines were either infected or transfected with WT, UHRF1- AID, DNMT1- AID, or UHRF1/DNMT1- AID. To ensure stable expression of the target genes or shRNA, the infected or transfected cells were incubated with \(10 \mu \mathrm{g / mL}\) Blasticidin for a period of one week, allowing for the selection of stable cell populations.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 398, 300, 414]]<|/det|>
+## Western blot analysis  
+
+<|ref|>text<|/ref|><|det|>[[118, 415, 882, 536]]<|/det|>
+Cells were harvested after trypsinization, washed twice with PBS, and lysed with RIPA buffer (Sigma- Aldrich) with protease inhibitor (1 mM phenylmethanesulfonyl fluoride and \(1 \times\) Complete Protease Inhibitor Cocktail; Roche), then sonicated with a Bioreutor (Diagenode). The sonicated samples were centrifuged at 16,000 g for 15 min, then the supernatants were subjected to the Bradford Protein Assay Kit (BioRad). Equivalent amounts of protein were resolved by SDS- PAGE and then transferred to a nitrocellulose membrane. The primary antibodies used for western blot analysis are listed in Supplementary File 2.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 555, 311, 570]]<|/det|>
+## Cell proliferation assay  
+
+<|ref|>text<|/ref|><|det|>[[118, 571, 882, 675]]<|/det|>
+For cell proliferation studies, HCT116 cells were seeded at a density of 5,000 cells per well in a 96- well plate. They were then treated with \(0.2 \mu \mathrm{g / mL}\) Dox and \(20 \mu \mathrm{M}\) auxinole for one day. Following this, the medium was replaced with fresh medium containing \(0.2 \mu \mathrm{g / mL}\) Dox and \(500 \mu \mathrm{M}\) IAA. Throughout the experiment, images were captured every 2 hours using an IncuCyte ZOOM microscope (Essen Bioscience). The IncuCyte ZOOM software was utilized to determine the cell confluency (%) based on the acquired images.  
+
+<|ref|>text<|/ref|><|det|>[[118, 676, 881, 727]]<|/det|>
+To obtain cell count data and assess cell viability, trypan blue staining was performed after every 4 days of auxin treatment. The TC20 Automated Cell Counter (BioRad) was used to obtain the cell count data and calculate the cell viability rate.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 746, 336, 762]]<|/det|>
+## DNA methylation analysis  
+
+<|ref|>text<|/ref|><|det|>[[118, 763, 881, 849]]<|/det|>
+LUMA and Pyrosequencing analyses were conducted following standard procedures. Whole- genome bisulfite sequencing (WGBS) libraries were prepared using the tPBAT protocol, as described by Miura et al66,67. The library preparation involved using 100 ng of genomic DNA spiked with \(1\%\) (w/w) of unmethylated lambda DNA from Promega. Subsequently, sequencing was carried out by Macrogen Japan Inc. utilizing the HiSeq X Ten system.  
+
+<|ref|>text<|/ref|><|det|>[[118, 850, 881, 901]]<|/det|>
+To process the sequenced reads, BMap was employed to map them to the hg38 reference genome. The mapping information was then summarized using an in- house pipeline, which has been previously described67. Custom scripts for this pipeline can be
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 881, 118]]<|/det|>
+accessed via GitHub at the following link: https://github.com/FumihitoMiura/Project- 2. A summary of the mapping information can be found in Supplementary File 3.  
+
+<|ref|>text<|/ref|><|det|>[[118, 120, 881, 188]]<|/det|>
+Once the methyl reports data was obtained, methylKit was utilized to determine the methylation levels of individual CpG sites and identify differential methylated regions (DMRs). In this analysis, DMRs were defined as having a methylation difference greater than \(25\%\) and a q- value lower than 0.01.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 207, 345, 222]]<|/det|>
+## Flanking sequence analysis  
+
+<|ref|>text<|/ref|><|det|>[[118, 224, 882, 328]]<|/det|>
+Genome- wide DNA methylation profiles were used to extract methylation level of individual CpG sites and their flanking sequences as described earlier<sup>68</sup>. CpGs with sequences coverage \(> = 10\) were included in the downstream analysis. Enzymes' flanking sequence profiles were combined from published data<sup>37- 40,42</sup>. Pearson r- values were determined with Microsoft Excel. Symmetrical preference profiles for DNMT3A and DNMT3B were generated by averaging the preferences of pairs of corresponding complementary flanks<sup>36</sup>.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 347, 263, 362]]<|/det|>
+## ChIP-seq analysis  
+
+<|ref|>text<|/ref|><|det|>[[118, 363, 882, 450]]<|/det|>
+ChIP- seq data for HCT116 cells was obtained from ENCODE. Upon downloading the data, we performed quality checks on the reads using FASTQC (v0.11.9, available at https://www.bioinformatics.babraham.ac.uk/projects/fastqc). Reads with low quality and adaptor sequences were removed using Trimmomatic with default settings (version 0.38). Subsequently, the reads were aligned to the hg38 reference genome using bowtie 2 (v2.4.5).  
+
+<|ref|>text<|/ref|><|det|>[[118, 450, 882, 588]]<|/det|>
+To calculate the histone read coverage within each CGI (CpG island), we utilized the BEDtools coverage function. Initially, CGIs with less than 4 read counts in the ChIP- seq data were excluded to avoid including randomly mapped regions. The read counts were then adjusted to counts per 10 million based on the total number of mapped reads per sample. Additionally, the counts were divided by the input read count to normalize the read counts. To prevent normalized counts from becoming infinite in regions where the input sample had zero reads, an offset of 0.5 was added to all windows before scaling and input normalization. Regions where the coverage was zero in all samples were removed from the analysis.  
+
+<|ref|>text<|/ref|><|det|>[[118, 589, 881, 640]]<|/det|>
+In order to statistically analyze differences in histone modification levels, we compared the normalized read depths across CGIs using a Wilcoxon rank sum test. This test allowed us to assess the significance of differences in histone modification levels between samples.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 659, 300, 674]]<|/det|>
+## Chromatome analysis  
+
+<|ref|>text<|/ref|><|det|>[[118, 676, 576, 692]]<|/det|>
+We followed the protocol we have recently published<sup>41</sup>.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 711, 647, 727]]<|/det|>
+## Transfection and co-immunoprecipitation with GFP trap beads  
+
+<|ref|>text<|/ref|><|det|>[[118, 729, 882, 848]]<|/det|>
+In a \(10\mathrm{cm}\) dish with HCT116 cells at \(60\%\) confluency, 12 micrograms of GFP- tagged plasmid (GFP, hUHRF1, UBL, TTD, PHD, SRA, RING, \(\Delta\) UBL, \(\Delta\) TTD, \(\Delta\) PHD, \(\Delta\) SRA, \(\Delta\) RING, hUHRF1- TTD- mut) and 12 micrograms of dsRed- tagged plasmid (DNMT3A, DNMT3B) were transfected using 60 \(\mu \mathrm{L}\) Lipofectamine 2000 (Thermo Fisher Scientific). After a 3- hour incubation with Lipofectamine, the medium was replaced with McCoy's 5A medium and incubated for 1 day. The transfected cells were then collected by trypsinization, washed twice with PBS, and subjected to co- immunoprecipitation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 850, 881, 901]]<|/det|>
+Co- immunoprecipitation was carried out following the manufacturer's protocol for GFP- Trap Agarose (chromotek). The collected cells were suspended in \(200\mu \mathrm{L}\) lysis buffer (10 mM Tris/Cl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, \(0.5\%\) NP- 40, \(2.5\mathrm{mM}\) MgCl2, \(1\mathrm{mM}\) PMSF,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 883, 170]]<|/det|>
+Protease inhibitor cocktail (Roche)) and incubated on ice for 30 minutes. The lysed samples were centrifuged at 16,000 g for 10 minutes at \(4^{\circ}C\) . A portion of the supernatant was collected as input, and the remaining supernatant was combined with dilution buffer ( \(10mM\) Tris/Cl pH 7.5, \(150mM\) NaCl, \(0.5mM\) EDTA, \(1mM\) PMSF, Protease inhibitor cocktail (Roche)) to a final volume of \(500\mu L\) .  
+
+<|ref|>text<|/ref|><|det|>[[118, 172, 882, 241]]<|/det|>
+Subsequently, \(30\mu L\) of GFP- Trap Agarose, pre- equilibrated with dilution buffer, was added to each lysate sample. The samples were incubated overnight at \(4^{\circ}C\) with gentle rotation. The GFP- Trap Agarose was then washed 5 times with lysis buffer and boiled for 10 minutes with SDS- PAGE sample buffer to elute the bound proteins for further analysis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 259, 696, 276]]<|/det|>
+## RNA extraction and quantitative reverse transcription PCR (RT-qPCR)  
+
+<|ref|>text<|/ref|><|det|>[[118, 277, 883, 397]]<|/det|>
+Total RNA was extracted from cells with RNeasy Plus Mini kit (Qiagen) according to the manufacturer's instructions and quantified using Qubit RNA BR Assay kit on Qubit 2.0 Fluorometer (Thermo Fisher Scientific). For RT- qPCR, total RNA was reverse transcribed using SuperScript IV Reverse Transcriptase (Thermo Fisher Scientific) and random primers (Promega). RT- qPCR was performed using Power SYBR Green (Applied Biosystems) following to manufacture protocol with TET2 and internal control (TBP1 and PGK1) primers. RT- qPCR primer sequences are available in Supplementary File 1.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 415, 256, 430]]<|/det|>
+## Data availability  
+
+<|ref|>text<|/ref|><|det|>[[118, 432, 712, 466]]<|/det|>
+The WGBS data has been submitted to GEO under reference GSE236026 The token for reviewer access is eribksakjbpgzgt  
+
+<|ref|>text<|/ref|><|det|>[[118, 484, 883, 518]]<|/det|>
+The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE<sup>69</sup> partner repository with the dataset identifier PXD043254  
+
+<|ref|>text<|/ref|><|det|>[[118, 537, 481, 587]]<|/det|>
+The reviewer account details are:  Username: reviewer_pxd043254@ebi.ac.uk  Password: PxYbEr4G
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 213, 100]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[113, 115, 884, 884]]<|/det|>
+1. Luo, C., Hajkova, P. & Ecker, J. R. Dynamic DNA methylation: In the right place at the right time. Science 361, 1336–1340 (2018).  
+2. Greenberg, M. V. C. & Bourc’his, D. The diverse roles of DNA methylation in mammalian development and disease. Nat. Rev. Mol. Cell Biol. 20, 590–607 (2019).  
+3. Yang, J., Bashkenova, N., Zang, R., Huang, X. & Wang, J. The roles of TET family proteins in development and stem cells. Development 147, dev183129 (2020).  
+4. Gujar, H., Weisenberger, D. J. & Liang, G. The Roles of Human DNA Methyltransferases and Their Isoforms in Shaping the Epigenome. Genes (Basel) 10, (2019).  
+5. Petryk, N., Bultmann, S., Bartke, T. & Defossez, P.-A. Staying true to yourself: mechanisms of DNA methylation maintenance in mammals. Nucleic Acids Res (2020) doi:10.1093/nar/gkaa1154.  
+6. Gowher, H. & Jeltsch, A. Mammalian DNA methyltransferases: new discoveries and open questions. Biochem. Soc. Trans. 46, 1191–1202 (2018).  
+7. Kikuchi, A. et al. Structural basis for activation of DNMT1. Nat Commun 13, 7130 (2022).  
+8. Sharif, J. et al. The SRA protein Np95 mediates epigenetic inheritance by recruiting Dnmt1 to methylated DNA. Nature 450, 908–912 (2007).  
+9. Bostick, M. et al. UHRF1 plays a role in maintaining DNA methylation in mammalian cells. Science 317, 1760–1764 (2007).  
+10. Mancini, M., Magnani, E., Macchi, F. & Bonapace, I. M. The multi-functionality of UHRF1: epigenome maintenance and preservation of genome integrity. Nucleic Acids Research 49, 6053–6068 (2021).  
+11. Arita, K., Ariyoshi, M., Tochio, H., Nakamura, Y. & Shirakawa, M. Recognition of hemimethylated DNA by the SRA protein UHRF1 by a base-flipping mechanism. Nature 455, 818–821 (2008).  
+12. Arita, K. et al. Recognition of modification status on a histone H3 tail by linked histone reader modules of the epigenetic regulator UHRF1. Proc. Natl. Acad. Sci. U.S.A. 109, 12950–12955 (2012).  
+13. Ferry, L. et al. Methylation of DNA Ligase 1 by G9a/GLP Recruits UHRF1 to Replicating DNA and Regulates DNA Methylation. Mol. Cell 67, 550-565.e5 (2017).  
+14. Kori, S. et al. Structure of the UHRF1 Tandem Tudor Domain Bound to a Methylated Non-histone Protein, LIG1, Reveals Rules for Binding and Regulation. Structure (2018) doi:10.1016/j.str.2018.11.012.  
+15. DaRosa, P. A. et al. A Bifunctional Role for the UHRF1 UBL Domain in the Control of Hemi-methylated DNA-Dependent Histone Ubiquitylation. Mol. Cell 72, 753-765.e6 (2018).  
+16. Foster, B. M. et al. Critical Role of the UBL Domain in Stimulating the E3 Ubiquitin Ligase Activity of UHRF1 toward Chromatin. Mol. Cell 72, 739-752.e9 (2018).  
+17. Nishiyama, A. et al. Uhrf1-dependent H3K23 ubiquitylation couples maintenance DNA methylation and replication. Nature 502, 249–253 (2013).  
+18. Ishiyama, S. et al. Structure of the Dnmt1 Reader Module Complexed with a Unique Two-Mono-Ubiquitin Mark on Histone H3 Reveals the Basis for DNA Methylation Maintenance. Mol. Cell 68, 350-360.e7 (2017).  
+19. Nishiyama, A. et al. Two distinct modes of DNMT1 recruitment ensure stable maintenance DNA methylation. Nat Commun 11, 1222 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 85, 881, 888]]<|/det|>
+20. Nishiyama, A. & Nakanishi, M. Navigating the DNA methylation landscape of cancer. Trends Genet 37, 1012–1027 (2021).21. Ming, X. et al. Kinetics and mechanisms of mitotic inheritance of DNA methylation and their roles in aging-associated methylome deterioration. Cell Res. (2020) doi:10.1038/s41422-020-0359-9.22. Mudbhary, R. et al. UHRF1 overexpression drives DNA hypomethylation and hepatocellular carcinoma. Cancer Cell 25, 196–209 (2014).23. Kong, X. et al. Defining UHRF1 Domains that Support Maintenance of Human Colon Cancer DNA Methylation and Oncogenic Properties. Cancer Cell 35, 633-648.e7 (2019).24. Vaughan, R. M. et al. The histone and non-histone methyllysine reader activities of the UHRF1 tandem Tudor domain are dispensable for the propagation of aberrant DNA methylation patterning in cancer cells. Epigenetics Chromatin 13, 44 (2020).25. Cai, Y. et al. Critical threshold levels of DNA methyltransferase 1 are required to maintain DNA methylation across the genome in human cancer cells. Genome Res. 27, 533–544 (2017).26. Kanemaki, M. T. Ligand-induced degrons for studying nuclear functions. Curr Opin Cell Biol 74, 29–36 (2022).27. Barretina, J. et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature 483, 603–607 (2012).28. Natsume, T., Kiyomitsu, T., Saga, Y. & Kanemaki, M. T. Rapid Protein Depletion in Human Cells by Auxin-Inducible Degron Tagging with Short Homology Donors. Cell Rep 15, 210–218 (2016).29. Yesbolatova, A., Natsume, T., Hayashi, K.-I. & Kanemaki, M. T. Generation of conditional auxin-inducible degron (AID) cells and tight control of degron-fused proteins using the degradation inhibitor auxinole. Methods 164–165, 73–80 (2019).30. Yesbolatova, A. et al. The auxin-inducible degron 2 technology provides sharp degradation control in yeast, mammalian cells, and mice. Nat Commun 11, 5701 (2020).31. Scelfo, A. et al. Tunable DNMT1 degradation reveals cooperation of DNMT1 and DNMT3B in regulating DNA methylation dynamics and genome organization. 2023.05.04.539406 Preprint at https://doi.org/10.1101/2023.05.04.539406 (2023).32. Chen, T. et al. Complete inactivation of DNMT1 leads to mitotic catastrophe in human cancer cells. Nat Genet 39, 391–396 (2007).33. Baubec, T. et al. Genomic profiling of DNA methyltransferases reveals a role for DNMT3B in genic methylation. Nature 520, 243–247 (2015).34. Masalmeh, R. H. A. et al. De novo DNA methyltransferase activity in colorectal cancer is directed towards H3K36me3 marked CpG islands. Nat Commun 12, 694 (2021).35. Janssen, S. M. & Lorincz, M. C. Interplay between chromatin marks in development and disease. Nat Rev Genet 23, 137–153 (2022).36. Emperle, M. et al. Mutations of R882 change flanking sequence preferences of the DNA methyltransferase DNMT3A and cellular methylation patterns. Nucleic Acids Res 47, 11355–11367 (2019).37. Adam, S. et al. DNA sequence-dependent activity and base flipping mechanisms of DNMT1 regulate genome-wide DNA methylation. Nat Commun 11, 3723 (2020).38. Gao, L. et al. Comprehensive structure-function characterization of DNMT3B and DNMT3A reveals distinctive de novo DNA methylation mechanisms. Nat Commun 11, 3355 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 84, 880, 888]]<|/det|>
+39. Dukatz, M. et al. Complex DNA sequence readout mechanisms of the DNMT3B DNA methyltransferase. Nucleic Acids Res 48, 11495-11509 (2020).
+40. Dukatz, M. et al. DNA methyltransferase DNMT3A forms interaction networks with the CpG site and flanking sequence elements for efficient methylation. J Biol Chem 298, 102462 (2022).
+41. Ugur, E. et al. Comprehensive chromatin proteomics resolves functional phases of pluripotency and identifies changes in regulatory components. Nucleic Acids Res gkad058 (2023) doi:10.1093/nar/gkad058.
+42. Adam, S. et al. Flanking sequences influence the activity of TET1 and TET2 methylcytosine dioxygenases and affect genomic 5hmC patterns. Commun Biol 5, 92 (2022).
+43. Ding, G. et al. Regulation of Ubiquitin-like with Plant Homeodomain and RING Finger Domain 1 (UHRF1) Protein Stability by Heat Shock Protein 90 Chaperone Machinery\*. Journal of Biological Chemistry 291, 20125-20135 (2016).
+44. Lam, A. J. et al. Improving FRET dynamic range with bright green and red fluorescent proteins. Nat Methods 9, 1005-1012 (2012).
+45. de Almeida, M. et al. AKIRIN2 controls the nuclear import of proteasomes in vertebrates. Nature 599, 491-496 (2021).
+46. Schraivogel, D. et al. High-speed fluorescence image-enabled cell sorting. Science 375, 315-320 (2022).
+47. Ginno, P. A. et al. A genome-scale map of DNA methylation turnover identifies site-specific dependencies of DNMT and TET activity. Nature Communications 11, 2680 (2020).
+48. Ngan, K. C.-H. et al. Activity-based CRISPR scanning uncovers allostery in DNA methylation maintenance machinery. Elife 12, e80640 (2023).
+49. Tian, Y. et al. UHRF1 contributes to DNA damage repair as a lesion recognition factor and nuclease scaffold. Cell Rep 10, 1957-1966 (2015).
+50. Meilinger, D. et al. Np95 interacts with de novo DNA methyltransferases, Dnmt3a and Dnmt3b, and mediates epigenetic silencing of the viral CMV promoter in embryonic stem cells. EMBO Rep 10, 1259-1264 (2009).
+51. Zhang, J. et al. S phase-dependent interaction with DNMT1 dictates the role of UHRF1 but not UHRF2 in DNA methylation maintenance. Cell Res. 21, 1723-1739 (2011).
+52. Maenohara, S. et al. Role of UHRF1 in de novo DNA methylation in oocytes and maintenance methylation in preimplantation embryos. PLoS Genet. 13, e1007042 (2017).
+53. Chang, Y. et al. MPP8 mediates the interactions between DNA methyltransferase Dnmt3a and H3K9 methyltransferase GLP/G9a. Nat Commun 2, 533 (2011).
+54. Sharma, S., Carvalho, D. D. D., Jeong, S., Jones, P. A. & Liang, G. Nucleosomes Containing Methylated DNA Stabilize DNA Methyltransferases 3A/3B and Ensure Faithful Epigenetic Inheritance. PLOS Genetics 7, e1001286 (2011).
+55. Arroyo, M. et al. Isoform-specific and ubiquitination dependent recruitment of Tet1 to replicating heterochromatin modulates methylcytosine oxidation. Nat Commun 13, 5173 (2022).
+56. Sharif, J. et al. Activation of Endogenous Retroviruses in Dnmt1(-/-) ESCs Involves Disruption of SETDB1-Mediated Repression by NP95 Binding to Hemimethylated DNA. Cell Stem Cell 19, 81-94 (2016).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 84, 880, 626]]<|/det|>
+57. Dawson, M. A. The cancer epigenome: Concepts, challenges, and therapeutic opportunities. Science 355, 1147–1152 (2017).58. Zeidan, A. M., Kharfan-Dabaja, M. A. & Komrokji, R. S. Beyond hypomethylating agents failure in patients with myelodysplastic syndromes. Curr Opin Hematol 21, 123–130 (2014).59. Pappalardi, M. B. et al. Discovery of a first-in-class reversible DNMT1-selective inhibitor with improved tolerability and efficacy in acute myeloid leukemia. Nat Cancer 2, 1002–1017 (2021).60. Chen, Q. et al. GSK-3484862 targets DNMT1 for degradation in cells. NAR Cancer 5, zcad022 (2023).61. Senisterra, G. et al. Discovery of Small-Molecule Antagonists of the H3K9me3 Binding to UHRF1 Tandem Tudor Domain. SLAS Discov 23, 930–940 (2018).62. Chang, L. et al. Discovery of small molecules targeting the tandem tudor domain of the epigenetic factor UHRF1 using fragment-based ligand discovery. Sci Rep 11, 1121 (2021).63. Kori, S. et al. Structure-based screening combined with computational and biochemical analyses identified the inhibitor targeting the binding of DNA Ligase 1 to UHRF1. Bioorg Med Chem 52, 116500 (2021).64. Liu, W. H. et al. Discovery and Mechanism of Small Molecule Inhibitors Selective for the Chromatin-Binding Domains of Oncogenic UHRF1. Biochemistry 61, 354–366 (2022).65. Ran, F. A. et al. Genome engineering using the CRISPR-Cas9 system. Nat Protoc 8, 2281–2308 (2013).66. Miura, F., Enomoto, Y., Dairiki, R. & Ito, T. Amplification-free whole-genome bisulfite sequencing by post-bisulfite adaptor tagging. Nucleic Acids Res 40, e136 (2012).67. Miura, F. et al. Highly efficient single-stranded DNA ligation technique improves low-input whole-genome bisulfite sequencing by post-bisulfite adaptor tagging. Nucleic Acids Res 47, e85 (2019).68. Adam, S., Klingel, V., Radde, N. E., Bashtrykov, P. & Jeltsch, A. On the accuracy of the epigenetic copy machine: comprehensive specificity analysis of the DNMT1 DNA methyltransferase. Nucleic Acids Res gkad465 (2023) doi:10.1093/nar/gkad465.69. Deutsch, E. W. et al. The ProteomeXchange consortium at 10 years: 2023 update. Nucleic Acids Res 51, D1539–D1548 (2023).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 275, 100]]<|/det|>
+## Acknowledgments  
+
+<|ref|>text<|/ref|><|det|>[[117, 102, 883, 362]]<|/det|>
+Work in the lab of Pierre- Antoine Defossez is supported by the Fondation ARC (PGA1 RF20180206807), Agence Nationale de la Recherche (ANR- 15- CE12- 0012- 01), Institut National du Cancer (INCa PLBio 2015- 1- PLBio- 01- DR A- 1), and Fondation pour la Recherche Medicale. This study was also supported by the LabEx "Who Am I?" #ANR- 11- LABX- 0071 and the Universite de Paris IdEx #ANR- 18- IDEX- 0001 funded by the French Government through its "Investments for the Future" program. KY was supported by ARC foundation CDD Postdoctorant en France (Ref. No. PDF20181208337), Labex "Who Am I?" postdoctoral support, and a JSPS Postdoctoral Fellowship for Research Abroad (Ref. No. 202260432). The work of FM and TI was supported by Platform Project for Supporting Drug Discovery and Life Science Research (Basis for Supporting Innovative Drug Discovery and Life Science Research (BINDS)) from AMED under Grant Number JP20am0101103 (support number 2652). AJ was supported by Deutsche Forschungsgemeinschaft (DFG) JE252/48. DF has received support for this project by ARC labellisation program 2019. The authors greatly acknowledge the Cell and Tissue Imaging (PICT- IBisA), Institut Curie, member of the French National Research Infrastructure France- Biolmaging (ANR10- INBS- 04).  
+
+<|ref|>text<|/ref|><|det|>[[118, 379, 881, 431]]<|/det|>
+We are very grateful to Allison Bardin and Raphael Margueron for useful advice. We thank the following colleagues for the gift of useful reagents: Alexis Gautreau, Olivier Bernard, Michaela Fontenay.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 450, 298, 466]]<|/det|>
+## Author Contributions  
+
+<|ref|>text<|/ref|><|det|>[[115, 467, 884, 693]]<|/det|>
+KY, XC, BR and PAD designed and performed experiments, KY, XC, BR, AS and ML established cell lines and rescue constructs for this study. KY, XC, BR, LF, and ML performed biological experiments (WB, PCR, and cell culture). KY, BR, LF, FB performed DNA methylation analysis (LUMA and LC- MS). CSL performed microscopy analysis. FM prepared WGBS libraries, then sequenced them with NGS. PB and AJ conducted the analyses shown in Fig. 4C and 6A. KY, PB, FM, OK performed bioinformatic analyses. EU performed proteome analysis. KY, PBA, HL, MTK, DF, AJ, TI, and PAD acquired funding. KY and PAD supervised the project. KY wrote the original draft of the manuscript. AJ, DF, BR and PAD reviewed and edited the draft.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 712, 383, 727]]<|/det|>
+## Competing Interests Statement  
+
+<|ref|>text<|/ref|><|det|>[[118, 729, 555, 745]]<|/det|>
+The authors declare no competing financial interests.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 764, 375, 779]]<|/det|>
+## Materials and correspondence  
+
+<|ref|>text<|/ref|><|det|>[[118, 782, 390, 849]]<|/det|>
+Requests should be addressed to yamako0801@icloud.com and pierre- antoine.defossez@cnrs.fr
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[39, 17, 930, 550]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[35, 560, 950, 802]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[45, 857, 909, 874]]<|/det|>
+<center>Figure 1. Establishment and validation of endogenous AID-tagged UHRF1 and/or DNMT1 HCT116 cells. </center>  
+
+<|ref|>text<|/ref|><|det|>[[45, 873, 951, 963]]<|/det|>
+(A) Schematic of the CRISPR/Cas9 genome editing strategy to endogenously tag UHRF1 with mAID/mClover and DNMT1 with mAID/mRuby2. (B) Order of events for the generation of the different cell lines. (C) Immunoblot images for validation of endogenous AID-tagged UHRF1 and/or DNMT1 HCT116 cells. (D) Representative fluorescence images on UHRF1-AID/DNMT1-AID HCT116 cells showing that tagged UHRF1 and DNMT1 co-localize. (E) Quantification of the DNA methylation level in each HCT116 cell line with LUMA, LC-MS/MS, or WGBS. Tukey HSD test: \(*p < 0.05\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[42, 24, 950, 750]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[33, 858, 962, 888]]<|/det|>
+<center>Figure S1. Establishment and validation of endogenous AID-tagged UHRF1 and DNMT1 DLD1 cells and further validations. </center>  
+
+<|ref|>text<|/ref|><|det|>[[33, 888, 962, 963]]<|/det|>
+(A) Procedure followed to establish the UHRF1-AID and DNMT1-AID cells in the DLD1 background. (B) Immunoblot for validation of endogenous AID-tagged UHRF1 or DNMT1 DLD1 cells. (C) Cell proliferation data on HCT116 derivatives without auxin (Incucyte videomicroscopy). (D) Cell proliferation data on HCT116 derivatives without auxin (Cell counting). (E) Cell proliferation data on DLD1 derivatives without auxin (Cell counting). (F) Quantification of the DNA methylation level in DLD1 derivatives, without auxin (LUMA). Tukey HSD test: \(^{**}p < 0.01\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[30, 13, 480, 390]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[30, 408, 475, 490]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[30, 512, 475, 644]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[30, 696, 475, 828]]<|/det|>
+ 
+
+<|ref|>image_caption<|/ref|><|det|>[[45, 863, 951, 985]]<|/det|>
+<center>Figure 2. The depletion of UHRF1 and DNMT1 is efficient, negatively affects growth, and can be rescued genetically. (A) Immunoblot of HCT116 cells following treatment with Auxin (IAA) at the indicated time points (hours) and before treatment (NT). (B) Cell proliferation of the HCT116 derivatives in the continuous presence of auxin for the indicated durations (Incucyte videomicroscopy). Error bars represent the SEM of biological triplicates. (C) Cell proliferation of the HCT116 derivatives in the continuous presence of auxin for the indicated durations (cell counting). The error bars represent the SEM of biological triplicates. (D) Schematic of the rescue experiments. (E) UHRF1 domain map showing the mutants studied (left panel) and corresponding cell proliferation analysis (Cell count, right panel). Error bars represent the SEM of biological triplicates. (F) Same as panel E, but for DNMT1.</center> 
+
+<|ref|>image<|/ref|><|det|>[[500, 10, 978, 330]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[500, 333, 978, 490]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[500, 504, 978, 660]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[500, 682, 978, 857]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[30, 24, 965, 700]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[45, 833, 950, 864]]<|/det|>
+<center>Figure S2. Growth of DLD1 derivatives after UHRF1 or DNMT1 degradation; additional controls on the HCT116 derivatives and the rescue experiment. </center>  
+
+<|ref|>text<|/ref|><|det|>[[45, 864, 950, 972]]<|/det|>
+(A) Immunoblot of DLD1 cells following treatment with Auxin (5-ph-IAA) at the indicated time points (hours) and before treatment (NT). (B) Cell proliferation of DLD1 derivatives after auxin addition (Cell counting). Error bars represent the SEM of biological triplicates. (C) Cell viability of HCT116 derivatives following UHRF1 or DNMT1 degradation (trypan blue staining). Error bars represent the SEM of biological triplicates. (D) Immunoblot images for validation of exogenous UHRF1 rescue constructs. The pink arrow indicates endogenous UHRF1 tagged with AID and mClover. The purple arrow indicates exogenous UHRF1 tagged with V5. (E) Immunoblot images for validation of exogenous DNMT1 rescue constructs. Legend as in panel D.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[45, 24, 940, 650]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[45, 911, 950, 972]]<|/det|>
+<center>Figure 3. UHRF1-depleted cells show more severe DNA hypomethylation than DNMT1-depleted cells. (A) Global DNA methylation analysis in the indicated HCT116 derivatives after auxin treatment for the indicated duration (LUMA). Error bars represent the SEM of biological triplicates. (B) As in panel A, but the quantitation of 5-mC was done by LC-MS/MS. (C) As in panel A, but the quantitation of DNA methylation was done by WGBS. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[240, 15, 721, 277]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[66, 300, 888, 864]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[50, 872, 925, 965]]<|/det|>
+<center>Figure S3. Validation of the effects of UHRF1 and DNMT1 degradation on DNA methylation in DLD1 cells; identification of the domains essential for DNA methylation. (A) Global DNA methylation analysis in the indicated DLD1 derivatives after auxin treatment for the indicated duration (LUMA). Error bars represent the SEM of biological triplicates. (B) Global DNA methylation analysis in the indicated HCT116 UHRF1-AID rescue lines. Error bars represent the SEM of 3 independent experiments. (C) As in panel B, but for DNMT1-AID. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[30, 12, 960, 300]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[50, 305, 500, 485]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[520, 305, 960, 476]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[45, 508, 900, 666]]<|/det|>  
+
+<|ref|>table<|/ref|><|det|>[[150, 664, 844, 790]]<|/det|>
+
+<table><tr><td rowspan="3"></td><td colspan="5">WGBS data in HCT116:</td></tr><tr><td colspan="3">DNMT1-AID</td><td colspan="2">UHRF1-AID</td></tr><tr><td>No auxin</td><td>4 days auxin</td><td>No auxin</td><td>4 days auxin</td><td></td></tr><tr><td rowspan="2">Correlation with enzymatic data for:</td><td>DNMT1</td><td>0.423</td><td>0.183</td><td>0.436</td><td>-0.078</td></tr><tr><td>DNMT3A</td><td>0.443</td><td>0.338</td><td>0.430</td><td>0.070</td></tr><tr><td></td><td>DNMT3B</td><td>-0.005</td><td>-0.315</td><td>0.040</td><td>-0.451</td></tr></table>  
+
+<|ref|>image<|/ref|><|det|>[[856, 667, 904, 792]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[46, 811, 949, 844]]<|/det|>
+Figure 4. Greater loss of DNA methylation upon UHRF1 depletion than upon DNMT1 depletion; UHRF1 regulates DNA methylation at DNMT1, DNMT3A and DNMT3B sites.  
+
+<|ref|>text<|/ref|><|det|>[[46, 843, 950, 963]]<|/det|>
+(A) Volcano plot of differentially methylated regions (DMRs, 1kb bins) after 4 days of depletion of UHRF1 and/or DNMT1. Blue dots: hypomethylated regions (>25% loss of methylation, q-value < 0.01), red dots: hypermethylated regions (>25% gain of methylation, q-value < 0.01). Gray dots: no significant change. (B) Venn diagram of the hypomethylated regions in the indicated cell lines, 4 days after depletion of the proteins. (C) Workflow used to quantitatively compare WGBS methylation values to the in vitro preferences of DNMT1, DNMT3A and DNMT3B. (D) Higher levels of H3K36me3 correlate with larger losses of DNA methylation in CpG islands. The CGIs were ranked by H3K36me3 level in HCT116 cells and split into 10 equally sized bins. Lines = median; box = 25th–75th percentile; whiskers = 1.5 × interquartile range from box.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[33, 15, 940, 202]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[50, 252, 940, 500]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[35, 516, 962, 799]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[45, 812, 950, 842]]<|/det|>
+<center>Figure S4. Greater loss of DNA methylation upon UHRF1 depletion than upon DNMT1 depletion; additional data. </center>  
+
+<|ref|>text<|/ref|><|det|>[[45, 842, 951, 958]]<|/det|>
+(A) Volcano plot of differentially methylated regions (DMRs, 1kb bins) after 2 days of depletion of UHRF1 and/or DNMT1. Blue dots: hypomethylated regions ( \(>25\%\) loss of methylation, q-value \(< 0.01\) ), red dots: hypermethylated regions ( \(>25\%\) gain of methylation, q-value \(< 0.01\) ). Gray dots: no significant change. (B) Venn diagram of the hypomethylated regions in the indicated cell lines, 2 days after depletion of the proteins. (C) Boxplots of CpG methylation (\%) in the indicated regions and conditions. Promoters: from -200 to +1000 bps from TSS; Gene bodies obtained from hg38 refFlat by removal of the promoter regions; Intergenic regions: the whole genome minus promoters and gene bodies. (D) Lack of correlation between H3K79me2 levels and DNA methylation loss at CpG islands; legend as in Panel 4D.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[26, 20, 960, 450]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[28, 458, 960, 800]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[45, 860, 770, 876]]<|/det|>
+<center>Figure 5. Physical and functional interaction between UHRF1, DNMT3A, and DNMT3B. </center>  
+
+<|ref|>text<|/ref|><|det|>[[45, 876, 951, 964]]<|/det|>
+(A) Western blotting after the indicated co-immunoprecipitation experiments. hUHRF1: Full-length protein. The other names indicate isolated domains, as depicted in Figure 2E. (B) Same as in A, except we used truncated constructs in which the indicated domains were deleted from the full-length protein. (C) Same as in A, except we used a full-length UHRF1 protein with a point mutation in the Tandem Tudor Domain (Y188A). (D) Western blotting showing abundance of the indicated proteins in total cell extracts. (E) Quantitation of the loss of DNA methylation in the indicated cell lines after 8 days of protein depletion, by LC-MS/MS. Tukey HSD test: N.S. \(p > 0.05\) , \(^{**}p < 0.01\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[28, 20, 70, 39]]<|/det|>
+
+<|ref|>image<|/ref|><|det|>[[512, 20, 545, 39]]<|/det|>
+
+<|ref|>table<|/ref|><|det|>[[171, 72, 479, 150]]<|/det|>
+
+<table><tr><td colspan="3">GFP</td><td colspan="3">GFP-UHRF1</td></tr><tr><td>Input</td><td>IP</td><td>Input</td><td>IP</td><td>Input</td><td>IP</td></tr><tr><td>-</td><td>+</td><td>-</td><td>+</td><td>-</td><td>+</td></tr></table>
+
+<|ref|>image<|/ref|><|det|>[[28, 123, 479, 280]]<|/det|>
+
+<|ref|>image<|/ref|><|det|>[[512, 123, 965, 280]]<|/det|>
+
+<|ref|>image<|/ref|><|det|>[[28, 317, 69, 336]]<|/det|>
+
+<|ref|>image<|/ref|><|det|>[[172, 315, 360, 336]]<|/det|>
+
+<|ref|>image<|/ref|><|det|>[[428, 317, 468, 336]]<|/det|>
+
+<|ref|>image<|/ref|><|det|>[[512, 315, 551, 336]]<|/det|>
+
+<|ref|>image<|/ref|><|det|>[[66, 355, 409, 375]]<|/det|>
+
+<|ref|>image<|/ref|><|det|>[[57, 378, 409, 440]]<|/det|>
+
+<|ref|>text<|/ref|><|det|>[[70, 466, 405, 494]]<|/det|>
+"Chromatome" identification by proteomics at Days 0, 2, and 4, of auxin treatment
+
+<|ref|>image<|/ref|><|det|>[[424, 355, 777, 490]]<|/det|>
+
+<|ref|>image<|/ref|><|det|>[[388, 530, 674, 789]]<|/det|>
+
+<|ref|>image_caption<|/ref|><|det|>[[54, 876, 960, 963]]<|/det|>
+<center>Figure S5. Additional controls for the UHRF1/DNMT3A/DNMT3B interaction; chromatome experiments reveal the effect of UHRF1 depletion on DNMT3B abundance; validation of the DNMT3A/DNMT3B KOs (A-B) Western blotting after the indicated co-immunoprecipitation experiments, without or with Ethidium Bromide (20 ug/mL). (C) Flowchart of the chromatome experiments. (D) Illustration of the chromatome results for DNMT3B, which is less abundant upon UHRF1 depletion. (E) Validation by western blotting of the DNMT3A and DNMT3B CRISPR KOs in UHRF1-AID or DNMT1-AID HCT116 cells.</center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[64, 20, 863, 120]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[46, 20, 163, 35]]<|/det|>
+<center>A Flowchart </center>  
+
+<|ref|>table<|/ref|><|det|>[[155, 180, 850, 285]]<|/det|>
+
+<table><tr><td rowspan="3"></td><td rowspan="3"></td><td colspan="4">WGBS data in:</td></tr><tr><td colspan="2">DNMT1-AID</td><td colspan="2">UHRF1-AID</td></tr><tr><td>No auxin</td><td>4 days auxin</td><td>No auxin</td><td>4 days auxin</td></tr><tr><td rowspan="2">Correlation with in vitro</td><td>TET1</td><td>-0.006</td><td>-0.330</td><td>0.037</td><td>-0.498</td></tr><tr><td>TET2</td><td>-0.118</td><td>-0.451</td><td>-0.070</td><td>-0.579</td></tr></table>  
+
+<|ref|>image<|/ref|><|det|>[[42, 310, 950, 644]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[46, 312, 456, 344]]<|/det|>
+<center>B RT-qPCR (HCT116) compensated with each shCtrl samples </center>  
+<|ref|>image_caption<|/ref|><|det|>[[46, 872, 583, 888]]<|/det|>
+<center>Figure 6. UHRF1 protects against active demethylation by TET2. </center>  
+
+<|ref|>text<|/ref|><|det|>[[46, 888, 950, 964]]<|/det|>
+(A) Workflow used to quantitatively compare WGBS methylation values to the in vitro preferences of TET1 and TET2. (B) RT-qPCR analysis for validation of TET2 knockdown for HCT116 UHRF1 and/or DNMT1-AID cell lines. (C) Global DNA methylation analysis (LUMA) for HCT116 UHRF1 and/or DNMT1-AID cell lines combined with TET2 knockdown. Error bars represent the SEM of 3 independent experiments. Student t-test: N.S. \(p > 0.05\) , \(***p < 0.001\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[35, 20, 960, 680]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[45, 887, 840, 903]]<|/det|>
+<center>Figure S6. UHRF1 protects against active demethylation by TET2: additional data and controls. </center>  
+
+<|ref|>text<|/ref|><|det|>[[45, 903, 950, 963]]<|/det|>
+(A) Global DNA methylation analysis (LUMA) for HCT116 UHRF1 and/or DNMT1-AID cell lines combined with TET2 knockdown, in the absence of auxin. Error bars represent the SEM of 3 independent experiments. Student t-test: N.S. \(p > 0.05\) , \(*p < 0.05\) , \(**p < 0.01\) . (B) As in Panel A, but following 8 days of auxin treatment. (B) Growth curves of the indicated cell lines, in the presence of auxin (Cell counts).
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[138, 32, 366, 46]]<|/det|>
+# UHRF1 canonical function:  
+
+<|ref|>text<|/ref|><|det|>[[46, 47, 457, 63]]<|/det|>
+DNA methylation maintenance by DNMT1 activation  
+
+<|ref|>title<|/ref|><|det|>[[592, 33, 870, 47]]<|/det|>
+# UHRF1 non-canonical functions:  
+
+<|ref|>text<|/ref|><|det|>[[515, 48, 952, 80]]<|/det|>
+- De novo methylation through DNMT3A and DNMT3B- Inhibition of TET activity  
+
+<|ref|>image<|/ref|><|det|>[[30, 120, 960, 370]]<|/det|>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 670, 285]]<|/det|>
+SupplementaryFile1Oligonucleotidesequences.xlsx  SupplementaryFile2Antibodies.xlsx  SupplementaryFile3WGBSbasicmetrics.xlsx  SupplementaryMovie1UHRF1mClover.mp4  SupplementaryMovie2DNMT1mRuby.mp4  SupplementaryMovie3UHRF1mCloverGreenDNMT1mRubyRed.mp4
+
+<--- Page Split --->

preprint/preprint__7e1dbd5e4b7563929632d79cc02b1d840b5d2ecb348dcd394110c3e708654a44/preprint__7e1dbd5e4b7563929632d79cc02b1d840b5d2ecb348dcd394110c3e708654a44.mmd

@@ -0,0 +1,397 @@
+
+# Projected increase of Arctic coastal erosion and its sensitivity to warming in the 21st Century  
+
+David Nielsen ( \(\boxed{ \begin{array}{r l} \end{array} }\) david.nielsen@uni- hamburg.de) University of Hamburg https://orcid.org/0000- 0003- 4201- 0373  
+
+Patrick Pieper University of Hamburg  
+
+Armineh Barkhordarian University of Hamburg  
+
+Paul Overduin Alfred Wegener Institute for Polar and Marine Research  
+
+Tatiana Ilyina Max Planck Institute for Meteorology https://orcid.org/0000- 0002- 3475- 4842  
+
+Victor Brovkin Max Plank Institute for Meteorology https://orcid.org/0000- 0001- 6420- 3198  
+
+Johanna Baehr University of Hamburg  
+
+Mikhail Dobrynin University of Hamburg  
+
+## Article  
+
+Keywords: climate change, erosion coastal erosion, permafrost  
+
+Posted Date: June 25th, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 634673/v1  
+
+License: © \(\circledast\) 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 Climate Change on February 14th, 2022. See the published version at https://doi.org/10.1038/s41558- 022- 01281- 0.
+
+<--- Page Split --->
+
+# Projected increase of Arctic coastal erosion and its sensitivity to warming in the \(21^{\mathrm{st}}\) Century  
+
+David Marcolino Nielsen \(^{1,2,*}\) , Patrick Pieper \(^{1}\) , Arminen Barkhordarian \(^{1}\) , Paul Overduin \(^{3}\) , Tatiana Ilyina \(^{4}\) , Victor Brovkin \(^{1,4}\) , Johanna Baehr \(^{1}\) , and Mikhail Dobrynin \(^{5,1}\)  
+
+\(^{1}\) Center for Earth System Research and Sustainability (CEN), Universität Hamburg, Hamburg, Germany \(^{2}\) International Max Planck Research School on Earth System Modelling, Max Planck Institute for Meteorology, Hamburg, Germany \(^{3}\) Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research, Potsdam, Germany \(^{4}\) Max Planck Institute for Meteorology, Hamburg, Germany \(^{5}\) Deutscher Wetterdienst, Hamburg, Germany \(^{*}\) Correspondence to: david.nielsen@uni- hamburg.de  
+
+June 16, 2021  
+
+## 1 Abstract  
+
+2 Arctic coastal erosion damages infrastructure, threatens coastal communities, and releases organic carbon from permafrost. However, the magnitude, timing and sensitivity of coastal erosion increase to global warming remain unknown. Here, we project the Arctic- mean erosion rate to roughly double by 2100 and very likely exceed its historical range of variability by mid- \(21^{\mathrm{st}}\) century. The sensitivity of erosion to warming also doubles, reaching \(0.4 - 0.5 \mathrm{m}\) year \(^{- 1} \circ \mathrm{C}^{- 1}\) and \(2.3 - 2.8 \mathrm{TgC}\) year \(^{- 1} \circ \mathrm{C}^{- 1}\) by the end of the century under moderate and high- emission scenarios. Our first \(21^{\mathrm{st}}\) - century pan- Arctic coastal erosion rate projections should inform policy makers on coastal conservation and socioeconomic planning. Our organic carbon flux projections also lay out the path for future work to investigate the impact of Arctic coastal erosion on the changing Arctic Ocean, on its role as a global carbon sink, and on the permafrost- carbon feedback.  
+
+## 12 Main  
+
+13 Arctic coast erosion is caused by a combination of thermal and mechanical drivers. Permafrost 14 thaw and ground- ice melt lead to soil decohesion and slumping, while surface ocean waves me- 15 chanically abrade the Arctic coast [1]. Sea- ice loss expands the fetch for waves [2, 3], and prolongs 16 the open- water season, increasing the vulnerability of the Arctic coast to erosion [4, 5]. In the past 17 decades, coastal retreat rates have increased throughout the Arctic, often by a factor of two or more 18 [6- 10]. The historical acceleration of erosion in the Arctic is linked with the observed decreasing
+
+<--- Page Split --->
+
+sea- ice cover [2, 4, 11], increasing air surface [12, 13] and permafrost temperatures [14]. As for the future, Arctic surface air temperature is projected to exceed its natural range of variability within the next decades [15]. Arctic sea ice decline has already exceeded natural variability [15], and summer ice- free conditions are projected by mid- \(21^{\mathrm{st}}\) century [16]. New regimes of surface waves are also projected in the Arctic Ocean and along the coast [17- 19]. Consequently, Arctic coastal erosion rates are expected to increase in the coming decades. However, the extent of this increase is still unknown, as no projections of Arctic coastal erosion rates are available. To fill this gap, we present the first \(21^{\mathrm{st}}\) - century projections of coastal erosion at the pan- Arctic scale.  
+
+The thawing of permafrost globally releases organic carbon (OC) and increases atmospheric and oceanic greenhouse gas concentrations, feeding back to further warming [20- 23]. Arctic coastal erosion alone releases about as much OC as all the Arctic rivers combined [23, 24], fueling about one- fifth of Arctic marine primary production [25]. Despite consistent improvements in the representation of permafrost dynamics [26, 27], the current generation of Earth system models (ESMs) does not account for abrupt permafrost thaw, which may cause projections of OC losses to be largely underestimated [28, 29]. Arctic coastal erosion is one form of abrupt permafrost thaw [22] and a relevant component of the Arctic carbon cycle [23, 30]. Nonetheless, it has not been considered in climate projections so far. The scale mismatch between Arctic coastal erosion and modern ESMs requires the development of holistic models, that account for the key large- scale processes to bridge this gap [30- 32].  
+
+In this study, we present a novel approach to represent Arctic coastal erosion at the scales of modern ESMs. We develop a semi- empirical Arctic coastal erosion model combining observations from the Arctic Coastal Dynamics (ACD) database [33], climate reanalyses, ESM and ocean surface wave simulations. Our model considers the main thermal and mechanical drivers of erosion as dynamical variables, represented by yearly- accumulated positive temperatures and significant wave heights, and constant ground- ice content from observations. Our approach allows us to make \(21^{\mathrm{st}}\) - century projections of coastal erosion at the pan- Arctic scale. We quantify the magnitude, timing and sensitivity of Arctic coastal erosion and its associated OC loss in the context of climate change.  
+
+## Emergence of Arctic coastal erosion  
+
+We project the Arctic- mean coastal erosion rate to increase from \(0.9 \pm 0.4 \mathrm{m / year}\) during the historical period (1850- 1950) to between \(2.0 \pm 0.7\) and \(2.6 \pm 0.8 \mathrm{m / year}\) by the end of the \(21^{\mathrm{st}}\) Century (2081- 2100), in the context of anthropogenic climate change, according to the socio- economic pathway (SSP) scenarios SSP2- 4.5 and SSP5- 8.5, respectively (Fig. 1a). This translates to an increase of the Arctic- mean coastal erosion rate by a factor of about between 2.2 and 2.9 by the end of the century with respect to the historical period. The SSP2- 4.5 and SSP5- 8.5 scenarios describe medium and high radiative forcings due to greenhouse gas emissions [34], respectively, and include the pathway of the current cumulative \(\mathrm{CO}_{2}\) emissions [35]. In both scenarios, our projections show that the Arctic- mean erosion exceeds its historical range of variability before the end of the century (Fig. 1b).
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1: Arctic coastal erosion projections. a) Time evolution of the Arctic-mean coastal erosion rate, expressed as the combined effect of its thermal and mechanical drivers. b) Yearly probabilities that the Arctic-mean coastal erosion rate leaves the historical range of variability, calculated from distributions of ensemble spread and erosion model uncertainties (see Methods). In both scenarios, it is very likely ( \(>90\%\) probability) that the Arctic-mean erosion emerges from its historical range by mid \(21^{\mathrm{st}}\) century, although the exact time of emergence is sensitive to our erosion model uncertainties. The thermal (c) and mechanical (d) drivers of erosion, expressed as yearly-accumulated daily positive degrees and significant wave heights, respectively. The erosion time series depict long-term means and therefore show little interannual variability in comparison to its drivers. </center>  
+
+We find it likely ( \(\geq 66\%\) probability) that the Arctic- mean erosion exceeds its historical range by around 2023, and very likely ( \(\geq 90\%\) probability) by 2049 (Fig. 1b), considering the largest distributions of uncertainties in our projections (i.e. ensemble spread and erosion model uncertainties). The emergence of the Arctic- mean erosion rate would very likely have happened by around 2010, if we take only the ensemble spread to define the historical range. Significant differences in projections between the two scenarios are only noticeable in the second half of the century, after a complete emergence from the historical range. Our erosion time- of- emergence estimates reflect those of its drivers, which take place around mid- \(21^{\mathrm{st}}\) Century (Fig. 1c,d), in accordance with previous studies [15, 16].  
+
+Arctic coastal erosion is typically caused by a combination of thermo- denudation (TD) and thermo- abrasion (TA) [1], which act together to thaw permafrost, melt ground ice, abrade and transport coastal material off shore. We take yearly- accumulated daily positive temperatures and significant wave heights to represent TD and TA: hereafter, the thermal and mechanical drivers of erosion, respectively. As various landform types compose the Arctic coast, the relative contribution of the thermal and mechanical drivers differs at the local scale. Erosion is predominantly thermally driven at retrogressive thaw slumps, observed at the Bykovsky Peninsula, Laptev Sea [36], and in
+
+<--- Page Split --->
+
+the Mackenzie Delta region – Beaufort Sea [37, 38], for example (Fig. 2a), as the sediment transport from ocean waves play a secondary role in coastal retreat in such formations. Erosion is also predominantly thermally driven in enclosed bays and in coastal segments protected by spits and barrier islands, where the fetch for ocean waves is limited [39], although barrier island themselves are often susceptible to wave abrasion [40]. In contrast, erosion of ice-rich cliffs, which occur extensively along the Beaufort and Laptev Sea coast for example [6–8], requires the mechanical action from ocean waves to open notches at the land-sea interface, causing the subsequent failure of often still frozen large blocks of permafrost. In some locations, the relative contribution of the thermal and mechanical drivers is more balanced than described above. At Muostakh Island in the Laptev Sea, for example, thermo-denudation and abrasion are estimated to contribute similarly to maintain erosion rates of up to \(25\mathrm{m / year}\) [8]. In our erosion model, we initially assume equal contributions from the thermal and mechanical drivers at the pan-Arctic scale during the observational period. This assumes that deviations occur comparably in both directions. We also make extreme \(10 - 90\%\) and \(90 - 10\%\) scenarios of relative thermal-mechanical contributions to test the sensitivity of our results to that assumption (see Methods and Table S1). Attributing \(90\%\) of mechanical contribution yields about \(15 - 20\%\) larger Arctic-mean coastal erosion projections by 2100 (and vice- versa), because the Arctic-mean wave exposure increases more than the thawing temperature exposure along the coast, with respect to their historical values (Fig. S1a).  
+
+## Spatial variability of erosion  
+
+The thermal and mechanical drivers of erosion explain about \(36 - 47\%\) of its observed spatial variability in multiple linear regression models. On one hand, wave exposure, combined with ground- ice content, best explains the spatial variability of erosion in most of the coastal segments \((r = 0.69 \pm 0.12\) , mean \(\pm 2\sigma\) , Fig. 2b), where erosion is not extremely high \((\sim 90^{\mathrm{th}}\) percentile, \(< 2.5\mathrm{m / year}\) ). The local wave exposure information indeed integrates several important sources of erosion variability. Not only does wave exposure promote cliff abrasion and subsequent sediment transport, but it is also proportional to open- water season (OWS) duration, which has been suggested to be the first- order driver of coastal erosion rate variability [2, 32]. In addition, sea- ice melt, and thus increasing OWS duration, responds to increasing surface air temperature, which also drives permafrost thaw and thus erosion by thermo- denudation. On the other hand, spatial differences among segments of extremely high long- term erosion rates are best characterized by thawing temperature exposure combined with ground- ice content \((r = 0.61 \pm 0.42\) , Fig. 2c). This suggests that thermo- denudation plays a more important role in driving coastal erosion rates at extreme- erosion segments, than at non- extreme ones. Among both extreme and non- extreme erosion segments, ground ice adds explanatory power, as it increases the susceptibility of permafrost to thaw and hence erosion. Our results are in accordance with previous work, which reported weak spatial correlations between ground- ice content and erosion rates [33]. Strong temporal correlations between erosion and thawing temperature exposure have also been reported for Muostakh Island – Laptev Sea [8], where erosion rates are often in the range between 10 and \(20\mathrm{m / year}\) [8, 41]. We further combine the temporal evolution of the Arctic- mean erosion with its spatial
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2: Observed and modelled erosion rate spatial variability. a) Observed long-term coastal erosion mean rates from the ACD database [33] used in this study (see Methods). Modelled against observed erosion rates in (b) non-extreme and (c) extreme erosion segments. Observed values are denoted by colored circles on the maps and on the scatter plots. Uncertainties represent \(2\sigma\) confidence intervals from the distribution of regression coefficients. Modelled historical-mean (1850-1900) (d) and end-of-the-century (2081-2100) erosion rates according to the SSP2-4.5 (e) and SSP5-8.5 (f) scenarios. The histograms in g display the historical and projected erosion time-means from the maps in d, e and f. Distributions shift and spread over time. </center>  
+
+distribution to make projections of erosion rates at the coastal segment resolution (Fig. 2d- f).  
+
+The geographical distribution of low and high- erosion segments does not change substantially from observations over time in our projections, which is partially a consequence of our model design, as explained by the three following reasons. First, we assume that the spatial model coefficients, empirically determined, remain unchanged throughout our simulations. Second, ground- ice content, an explanatory variable in our regression model, is also assumed constant over time. Third, our regression model accounts for only a fraction of the spatial variability in erosion, and may thus underestimate larger spatial changes to occur over time. Moreover, and independent from model design, local anomalies of the dynamical variables (i.e. local wave and thawing temperature exposure) are smaller in magnitude than their Arctic- mean increase. Therefore, our modelled changes in the spatial variability of erosion are small in comparison to its Arctic- mean increase. Nonetheless, our modelled spatial spread of erosion increases with time (Fig. 2g). The \(5^{\text{th}}\) - \(95^{\text{th}}\) percentile range of erosion rate distributions increases from 3.6 (0- 3.6) m/year in the historical period to 3.9 (0.9- 4.8) and 4.2 (1.4- 5.7) m/year in the SSP2- 4.5 and SSP5- 8.5 scenarios, respectively. Temporally resolved erosion rate observations are rare, often sparse in time, and only available at a relatively small number of locations [10]. Only with such observations, temporally resolved and
+
+<--- Page Split --->
+
+at the pan-Arctic scale, would empirical models be able to better constrain the temporal evolution of spatial variability of coastal erosion.  
+
+## Spatial variability of organic carbon losses  
+
+The pan- Arctic OC loss from coastal erosion increases from 6.9 (1.5- 12.3) TgC year \(^{- 1}\) during the historical period to between 13.1 (6.4- 19.7) TgC year \(^{- 1}\) and 17.2 (9.0- 25.4) TgC year \(^{- 1}\) by the end of the century in the SSP2.4- 5 and SSP5- 8.5 scenarios, respectively (Fig. 3). For the present- day climate (i.e. the period for which erosion observations are available), we estimate a pan- Arctic OC loss from coastal erosion of 8.5 (3.3- 13.7) TgC year \(^{- 1}\) . Both our simulated present- climate mean and uncertainty range are comparable with previous estimates from observations [24, 33]. Our projections suggest a pan- Arctic OC flux increase by a factor of between 1.5 and 2.0 with respect to the present- day climate, or by a factor of between 1.9 and 2.5 by 2100 with respect to the historical period.  
+
+![](images/Figure_3.jpg)
+
+<center>Figure 3: Projected organic carbon loss. Changes in organic carbon released annually by coastal erosion according to observations-based estimates and in our model simulations for the historical period (1850-1950), current climate (according to observations from the ACD [33]) and at the end of the \(21^{\mathrm{st}}\) century (2081-2100) in the two future scenarios. The height of bars represent the total uncertainty of our projections, which we disentangle between ensemble spread, spatial and temporal erosion model components. Most of the uncertainties originate from the empirical estimates of the erosion model parameters (76-97%) and the smallest fraction to the ensemble spread (3-24%). </center>  
+
+The Laptev and East Siberian Seas (LESS, Fig.2a) together account for about three quarters of the pan- Arctic OC losses in our simulations, in accordance with observations- based estimates [24]. This also holds truth for future scenarios. The reason for the relatively high OC fluxes from the LESS coast is twofold. First, the region comprises coastal segments of extremely rapid erosion, often between 10 and \(20\mathrm{m / year}\) [8, 41]. Second, the LESS coast is dominated by Yedoma ice- complex deposits, where ground- ice concentration reaches more than \(80\%\) of soil volume [8, 42], and organic- carbon content is extremely high, reaching about \(5\%\) of weight [33]. From the LESS, we simulate a present- climate OC flux of 6.5 (2.4- 10.6) TgC year \(^{- 1}\) , comparable to the 2.9- 11.0 TgC year \(^{- 1}\) range estimated by Wegner et al. (2015) [24], and comprising the ACD
+
+<--- Page Split --->
+
+value of \(7.7 \mathrm{TgC}\) year \(^{- 1}\) . In an extensive campaign over the LESS continental shelf, Vonk et al. (2012)[23] determined that about \(20 \mathrm{TgC}\) year \(^{- 1}\) are buried in the LESS sediment, which would originate from a combination of coastal and seafloor erosion. Accounting for degradation before burial and assuming an equal contribution from coastal and subsea erosion, about 11 (7- 15) \(\mathrm{TgC}\) year \(^{- 1}\) would be released by coastal erosion alone. The LESS estimate of Vonk et al. (2012) [23] is \(43 - 57\%\) larger than other observations- based estimates [24] and about \(69\%\) larger than our present- climate modelled value. These differences are likely due to extensive and high- resolution sampling, allowing for more accurate upscaling [23]. However, the uncertainties associated with the contribution between coastal and subsea erosion comprehend our modelled range (their Table S6 [23]). Therefore, an underestimation from our side is not conclusive. From the LESS coast, we project an increase in OC fluxes from 5.3 (1.0- 9.6) \(\mathrm{TgC}\) year \(^{- 1}\) in the historical period to 9.6 (5.7- 13.4) \(\mathrm{TgC}\) year \(^{- 1}\) in the SSP2- 4.5 and 12.4 (7.8- 17.1) \(\mathrm{TgC}\) year \(^{- 1}\) in the SSP5- 8.5 scenarios by 2100, which translates to an increase by a factor of between 1.8 and 2.3.  
+
+The Beaufort Sea coast accounts for about half of the remaining fraction of pan- Arctic OC flux, releasing 0.9 (0.4- 1.4) \(\mathrm{TgC}\) year \(^{- 1}\) during the present climate in our simulations, in agreement with the 0.7 \(\mathrm{TgC}\) year \(^{- 1}\) estimates from the ACD [33], however larger than previous estimates of 0.2- 0.4 \(\mathrm{TgC}\) year \(^{- 1}\) [24] (Fig. 3). Hotspots of extreme erosion are also observed in the Beaufort Sea coast. Extensive field work has been recently carried out, especially in the Yukon coast region, showing increasing erosion rates and suggesting that the associated OC fluxes could have been previously underestimated [9, 22, 43- 45]. We project an OC flux increase from the Beaufort Sea coast from 0.7 (0.2- 1.2) \(\mathrm{TgC}\) year \(^{- 1}\) in the historical period to between 1.6 (0.9- 2.3) \(\mathrm{TgC}\) year \(^{- 1}\) and 2.3 (1.4- 3.1) \(\mathrm{TgC}\) year \(^{- 1}\) by 2100 in the SSP2- 4.5 and SSP5- 8.5 scenarios, respectively, translating to an increase by a factor of between 2.3 and 3.3. The remaining marginal Arctic Seas contribute with yearly OC fluxes at absolute amounts similar to those from Beaufort Sea in our projections, accounting for about \(12 - 14\%\) of the pan- Arctic totals.  
+
+Coastal erosion is estimated to sustain about one fifth of the total Arctic marine primary production at present- climate conditions [25]. Therefore, the projected additional OC loss could have a substantial impact on the Arctic marine biogeochemistry. However, the fate of the organic carbon released by Arctic coastal erosion is currently under active debate. Field work has shown that between about \(13\%\) and \(65\%\) of the OC released into the ocean by coastal erosion could settle in the marine sediment [44- 46], slowing down remineralization. In the sediment, organic matter degradation would then take place at millennial time scales [47]. However, in the shallow nearshore zone, resuspension driven by waves and storm activity increases the residence time of OC in the water column, and allows for more effective remineralization [48]. Moreover, partial degradation of the eroded material takes place before it enters the ocean, releasing greenhouse gases directly to the atmosphere [22, 23, 49]. The OC degradation time scale thus also depends on its transit time onshore [49]. It is therefore challenging to determine short- term impacts from the projected additional OC fluxes from coastal erosion, as large uncertainties still remain regarding pathways of OC degradation.  
+
+We partition the uncertainty sources in our projections between three sources: ensemble spread,
+
+<--- Page Split --->
+
+temporal, and spatial erosion model components (see Methods). Our erosion model contributes the most to the uncertainties in our simulations: from about \(76\%\) of the total uncertainty range in the historical period and up to \(97\%\) by the end of the century in SSP5- 8.5. The ensemble spread is responsible for the remaining \(24\%\) of the total uncertainty during the historical period, and for only \(3\%\) to \(6\%\) of the total range at the end of the future scenarios. The spatial component of the erosion model accounts for about half of the total range of uncertainties, on average, without significant changes in proportion over time. The fraction of uncertainties stemming from the temporal model component increases from about \(33\%\) of the total range in the historical period to about \(55\%\) by the end of the century in SSP5- 8.5 due to the increasing magnitude of the erosion drivers. The distribution of sources of uncertainties in our projections is qualitatively similar between the pan- Arctic and the regional totals.  
+
+## Sensitivity of erosion and carbon losses to climate change  
+
+The sensitivity of Arctic coastal erosion to climate change increases over time in our simulations, and is tightly related with the Arctic amplification (AA) [12] after its onset. Arctic coastal erosion increases more rapidly in response to increasing global mean surface air temperature (SAT) in the future scenarios than it does in the historical period. Before the mid 1970s, neither global nor Arctic- mean SAT decadal trends are consistently significantly positive yet (Fig. 4a). During this period, the correlation between the Arctic- mean erosion rate and the Arctic- mean SAT is weak ( \(r = 0.26 \pm 0.29\) , mean \(\pm 2\sigma\) range, Fig. 4b). However, after the 1970s, correlations between erosion and Arctic SAT increase substantially (SSP2- 4.5: \(r = 0.68 \pm 0.18\) , SSP5- 8.5: \(r = 0.93 \pm 0.06\) , 2081- 2100 means), driven by the concurrent increasing trends. This turning point is also marked by the AA onset, when the Arctic SAT starts increasing at a faster pace than the global SAT, i.e. the AA factor is consistently larger than 1 (Fig. 4c). Therefore, the sensitivity of erosion to global SAT reflects the sensitivity of Arctic SAT to global SAT – quantified as the AA factor – after the AA onset, given the strong correspondence between erosion and the Arctic SAT at that time (Fig. 4d). The sharp increase of erosion sensitivity and the AA factor to their maximum values in the early 2000s is a signature from the so- called "hiatus" in global warming [50]. Global mean SAT stalls between the late 1990s and the early 2010s, while the erosion drivers continue to increase (Fig. S1b,c). Sensitivity values level off in the second half of the \(21^{\text{st}}\) Century, when global mean SAT trends decelerate. End- of- century sensitivities are lowest in the SSP2- 4.5 scenario, when Arctic SAT trends decrease sharply to reach the also consistently decreasing global SAT trends, and the AA factor approaches one. In order to avoid the effect of the warming hiatus, we quantify erosion sensitivity considering the historical period until before the AA onset, and during the last 50 years in the scenario simulations.  
+
+The sensitivity of the Arctic- mean erosion rate to global mean SAT increases significantly from \(0.18 \pm 0.31 \text{ m year}^{- 1} \text{°C}^{- 1}\) on average during the historical period until 1975, to at least double (between \(0.40 \pm 0.16\) and \(0.48 \pm 0.21 \text{ m year}^{- 1} \text{°C}^{- 1}\) ) during the second half of the \(21^{\text{st}}\) Century following the SSP2- 4.5 and SSP5- 8.5 scenarios, respectively. This translates to an increase in the sensitivity of OC losses to climate warming from \(1.4 \text{ TgC year}^{- 1} \text{°C}^{- 1}\) in the historical period
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4: Sensitivity to climate change. a: 20-year running trends of global and Arctic mean surface air temperature (SAT). b: Correlations between Arctic-mean erosion rates and Arctic mean SAT. c: The Arctic Amplification (AA) factor, expressed as regression coefficients of Arctic SAT changes on global SAT. The AA onset is defined when the AA factor is larger than 1. d: Sensitivity of Arctic-mean erosion rates to climate, expressed regression coefficients on global SAT. Running-window lengths are 20 years in all plots. Different window lengths show qualitatively similar results (not shown). The AA onset (dashed blue line) takes place in 1976, when the Arctic SAT increases at a faster pace than the global mean SAT, i.e. the AA factor is larger then 1. After the the 1970s, the AA factor is consistently significantly larger than 1, except for late 21st-century in the SSP2-4.5 scenario, when global and Arctic mean SATs deaccelerated and 20-year trends are momentarily similar. </center>
+
+<--- Page Split --->
+
+before until 1975, on average, to between 2.3 and \(2.8\mathrm{TgC}\) year \(^{- 1}\circ \mathrm{C}^{- 1}\) following the SSP2- 4.5 and SSP5- 8.5 scenarios, respectively.  
+
+The sensitivity parameters are useful tools to assess the state of Arctic coastal erosion increase and the associated OC fluxes at intermediate states or policy- based targets of global warming. It must be noted, however, that the sensitivity parameters usually assume linear relationships between the forcing and outcome variables [51]. Similarly, in our erosion model, we assume that the linear combination of thermal and mechanical drivers of erosion provides us with first- order large- scale information on the time evolution of Arctic coastal erosion, associated with a range of uncertainties and scenarios of proportionality factors. Non- linear effects could emerge, for example, from earlier onsets of the storm season overlapping with longer- lasting positive temperatures into fall. We do not consider sea- level change in our projections. Adding sea- level change as a temporal driver of erosion would increase future erosion and the sensitivity parameters, if it increases proportionally faster than our thermal and mechanical drivers with respect to the historical period. We do not directly consider episodic water level changes due to storms, which are relevant for coastal abrasion and sediment transport. However, by using a global dynamical wave model, and integrating yearly wave exposure at the coastal- segment level, we do incorporate the effect of storms in our mechanical driver of erosion. Our erosion model, relatively simple in comparison with higher- resolution and process- based strategies [52- 57], does not intended to represent all processes, often of fine spatial scale (order of meters or less), associated with the erosion of the Arctic coast. Here, we empirically parameterize the role of the the main, first- order drivers of Arctic coastal erosion at larger- scales, compatible with the resolution and mechanisms represented in ESMs (order of tens or hundreds of kilometers). Future work on coastal erosion modelling is necessary to constrain our relatively large uncertainties. Nonetheless, our semi- empirical approach allows us to make Pan- Arctic projections of coastal erosion, its associated OC fluxes, and thus estimate the magnitude, timing and sensitivity of their increase to global warming.  
+
+## Conclusions  
+
+We present a semi- empirical model for coastal erosion to make \(21^{\mathrm{st}}\) - century pan- Arctic projections of erosion rates and associated organic carbon (OC) losses. Our model accounts for temporal and spatial variability of erosion, combining wave and thawing temperature exposure with ground- ice content as explanatory variables. With our approach, we are able to provide estimates of magnitude, timing and sensitivity of Arctic coastal erosion increase to climate change. The Arctic- mean erosion rate increases by a factor of between 2.2 and 2.9 from the historical period (1850- 1900) to the end of the \(21^{\mathrm{st}}\) Century following the SSP2- 4.5 and the SSP5- 8.5 scenarios, respectively. The associated pan- Arctic OC flux increases by a factor of 1.9- 2.5 at the same time, reaching up to 17.2 (9.0- 25.4, two standard- deviation range) \(\mathrm{TgC}\) year \(^{- 1}\) in the SSP5- 8.5 scenario. Our projections show that Arctic coastal erosion is very likely (at least \(90\%\) probability) to exceed its historical range of variability before end of the century, even in the intermediate- emission scenario. We estimate that the sensitivity of Arctic coastal erosion to climate also increases with time, following the Arctic amplification after its onset in the 1970s, due to the strong relationship between
+
+<--- Page Split --->
+
+erosion and Arctic SAT at that time. During the second half of the century, one degree of global warming is associated with an increase of the Arctic- mean erosion by about 0.4- 0.5 m/year and 2.3- 2.8 TgC/year of associated OC carbon loss, equivalent to about 5- 8% of the present- climate OC yearly flux from the Arctic rivers into the Ocean. Arctic coastal erosion will increase more rapidly in the future in response climate change, roughly doubling in rates by 2100, and likely reaching values unseen before in the past century. Our projections allow future work to investigate the impact of Arctic coastal erosion on the permafrost- climate feedback, and the future evolution of the Arctic Ocean's ecosystems and its role as a global carbon sink. Moreover, our results should also inform policy makers on coastal conservation and socioeconomic planning at the pan- Arctic level, focusing on the sustainable future of Arctic coastal communities.  
+
+## Acknowledgements  
+
+D.M.N., M.D., P.O., T.I. and V.B. are funded by European Union's Horizon 2020 research and innovation programme under grant agreement number 773421 - project "Nunataryuk". D.M.N., P.P., A.B. T.I., J.B. and V.B. are funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy - EXC 2037 'CLICCS - Climate, Climatic Change, and Society' - Project Number: 390683824, contribution to the Center for Earth System Research and Sustainability (CEN) of Universität Hamburg.  
+
+## Competing Interests  
+
+The authors declare no competing interests.  
+
+## Authors' contributions  
+
+D.M.N, M.D., J.B. and V.B. conceived and designed the study. D.M.N., P.P., M.D., J.B. and V.B. designed the erosion model. D.M.N. and M.D. performed the Ocean wave simulations. All authors analyzed and discussed the results. All authors wrote and reviewed the paper.  
+
+## Methods  
+
+## Data  
+
+Arctic coastal observations.  
+
+We use the Arctic Coastal Dynamics (ACD) database [33] as our observational reference. The ACD compiles several sources of data and provides a list of variables for a total of 1314 coastal segments along the Arctic coast, including: long- term erosion mean rates, organic carbon concentration, soil bulk density, ground- ice fraction, mean elevation and length. From the 1314 segments, we take those classified as erosive and non- lithified, which excludes segments from the rocky coasts in Greenland and in the Canadian Archipelago and other segments that present stable or
+
+<--- Page Split --->
+
+aggrading dynamics. We also select segments containing excess ice, which excludes all the nonerosive segments from Svalbard, for example. We this work with a subset of 306 coastal segments in our analysis.  
+
+## Reanalysis  
+
+We take 2- meter air temperature and significant wave heights from ERA20C reanalysis [58] as empirical variables in our coastal erosion model. Data are taken in the same periods for which the erosion rates are provided in the ACD. The temperature and wave data have \(\sim 1.12^{\circ}\) (atmosphere) and \(1.5^{\circ}\) (waves) horizontal resolution. We assign the closest land grid cell in ERA20C from its atmospheric grid to ACD segments, and two rows of adjacent cells from the ocean grid.  
+
+## Climate projections  
+
+To force our coastal erosion model, we use a 10- member ensemble of simulations from the Max Planck Institute Earth System Model (MPI- ESM) version 1.2 in its low- resolution configuration [59] performed for the Coupled Model Intercomparison Project phase 6 (CMIP6) [34]. In this configuration, the atmospheric component ECHAM6.3 has horizontal resolution of T63 ( \(\sim 200\) km), and 47 vertical levels. The oceanic component MPIOM1.6 uses the curvilinear grid GR1.5, which has mean horizontal resolution of \(\sim 150\) km and 40 vertical levels. We use the historical simulations (1850- 2014) and two future Shared Socioeconomic Pathway (SSP) scenarios for the \(21^{\mathrm{st}}\) century projections (2015- 2100), namely: the SSP2- 4.5 and the SSP5- 8.5, which represent a mid- range and a high- end emission scenario, respectively. This range of scenarios is realistic it terms of current cumulative \(\mathrm{CO}_2\) emissions [35].  
+
+## Ocean wave simulations  
+
+We use the wave model WAM [60] to generate a 10- member ensemble of global waves for historical, SSP2- 4.5 and SSP5- 8.5 scenarios, forced by the MPI- ESM ensemble. In our setup, WAM has \(1^{\circ}\) grid resolution and is forced with daily sea- ice concentration (threshold of \(15\%\) to define open- water), 6- hourly 10- meter winds, and a realistic ETOPO2- based bathymetry as boundary conditions.  
+
+## Semi-empirical Arctic coastal erosion model  
+
+We present a simplified model for Arctic coastal erosion, compatible with the scales of Earth system models. Our model considers the dominant physical thermal and mechanical drivers of erosion, also referred to as thermal- abrasion (TA) and thermal- denudation (TD) [1]. The model is constrained to only simulate erosion at the presence of ground ice and at the absence of coastal sea ice. We use an empirical approach to quantify the relationship between the physical drivers, constraints and the erosion rates, by comparing the observations from the ACD with ERA20C reanalysis. The empirically estimated parameters are then applied to all coastal segments, which provides us with erosion rates in the pan- Arctic scale. Our model has yearly time resolution, and the spatial resolution follows the definitions of the ACD coastal segments.
+
+<--- Page Split --->
+
+The total erosion \(E(\mathrm{t},\mathrm{x})\) [m year \(^{- 1}\) ], defined in every year \(t\) and coastal segment \(x(lat,lon)\) , is given as a combination of a temporal and a spatial component.  
+
+\[E(x,t) = \overline{{E}} (t) + \Delta E(x,t) \quad (1)\]  
+
+The temporal component represents the temporal evolution of the Arctic- mean erosion \(\overline{{E}} (t)\) [m year \(^{- 1}\) ]. The spatial component \(\Delta E(x,t)\) [m year \(^{- 1}\) ] represents local departures from the Arctic mean at every year and coastal segment, providing spatially distributed values of erosion. Hereafter, we use "Arctic mean", denoted by the overline, to refer to means along the Arctic coast. All data associated with ACD coastal segments are weighted by segment lengths in the computation of means.  
+
+## The temporal component  
+
+The temporal component of our model is a linear combination of Arctic means of the thermal and mechanical drivers of erosion.  
+
+\[\overline{{E}} (t) = a_{T D}\overline{{T}} (t) + a_{T A}\overline{{H}} (t) \quad (2)\]  
+
+The thermal driver of erosion is represented by Arctic- mean yearly- accumulated daily- mean positive 2- meter air temperatures \(\overline{{T}} (\mathrm{t})\) [°C day year \(^{- 1}\) ], also commonly known as positive degrees days or thawing- degree days. The mechanical driver of erosion is represented by Arctic- mean yearly- accumulated daily significant wave heights \(\overline{{H}} (\mathrm{t})\) [m day year \(^{- 1}\) ].  
+
+We empirically estimate the linear coefficients \(a_{T A}\) [m m \(^{- 1}\) day \(^{- 1}\) year] and \(a_{T D}\) [°C m \(^{- 1}\) day \(^{- 1}\) year] by scaling the Arctic- mean physical drivers, from ERA20C reanalysis, with the observed coastal erosion rates from the ACD. This is done for the reference time \(t_{obs}\) , during which observations are available.  
+
+\[\begin{array}{l}{a_{T A} = q\frac{\overline{{E}}_{obs}}{\overline{{H}} (t_{obs})}}\\ {a_{T D} = (1 - q)\frac{\overline{{E}}_{obs}}{\overline{{T}} (t_{obs})}} \end{array} \quad (3)\]  
+
+We assume that the thermal and mechanical drivers \(a_{T D}\overline{{T}} (t)\) and \(a_{T A}\overline{{H}} (t)\) contribute in equal proportions to the Arctic- mean erosion during the reference time. We do that by setting the proportionality factor \(q\) to 0.5. We test the sensitivity of our results to this assumption by making scenarios with \(q = 0.1\) and \(q = 0.9\) (see Table S1 and Fig. S1a in the supplementary material).  
+
+## The spatial component  
+
+The spatial component of our erosion model calculates local erosion anomalies with respect to the Arctic- mean temporal evolution, and consists of two multiple linear regression (MLR) models.
+
+<--- Page Split --->
+
+We split the coastal segments in two groups by classifying them between extreme and non- extreme with respect to erosion, using \(2.5\mathrm{m / year}\) as a threshold ( \(\sim 90th\) percentile). We do not find a distinct separation between extreme and non- extreme segments in terms of geographical location (Fig. 2a), neither in terms of coastal morphology. Both groups show similar distributions of ground- ice content, mean cliff height, bathymetric profile, bulk density, as well as mean thermal and mechanical forcings derived from thawing temperature and ocean waves, for example (not shown). We test a comprehensive number of combinations of dynamical and geomorphological parameters as explanatory variables in MLR models, simultaneously maximizing goodness- of- fit and penalizing model complexity (Table S3). We fit MLR models using the usual Ordinary Least Square (OLS) method. The goodness- of- fit of models is assessed with the proportion of explained variance and root- mean squared error (RMSE). Since increasing the number of combined explanatory variables necessarily increases the model fit and may lead to overfitting, we penalize model complexity by assessing the changes in the Akaike Information Criterion (AIC) in parallel. The best performing combination of covariates is the one which maximizes correlation (or proportion of explained variance) and minimizes RMSE and AIC (Fig. S2). We train the spatial component of our erosion model only on those segments classified as "high quality" with respect to erosion data. We include medium- quality segments to train the model for the high- erosion case to increase our sample size and thus also statistical robustness. We validate each combination of regression coefficients with unseen data by performing a leave- one- out cross validation test. We use a Bootstrap approach with 10 thousand sampling iterations to obtain distributions of model coefficient estimates, and thus their associated uncertainties.  
+
+Three variables compose the best performing combinations: a) daily- mean thawing temperature exposure, expressed as the yearly- accumulated daily positive temperature divided by the number of positive- temperature days per year \(T_{day}[\mathrm{^{\circ}C}\) year \(^{- 1}\) ], b) daily- mean wave exposure, expressed as the yearly- accumulated daily significant wave heights divided by the number of open- water days per year \(H_{day}[\mathrm{m}\) year \(^{- 1}\) ], and c) ground- ice content \(\theta [\%\) of soil volume]. On one hand, combining ground- ice content with daily- mean wave exposure \((\theta +H_{day})\) explains about \(47\%\) of the observed spatial variance among non- extreme (2.5 m/year threshold) erosion segments \((r = 0.69\) \(9 - 95^{\mathrm{th}}\) - percentile range: \(r = 0.60 - 0.78\) , Fig. 2b, Fig. S3a). On the other hand, combining ground- ice content with the daily- mean thawing temperature exposure \((\theta +T_{day})\) explains about \(36\%\) of the variance among extreme- erosion segments \((r = 0.61\) \(9 - 95^{\mathrm{th}}\) - percentile range: \(r = 0.31 - 0.94\) Fig. 2c, Fig. S3a). The linear regression coefficients \(b\) obtained with the selected variable combinations are statistically significant \((p< 0.01)\)  
+
+\[\Delta E(x,t) = \left\{ \begin{array}{ll}b_{\theta}\Delta \theta (x) + b_{H}\Delta H_{day}(x,t) & \mathrm{if} E_{obs}(x)< 2.5\mathrm{myear}^{-1}\\ b_{\theta}^{\prime}\Delta \theta (x) + b_{T}\Delta T_{day}(x,t) & \mathrm{otherwise} \end{array} \right. \quad (5)\]  
+
+Swapping the combinations and groups, that is, using \(\theta +H_{day}\) for the extreme and \(\theta +T_{day}\) for the non- extreme erosion segments, yields overall poorer fits (Fig. S3a,b) and less robust estimation of regression coefficients (Fig. S3c- e). We also test the sensitivity of these results to the choice
+
+<--- Page Split --->
+
+of the threshold to define extreme erosion. Allowing for an overlap between the extreme and non- extreme segments by lowering the threshold to \(2.0\mathrm{m / year}\) , for example, increases the robustness of the \(T_{day}\) regression coefficient estimate for the extreme group (Fig. S3d) by increasing the number of data points, and yields a similar fit to that of the higher threshold \((\theta +T_{day}\) in Fig. S3a,b) and also similar ground- ice coefficients \((\theta +T_{day}\) in Fig. 3Sc).  
+
+Finally, the total erosion is constrained to the open- water period, and set to zero whenever and wherever sea- ice concentration (SIC) is above \(15\%\) at the coast. Combining the temporal (Eq. 2) and spatial (Eq. 5) components into our total erosion model (Eq. 1), conditioned by open- water and the extreme- erosion threshold, our model assumes the complete form:  
+
+\[E(x,t) = \left\{ \begin{array}{ll}a_{TD}\overline{T} (t) + a_{TA}\overline{H} (t) + \left\{ \begin{array}{ll}b_{\theta}\Delta \theta (x) + b_{H}\Delta H_{day}(x,t) & \mathrm{if} E_{obs}(x)< 2.5\mathrm{m / year}\\ b_{\theta}^{\prime}\Delta \theta (x) + b_{T}\Delta T_{day}(x,t) & \mathrm{if} E_{obs}(x)\geq 2.5 \end{array} \right. & \mathrm{if} SIC(x)< 15\% \\ 0 & \mathrm{if} SIC(x)\geq 15\% \end{array} \right. \quad (6)\]  
+
+## Bias correction  
+
+Before forcing the erosion model with MPI- ESM data, we adjust the historical and scenario simulations for climate biases. The bias is removed between ERA20C data (used to estimate our model parameters) and MPI- ESM ensemble means at the coastal segments and reference periods from observations. The modelled distributions are shifted and scaled, so that their means and spread fit those of ERA20C at the reference time.  
+
+## Organic carbon fluxes  
+
+We translate linear erosion rates into volumetric erosion rates \(\mathrm{E}_{vol}[\mathrm{m}^3 \mathrm{year}^{- 1}]\) , sediment fluxes \(\mathrm{S}[\mathrm{Kg} \mathrm{year}^{- 1}]\) , and carbon fluxes \(C_{flux}\) [Kg year \(^{- 1}\) ], considering the mean geometry and ground properties of each coastal segment.  
+
+\[\begin{array}{c}{E_{vol}(x,t) = E(x,t)L(x)h(x)}\\ {S(x,t) = E_{vol}(x,t)(1 - \theta (x))\rho (x)}\\ {C_{flux}(x,t) = S(x,t)C_{conc}(x)} \end{array} \quad (7)\]  
+
+where \(L\) and \(h\) are the segments' mean length and elevation [m], \(\theta\) is the ground- ice content \([\% \mathrm{volume}]\) , \(\rho\) is the soil bulk density [Kg/m \(^3\) ], and \(C_{conc}\) is the organic carbon concentration \([\% \mathrm{weight}]\) . We integrate over the coastal segments:  
+
+\[\overline{C}_{flux}(t) = \sum_{x}C_{flux}(x,t) \quad (8)\]  
+
+to obtain the total Arctic flux.
+
+<--- Page Split --->
+
+## Sensitivity to climate change  
+
+We estimate the sensitivity of the organic carbon release by Arctic coastal erosion to climate change following the approach of Friedlingstein et al. (2006) [51]; however, with a simplified set of tools. In their work, Friedlingstein et al. compare pairs of "coupled" and "uncoupled" simulations, where the increasing atmospheric \(\mathrm{CO_2}\) concentration either affects climate, or is neutral in terms of radiative effect. This pairwise comparison is necessary because the land- atmosphere and ocean- atmosphere carbon fluxes respond to changes in both climate and atmospheric \(\mathrm{CO_2}\) concentrations. Therefore, the difference between their coupled and uncoupled simulations provide the isolated effect of the \(\mathrm{CO_2}\) - induced changes in climate on carbon fluxes from the effect of the changing atmospheric \(\mathrm{CO_2}\) concentration. In our case, changes in atmospheric \(\mathrm{CO_2}\) alone do not induce any Arctic coastal erosion response, if not by its radiative effect. An uncoupled simulation, where \(\mathrm{CO_2}\) does not induce a change in climate, would not yield any change in the organic carbon released by Arctic coastal erosion. Therefore, we can estimate the sensitivity of the organic carbon release by Arctic coastal erosion to climate \(\gamma\) [TgC year \(^{- 1}\mathrm{^\circ C^{- 1}}\) ] by comparing changes in global mean surface temperature and the resulting changes in carbon fluxes from erosion.  
+
+## Probability and onset of emergence from the historical range  
+
+We define the yearly probability density distribution of a modelled variable \(\psi\) as the normal distribution \(N(t)\) at year \(t\) . The mean of \(N(t)\) is the ensemble mean and its standard deviation is the ensemble standard deviation (plus the standard deviation of the distribution of erosion model uncertainties in specific situations, made clear in the text). Similarly, the historical range of a modelled variable \(\psi\) is the normal distribution fitted to its average over the period 1850- 1950 \(N_{hist}\) . We calculate the area of distributions \(A_{hist} = \int N_{hist}d\psi\) and \(A(t) = \int N(t)d\psi\) to determine their overlap \(A_{hist}\cap A(t)\) . We define the probability of emergence from the historical range \(P(t)\) , i.e. the probability that \(N(t)\) be different from \(N_{hist}\) , as the fraction of \(A(t)\) that emerges from \(A_{hist}\) :  
+
+\[P(t) = \frac{A(t) - A_{hist}\cap A(t)}{A(t)}\times 100[\% ] \quad (9)\]  
+
+We define the onset of emergence as the year when the ensemble mean is larger than \(\mu + 2\sigma\) from historical range \(N_{hist}\) .  
+
+## Estimation of uncertainties  
+
+All ranges of uncertainties, except when clearly stated otherwise, are calculated with a Bootstrap method, which suits cases where the number of data is relatively small. From any vector \(\mathbf{X}\) of arbitrary length, a large number (i.e. 10 thousand) of vectors \(\mathbf{X}^{i}\) \((i = 1, 2, \dots 10\mathrm{k})\) is generated by sampling with replacement from \(\mathbf{X}\) . The uncertainty of any statistics of \(\mathbf{X}\) is estimated from the distribution of \(i\) realizations of the statistics obtained from \(\mathbf{X}^{i}\) .
+
+<--- Page Split --->
+
+## References  
+
+[1] Aré, F. E. "Thermal abrasion of sea coasts (part I)". In: Polar Geography and Geology 12.1 (1988), pp. 1- 1. DOI: 10.1080/10889378809377343.  
+
+[2] Overeem, Irina et al. "Sea ice loss enhances wave action at the Arctic coast". In: Geophysical Research Letters 38.17 (2011). DOI: https://doi.org/10.1029/2011GL048681.  
+
+[3] Casas- Prat, Mercè and Wang, Xiaolan L. "Sea Ice Retreat Contributes to Projected Increases in Extreme Arctic Ocean Surface Waves". In: Geophysical Research Letters 47.15 (2020), e2020GL088100. DOI: https://doi.org/10.1029/2020GL088100.  
+
+[4] Barnhart, K. R., Overeem, I., and Anderson, R. S. "The effect of changing sea ice on the physical vulnerability of Arctic coasts". In: Cryosphere 8.5 (2014), pp. 1777- 1799. DOI: 10.5194/tc- 8- 1777- 2014.  
+
+[5] Barnhart, Katherine R. et al. "Mapping the future expansion of Arctic open water". In: Nature Climate Change 6.3 (2016), pp. 280- 285. DOI: 10.1038/nclimate2848.  
+
+[6] Jones, B M et al. "Increase in the rate and uniformity of coastline erosion in Arctic Alaska". In: Geophysical Research Letters 36.3 (2009). DOI: 10.1029/2008GL036205.  
+
+[7] Jones, Benjamin M et al. "A decade of remotely sensed observations highlight complex processes linked to coastal permafrost bluff erosion in the Arctic". In: Environmental Research Letters 13.11 (2018), p. 115001. DOI: 10.1088/1748- 9326/aaef471.  
+
+[8] Günther, F et al. "Observing Muostakh disappear: Permafrost thaw subsidence and erosion of a ground- ice- rich Island in response to arctic summer warming and sea ice reduction". In: Cryosphere 9.1 (2015), pp. 151- 178. DOI: 10.5194/tc- 9- 151- 2015.  
+
+[9] Irrgang, Anna M. et al. "Variability in Rates of Coastal Change Along the Yukon Coast, 1951 to 2015". In: Journal of Geophysical Research: Earth Surface 123.4 (2018), pp. 779- 800. DOI: https://doi.org/10.1002/2017JF004326.  
+
+[10] Jones B. M. Irrgang A. M., Farquharson A. M. et al. "Coastal Permafrost Erosion". In: Arctic Report Card 2020. Ed. by Frey K. E. et al. 2020. DOI: https://doi.org/10.25923/e47w-dw52.  
+
+[11] Stroeve, Julienne and Notz, Dirk. "Changing state of Arctic sea ice across all seasons". In: Environmental Research Letters 13.10 (2018), p. 103001. DOI: 10.1088/1748- 9326/aaed56.  
+
+[12] Serreze, MC et al. "The emergence of surface- based Arctic amplification". In: The Cryosphere 3.1 (2009), p. 11. DOI: https://doi.org/10.5194/tc- 3- 11- 2009.  
+
+[13] Cohen, J et al. "Recent Arctic amplification and extreme mid- latitude weather". In: Nature Geoscience 7.9 (2014), pp. 627- 637. DOI: 10.1038/ngeo2234.  
+
+[14] Biskaborn, Boris K. et al. "Permafrost is warming at a global scale". In: Nature Communications 10.1 (Dec. 2019), p. 264. DOI: 10.1038/s41467- 018- 08240- 4.  
+
+[15] Landrum, Laura and Holland, Marika M. "Extremes become routine in an emerging new Arctic". In: Nature Climate Change (2020), pp. 1- 8. DOI: https://doi.org/10.1038/s41558- 020- 0892- z.  
+
+[16] Notz, Dirk and Community, SIMIP. "Arctic Sea Ice in CMIP6". In: Geophysical Research Letters 47.10 (2020), e2019GL086749. DOI: https://doi.org/10.1029/2019GL086749.
+
+<--- Page Split --->
+
+[17] Dobrynin, M., Murawsky, J., and Yang, S. "Evolution of the global wind wave climate in CMIP5 experiments". In: Geophysical Research Letters 39.17 (Sept. 2012). DOI: 10.1029/2012GL052843. 
+
+[18] Dobrynin, Mikhail et al. "Detection and attribution of climate change signal in ocean wind waves". In: Journal of Climate 28.4 (2015), pp. 1578-1591. DOI: 10.1175/JCLI-D-13-00664.1. 
+
+[19] Casas-Prat, Mercè and Wang, Xiaolan L. "Projections of Extreme Ocean Waves in the Arctic and Potential Implications for Coastal Inundation and Erosion". In: Journal of Geophysical Research: Oceans 125.8 (2020), e2019JC015745. DOI: https://doi.org/10.1029/2019JC015745. 
+
+[20] Schaefer, Kevin et al. "The impact of the permafrost carbon feedback on global climate". In: Environmental Research Letters 9.8 (2014), p. 085003. DOI: 10.1088/1748-9326/9/8/085003. 
+
+[21] Schuur, Edward AG et al. "Climate change and the permafrost carbon feedback". In: Nature 520.7546 (2015), pp. 171-179. DOI: https://doi.org/10.1038/nature14338. 
+
+[22] Tanski, George et al. "Rapid CO 2 release from eroding permafrost in seawater". In: Geophysical Research Letters 46 (2019), pp. 11244-11252. DOI: 10.1029/2019GL084303. 
+
+[23] Vonk, Jorien E et al. "Activation of old carbon by erosion of coastal and subsea permafrost in Arctic Siberia". In: Nature 489.7414 (2012), pp. 137-140. DOI: https://doi.org/10.1038/nature11392. 
+
+[24] Wegner, Carolyn et al. "Variability in transport of terrigenous material on the shelves and the deep Arctic Ocean during the Holocene". In: Polar Research 34.1 (2015). DOI: 10.3402/polar.v34.24964. 
+
+[25] Terhaar, Jens et al. "Around one third of current Arctic Ocean primary production sustained by rivers and coastal erosion". In: Nature Communications 12.169 (2021). DOI: https://doi.org/10.1038/s41467-020-20470-z. 
+
+[26] Koven, Charles D, Riley, William J, and Stern, Alex. "Analysis of permafrost thermal dynamics and response to climate change in the CMIP5 Earth System Models". In: Journal of Climate 26.6 (2013), pp. 1877-1900. DOI: https://doi.org/10.1175/JCLI-D-12-00228.1. 
+
+[27] Burke, E. J., Zhang, Y., and Krinner, G. "Evaluating permafrost physics in the Coupled Model Intercomparison Project 6 (CMIP6) models and their sensitivity to climate change". In: The Cryosphere 14.9 (2020), pp. 3155-3174. DOI: 10.5194/tc-14-3155-2020. 
+
+[28] Turetsky, Merritt R et al. "Carbon release through abrupt permafrost thaw". In: Nature Geoscience 13.2 (2020), pp. 138-143. DOI: https://doi.org/10.1038/s41561-019-0526-0. 
+
+[29] Pihl, Erik et al. "10 New Insights in Climate Science 2020 - a Horizon Scan". In: Global Sustainability 4 (2021), pp. 1-65. DOI: 10.1017/sus.2021.2. 
+
+[30] Fritz, Michael, Vonk, Jorien E., and Lantuit, Hugues. "Collapsing Arctic coastlines". In: Nature Climate Change 7.1 (2017), pp. 6-7. DOI: 10.1038/nclimate3188. 
+
+[31] Turetsky, Merritt R et al. Permafrost collapse is accelerating carbon release. 2019. DOI: https://doi.org/10.1038/d41586-019-01313-4. 
+
+[32] Nielsen, David Marcolino et al. "Coastal Erosion Variability at the Southern Laptev Sea Linked to Winter Sea Ice and the Arctic Oscillation". In: Geophysical Research Letters 47.5 (2020), e2019GL086876. DOI: 10.1029/2019GL086876.
+
+<--- Page Split --->
+
+[33] Lantuit, Hugues et al. "The Arctic Coastal Dynamics Database: A New Classification Scheme and Statistics on Arctic Permafrost Coastlines". In: Estuaries and Coasts 35.2 (2012), pp. 383- 400. DOI: 10.1007/s12237- 010- 9362- 6.  
+
+[34] O'Neill, B. C. et al. "The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6". In: Geoscientific Model Development 9.9 (2016), pp. 3461- 3482. DOI: 10.5194/gmd- 9- 3461- 2016.  
+
+[35] Schwalm, Christopher R., Glendon, Spencer, and Duffy, Philip B. "RCP8.5 tracks cumulative \(\mathrm{CO_2}\) emissions". In: Proceedings of the National Academy of Sciences 117.33 (2020), pp. 19656- 19657. DOI: 10.1073/pnas.2007117117.  
+
+[36] Lantuit, Hugues et al. "Coastal erosion dynamics on the permafrost- dominated Bykovsky Peninsula, north Siberia, 1951- 2006". In: Polar Research 30. SUPPL.1 (2011). DOI: 10.3402/polar.v30i0. 7341.  
+
+[37] Lantz, Trevor C. and Kokelj, Steven V. "Increasing rates of retrogressive thaw slump activity in the Mackenzie Delta region, N.W.T., Canada". In: Geophysical Research Letters 35.6 (2008). DOI: https://doi.org/10.1029/2007GL032433.  
+
+[38] Lantuit, H. and Pollard, W.H. "Fifty years of coastal erosion and retrogressive thaw slump activity on Herschel Island, southern Beaufort Sea, Yukon Territory, Canada". In: Geomorphology 95.1 (2008), pp. 84- 102. DOI: https://doi.org/10.1016/j.geomorph.2006.07.040.  
+
+[39] Baranskaya, Alisa et al. "The Role of Thermal Denudation in Erosion of Ice- Rich Permafrost Coasts in an Enclosed Bay (Gulf of Kruzenstern, Western Yamal, Russia)". In: Frontiers in Earth Science 8 (2021), p. 659. DOI: 10.3389/feart.2020.566227.  
+
+[40] Farquharson, Louise M et al. "Temporal and spatial variability in coastline response to declining sea- ice in northwest Alaska". In: Marine Geology 404 (2018), pp. 71- 83. DOI: https://doi.org/10.1016/j.margeo.2018.07.007.  
+
+[41] Grigoriev, MN. "Coastal retreat rates at the Laptev Sea key monitoring sites". In: PANGAEA (2019). DOI: https://doi.org/10.1594/PANGAEA.905519.  
+
+[42] Fuchs, M. et al. "Carbon and nitrogen pools in thermokarst- affected permafrost landscapes in Arctic Siberia". In: Biogeosciences 15.3 (2018), pp. 953- 971. DOI: 10.5194/bg- 15- 953- 2018.  
+
+[43] Ramage, Justine L. et al. "Terrain controls on the occurrence of coastal retrogressive thaw slumps along the Yukon Coast, Canada". In: Journal of Geophysical Research: Earth Surface 122.9 (2017), pp. 1619- 1634. DOI: https://doi.org/10.1002/2017JF004231.  
+
+[44] Couture, Nicole J. et al. "Coastal Erosion of Permafrost Soils Along the Yukon Coastal Plain and Fluxes of Organic Carbon to the Canadian Beaufort Sea". In: Journal of Geophysical Research: Biogeosciences 123.2 (2018), pp. 406- 422. DOI: https://doi.org/10.1002/2017JG004166.  
+
+[45] Grotheer, H. et al. "Burial and Origin of Permafrost- Derived Carbon in the Nearshore Zone of the Southern Canadian Beaufort Sea". In: Geophysical Research Letters 47.3 (2020), e2019GL085897.  
+
+[46] Hilton, Robert G et al. "Erosion of organic carbon in the Arctic as a geological carbon dioxide sink". In: Nature 524.7563 (2015), pp. 84- 87. DOI: https://doi.org/10.1038/nature14653.  
+
+[47] Bröder, Lisa et al. "Bounding cross- shelf transport time and degradation in Siberian- Arctic land- ocean carbon transfer". In: Nature communications 9.1 (2018), pp. 1- 8. DOI: https://doi.org/10.1038/s41467- 018- 03192- 1.
+
+<--- Page Split --->
+
+576 [48] Jong, Dirk et al. "Nearshore Zone Dynamics Determine Pathway of Organic Carbon From Eroding Permafrost Coasts". In: Geophysical Research Letters 47.15 (2020), e2020GL088561. DOI: https://doi.org/10.1029/2020GL088561.  
+
+579 [49] Tanski, George et al. "Permafrost Carbon and CO2 Pathways Differ at Contrasting Coastal Erosion Sites in the Canadian Arctic". In: Frontiers in Earth Science 9 (2021), p. 207. DOI: 10.3389/feart.2021.630493.  
+
+582 [50] Kosaka, Yu and Xie, Shang- Ping. "Recent global- warming hiatus tied to equatorial Pacific surface cooling". In: Nature 501.7467 (2013), pp. 403- 407. DOI: https://doi.org/10.1038/nature12534.  
+
+585 [51] Friedlingstein, Pierre et al. "Climate- carbon cycle feedback analysis: results from the C4MIP model intercomparison". In: Journal of climate 19.14 (2006), pp. 3337- 3353. DOI: https://doi.org/10.1175/JCLI3800.1.  
+
+588 [52] Hoque, Md. Azharul and Pollard, Wayne H. "Arctic coastal retreat through block failure". In: Canadian Geotechnical Journal 46.10 (2009), pp. 1103- 1115. DOI: https://doi.org/10.1139/T09- 058.  
+
+591 [53] Ravens, Thomas M. et al. "Process- Based Coastal Erosion Modeling for Drew Point, North Slope, Alaska". In: Journal of Waterway, Port, Coastal, and Ocean Engineering 138.2 (2011), pp. 122- 130. DOI: 10.1061/(asce)ww.1943- 5460.0000106.  
+
+594 [54] Barnhart, Katherine R. et al. "Modeling erosion of ice- rich permafrost bluffs along the Alaskan Beaufort Sea coast". In: Journal of Geophysical Research: Earth Surface 119.5 (2014), pp. 1155- 1179. DOI: 10.1002/2013JF002845.  
+
+597 [55] Thomas, Matthew A. et al. "Geometric and Material Variability Influences Stress States Relevant to Coastal Permafrost Bluff Failure". In: Frontiers in Earth Science 8 (2020), p. 143. DOI: 10.3389/feart.2020.00143.  
+
+600 [56] Frederick, Jennifer et al. "A thermo- mechanical terrestrial model of Arctic coastal erosion". In: Journal of Computational and Applied Mathematics (2021), p. 113533. DOI: https://doi.org/10.1016/j.cam.2021.113533.  
+
+603 [57] Rolph, R. et al. "ArcticBeach v1.0: A physics- based parameterization of pan- Arctic coastline erosion". In: Geoscientific Model Development Discussions 2021 (2021), pp. 1- 26. DOI: 10.5194/gmd- 2021- 28.  
+
+606 [58] Poli, Paul et al. "ERA- 20C: An atmospheric reanalysis of the twentieth century". In: Journal of Climate 29.11 (2016), pp. 4083- 4097. DOI: 10.1175/JCLI- D- 15- 0556.1.  
+
+608 [59] Mauritsen, Thorsten et al. "Developments in the MPI- M Earth System Model version 1.2 (MPI- ESM1.2) and Its Response to Increasing CO2". In: Journal of Advances in Modeling Earth Systems 11.4 (2019), pp. 998- 1038. DOI: https://doi.org/10.1029/2018MS001400.  
+
+611 [60] The WAMDI Group. "The WAM Model—A Third Generation Ocean Wave Prediction Model". In: Journal of Physical Oceanography 18.12 (1988), pp. 1775- 1810.
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+CoastalErosionSuplement20210608. pdf
+
+<--- Page Split --->

preprint/preprint__7e1dbd5e4b7563929632d79cc02b1d840b5d2ecb348dcd394110c3e708654a44/preprint__7e1dbd5e4b7563929632d79cc02b1d840b5d2ecb348dcd394110c3e708654a44_det.mmd

@@ -0,0 +1,548 @@
+<|ref|>title<|/ref|><|det|>[[44, 107, 936, 177]]<|/det|>
+# Projected increase of Arctic coastal erosion and its sensitivity to warming in the 21st Century  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 610, 238]]<|/det|>
+David Nielsen ( \(\boxed{ \begin{array}{r l} \end{array} }\) david.nielsen@uni- hamburg.de) University of Hamburg https://orcid.org/0000- 0003- 4201- 0373  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 252, 283]]<|/det|>
+Patrick Pieper University of Hamburg  
+
+<|ref|>text<|/ref|><|det|>[[44, 289, 252, 330]]<|/det|>
+Armineh Barkhordarian University of Hamburg  
+
+<|ref|>text<|/ref|><|det|>[[44, 335, 532, 377]]<|/det|>
+Paul Overduin Alfred Wegener Institute for Polar and Marine Research  
+
+<|ref|>text<|/ref|><|det|>[[44, 382, 735, 425]]<|/det|>
+Tatiana Ilyina Max Planck Institute for Meteorology https://orcid.org/0000- 0002- 3475- 4842  
+
+<|ref|>text<|/ref|><|det|>[[44, 429, 727, 472]]<|/det|>
+Victor Brovkin Max Plank Institute for Meteorology https://orcid.org/0000- 0001- 6420- 3198  
+
+<|ref|>text<|/ref|><|det|>[[44, 476, 252, 516]]<|/det|>
+Johanna Baehr University of Hamburg  
+
+<|ref|>text<|/ref|><|det|>[[44, 521, 252, 562]]<|/det|>
+Mikhail Dobrynin University of Hamburg  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 604, 102, 621]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 641, 588, 661]]<|/det|>
+Keywords: climate change, erosion coastal erosion, permafrost  
+
+<|ref|>text<|/ref|><|det|>[[44, 679, 300, 698]]<|/det|>
+Posted Date: June 25th, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 717, 463, 736]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 634673/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 753, 910, 797]]<|/det|>
+License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 832, 933, 876]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Climate Change on February 14th, 2022. See the published version at https://doi.org/10.1038/s41558- 022- 01281- 0.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[157, 114, 864, 166]]<|/det|>
+# Projected increase of Arctic coastal erosion and its sensitivity to warming in the \(21^{\mathrm{st}}\) Century  
+
+<|ref|>text<|/ref|><|det|>[[120, 192, 898, 231]]<|/det|>
+David Marcolino Nielsen \(^{1,2,*}\) , Patrick Pieper \(^{1}\) , Arminen Barkhordarian \(^{1}\) , Paul Overduin \(^{3}\) , Tatiana Ilyina \(^{4}\) , Victor Brovkin \(^{1,4}\) , Johanna Baehr \(^{1}\) , and Mikhail Dobrynin \(^{5,1}\)  
+
+<|ref|>text<|/ref|><|det|>[[125, 252, 896, 395]]<|/det|>
+\(^{1}\) Center for Earth System Research and Sustainability (CEN), Universität Hamburg, Hamburg, Germany \(^{2}\) International Max Planck Research School on Earth System Modelling, Max Planck Institute for Meteorology, Hamburg, Germany \(^{3}\) Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research, Potsdam, Germany \(^{4}\) Max Planck Institute for Meteorology, Hamburg, Germany \(^{5}\) Deutscher Wetterdienst, Hamburg, Germany \(^{*}\) Correspondence to: david.nielsen@uni- hamburg.de  
+
+<|ref|>text<|/ref|><|det|>[[456, 425, 560, 440]]<|/det|>
+June 16, 2021  
+
+<|ref|>sub_title<|/ref|><|det|>[[94, 488, 235, 510]]<|/det|>
+## 1 Abstract  
+
+<|ref|>text<|/ref|><|det|>[[92, 528, 907, 739]]<|/det|>
+2 Arctic coastal erosion damages infrastructure, threatens coastal communities, and releases organic carbon from permafrost. However, the magnitude, timing and sensitivity of coastal erosion increase to global warming remain unknown. Here, we project the Arctic- mean erosion rate to roughly double by 2100 and very likely exceed its historical range of variability by mid- \(21^{\mathrm{st}}\) century. The sensitivity of erosion to warming also doubles, reaching \(0.4 - 0.5 \mathrm{m}\) year \(^{- 1} \circ \mathrm{C}^{- 1}\) and \(2.3 - 2.8 \mathrm{TgC}\) year \(^{- 1} \circ \mathrm{C}^{- 1}\) by the end of the century under moderate and high- emission scenarios. Our first \(21^{\mathrm{st}}\) - century pan- Arctic coastal erosion rate projections should inform policy makers on coastal conservation and socioeconomic planning. Our organic carbon flux projections also lay out the path for future work to investigate the impact of Arctic coastal erosion on the changing Arctic Ocean, on its role as a global carbon sink, and on the permafrost- carbon feedback.  
+
+<|ref|>sub_title<|/ref|><|det|>[[94, 771, 185, 792]]<|/det|>
+## 12 Main  
+
+<|ref|>text<|/ref|><|det|>[[92, 810, 907, 937]]<|/det|>
+13 Arctic coast erosion is caused by a combination of thermal and mechanical drivers. Permafrost 14 thaw and ground- ice melt lead to soil decohesion and slumping, while surface ocean waves me- 15 chanically abrade the Arctic coast [1]. Sea- ice loss expands the fetch for waves [2, 3], and prolongs 16 the open- water season, increasing the vulnerability of the Arctic coast to erosion [4, 5]. In the past 17 decades, coastal retreat rates have increased throughout the Arctic, often by a factor of two or more 18 [6- 10]. The historical acceleration of erosion in the Arctic is linked with the observed decreasing
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[87, 65, 907, 234]]<|/det|>
+sea- ice cover [2, 4, 11], increasing air surface [12, 13] and permafrost temperatures [14]. As for the future, Arctic surface air temperature is projected to exceed its natural range of variability within the next decades [15]. Arctic sea ice decline has already exceeded natural variability [15], and summer ice- free conditions are projected by mid- \(21^{\mathrm{st}}\) century [16]. New regimes of surface waves are also projected in the Arctic Ocean and along the coast [17- 19]. Consequently, Arctic coastal erosion rates are expected to increase in the coming decades. However, the extent of this increase is still unknown, as no projections of Arctic coastal erosion rates are available. To fill this gap, we present the first \(21^{\mathrm{st}}\) - century projections of coastal erosion at the pan- Arctic scale.  
+
+<|ref|>text<|/ref|><|det|>[[88, 237, 907, 468]]<|/det|>
+The thawing of permafrost globally releases organic carbon (OC) and increases atmospheric and oceanic greenhouse gas concentrations, feeding back to further warming [20- 23]. Arctic coastal erosion alone releases about as much OC as all the Arctic rivers combined [23, 24], fueling about one- fifth of Arctic marine primary production [25]. Despite consistent improvements in the representation of permafrost dynamics [26, 27], the current generation of Earth system models (ESMs) does not account for abrupt permafrost thaw, which may cause projections of OC losses to be largely underestimated [28, 29]. Arctic coastal erosion is one form of abrupt permafrost thaw [22] and a relevant component of the Arctic carbon cycle [23, 30]. Nonetheless, it has not been considered in climate projections so far. The scale mismatch between Arctic coastal erosion and modern ESMs requires the development of holistic models, that account for the key large- scale processes to bridge this gap [30- 32].  
+
+<|ref|>text<|/ref|><|det|>[[88, 471, 907, 660]]<|/det|>
+In this study, we present a novel approach to represent Arctic coastal erosion at the scales of modern ESMs. We develop a semi- empirical Arctic coastal erosion model combining observations from the Arctic Coastal Dynamics (ACD) database [33], climate reanalyses, ESM and ocean surface wave simulations. Our model considers the main thermal and mechanical drivers of erosion as dynamical variables, represented by yearly- accumulated positive temperatures and significant wave heights, and constant ground- ice content from observations. Our approach allows us to make \(21^{\mathrm{st}}\) - century projections of coastal erosion at the pan- Arctic scale. We quantify the magnitude, timing and sensitivity of Arctic coastal erosion and its associated OC loss in the context of climate change.  
+
+<|ref|>sub_title<|/ref|><|det|>[[90, 686, 536, 707]]<|/det|>
+## Emergence of Arctic coastal erosion  
+
+<|ref|>text<|/ref|><|det|>[[88, 718, 907, 930]]<|/det|>
+We project the Arctic- mean coastal erosion rate to increase from \(0.9 \pm 0.4 \mathrm{m / year}\) during the historical period (1850- 1950) to between \(2.0 \pm 0.7\) and \(2.6 \pm 0.8 \mathrm{m / year}\) by the end of the \(21^{\mathrm{st}}\) Century (2081- 2100), in the context of anthropogenic climate change, according to the socio- economic pathway (SSP) scenarios SSP2- 4.5 and SSP5- 8.5, respectively (Fig. 1a). This translates to an increase of the Arctic- mean coastal erosion rate by a factor of about between 2.2 and 2.9 by the end of the century with respect to the historical period. The SSP2- 4.5 and SSP5- 8.5 scenarios describe medium and high radiative forcings due to greenhouse gas emissions [34], respectively, and include the pathway of the current cumulative \(\mathrm{CO}_{2}\) emissions [35]. In both scenarios, our projections show that the Arctic- mean erosion exceeds its historical range of variability before the end of the century (Fig. 1b).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[122, 70, 900, 408]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 426, 906, 556]]<|/det|>
+<center>Figure 1: Arctic coastal erosion projections. a) Time evolution of the Arctic-mean coastal erosion rate, expressed as the combined effect of its thermal and mechanical drivers. b) Yearly probabilities that the Arctic-mean coastal erosion rate leaves the historical range of variability, calculated from distributions of ensemble spread and erosion model uncertainties (see Methods). In both scenarios, it is very likely ( \(>90\%\) probability) that the Arctic-mean erosion emerges from its historical range by mid \(21^{\mathrm{st}}\) century, although the exact time of emergence is sensitive to our erosion model uncertainties. The thermal (c) and mechanical (d) drivers of erosion, expressed as yearly-accumulated daily positive degrees and significant wave heights, respectively. The erosion time series depict long-term means and therefore show little interannual variability in comparison to its drivers. </center>  
+
+<|ref|>text<|/ref|><|det|>[[115, 582, 906, 771]]<|/det|>
+We find it likely ( \(\geq 66\%\) probability) that the Arctic- mean erosion exceeds its historical range by around 2023, and very likely ( \(\geq 90\%\) probability) by 2049 (Fig. 1b), considering the largest distributions of uncertainties in our projections (i.e. ensemble spread and erosion model uncertainties). The emergence of the Arctic- mean erosion rate would very likely have happened by around 2010, if we take only the ensemble spread to define the historical range. Significant differences in projections between the two scenarios are only noticeable in the second half of the century, after a complete emergence from the historical range. Our erosion time- of- emergence estimates reflect those of its drivers, which take place around mid- \(21^{\mathrm{st}}\) Century (Fig. 1c,d), in accordance with previous studies [15, 16].  
+
+<|ref|>text<|/ref|><|det|>[[115, 774, 906, 920]]<|/det|>
+Arctic coastal erosion is typically caused by a combination of thermo- denudation (TD) and thermo- abrasion (TA) [1], which act together to thaw permafrost, melt ground ice, abrade and transport coastal material off shore. We take yearly- accumulated daily positive temperatures and significant wave heights to represent TD and TA: hereafter, the thermal and mechanical drivers of erosion, respectively. As various landform types compose the Arctic coast, the relative contribution of the thermal and mechanical drivers differs at the local scale. Erosion is predominantly thermally driven at retrogressive thaw slumps, observed at the Bykovsky Peninsula, Laptev Sea [36], and in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 66, 907, 448]]<|/det|>
+the Mackenzie Delta region – Beaufort Sea [37, 38], for example (Fig. 2a), as the sediment transport from ocean waves play a secondary role in coastal retreat in such formations. Erosion is also predominantly thermally driven in enclosed bays and in coastal segments protected by spits and barrier islands, where the fetch for ocean waves is limited [39], although barrier island themselves are often susceptible to wave abrasion [40]. In contrast, erosion of ice-rich cliffs, which occur extensively along the Beaufort and Laptev Sea coast for example [6–8], requires the mechanical action from ocean waves to open notches at the land-sea interface, causing the subsequent failure of often still frozen large blocks of permafrost. In some locations, the relative contribution of the thermal and mechanical drivers is more balanced than described above. At Muostakh Island in the Laptev Sea, for example, thermo-denudation and abrasion are estimated to contribute similarly to maintain erosion rates of up to \(25\mathrm{m / year}\) [8]. In our erosion model, we initially assume equal contributions from the thermal and mechanical drivers at the pan-Arctic scale during the observational period. This assumes that deviations occur comparably in both directions. We also make extreme \(10 - 90\%\) and \(90 - 10\%\) scenarios of relative thermal-mechanical contributions to test the sensitivity of our results to that assumption (see Methods and Table S1). Attributing \(90\%\) of mechanical contribution yields about \(15 - 20\%\) larger Arctic-mean coastal erosion projections by 2100 (and vice- versa), because the Arctic-mean wave exposure increases more than the thawing temperature exposure along the coast, with respect to their historical values (Fig. S1a).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 472, 440, 494]]<|/det|>
+## Spatial variability of erosion  
+
+<|ref|>text<|/ref|><|det|>[[92, 504, 907, 929]]<|/det|>
+The thermal and mechanical drivers of erosion explain about \(36 - 47\%\) of its observed spatial variability in multiple linear regression models. On one hand, wave exposure, combined with ground- ice content, best explains the spatial variability of erosion in most of the coastal segments \((r = 0.69 \pm 0.12\) , mean \(\pm 2\sigma\) , Fig. 2b), where erosion is not extremely high \((\sim 90^{\mathrm{th}}\) percentile, \(< 2.5\mathrm{m / year}\) ). The local wave exposure information indeed integrates several important sources of erosion variability. Not only does wave exposure promote cliff abrasion and subsequent sediment transport, but it is also proportional to open- water season (OWS) duration, which has been suggested to be the first- order driver of coastal erosion rate variability [2, 32]. In addition, sea- ice melt, and thus increasing OWS duration, responds to increasing surface air temperature, which also drives permafrost thaw and thus erosion by thermo- denudation. On the other hand, spatial differences among segments of extremely high long- term erosion rates are best characterized by thawing temperature exposure combined with ground- ice content \((r = 0.61 \pm 0.42\) , Fig. 2c). This suggests that thermo- denudation plays a more important role in driving coastal erosion rates at extreme- erosion segments, than at non- extreme ones. Among both extreme and non- extreme erosion segments, ground ice adds explanatory power, as it increases the susceptibility of permafrost to thaw and hence erosion. Our results are in accordance with previous work, which reported weak spatial correlations between ground- ice content and erosion rates [33]. Strong temporal correlations between erosion and thawing temperature exposure have also been reported for Muostakh Island – Laptev Sea [8], where erosion rates are often in the range between 10 and \(20\mathrm{m / year}\) [8, 41]. We further combine the temporal evolution of the Arctic- mean erosion with its spatial
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 70, 900, 435]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 446, 905, 563]]<|/det|>
+<center>Figure 2: Observed and modelled erosion rate spatial variability. a) Observed long-term coastal erosion mean rates from the ACD database [33] used in this study (see Methods). Modelled against observed erosion rates in (b) non-extreme and (c) extreme erosion segments. Observed values are denoted by colored circles on the maps and on the scatter plots. Uncertainties represent \(2\sigma\) confidence intervals from the distribution of regression coefficients. Modelled historical-mean (1850-1900) (d) and end-of-the-century (2081-2100) erosion rates according to the SSP2-4.5 (e) and SSP5-8.5 (f) scenarios. The histograms in g display the historical and projected erosion time-means from the maps in d, e and f. Distributions shift and spread over time. </center>  
+
+<|ref|>text<|/ref|><|det|>[[115, 588, 864, 605]]<|/det|>
+distribution to make projections of erosion rates at the coastal segment resolution (Fig. 2d- f).  
+
+<|ref|>text<|/ref|><|det|>[[110, 608, 905, 925]]<|/det|>
+The geographical distribution of low and high- erosion segments does not change substantially from observations over time in our projections, which is partially a consequence of our model design, as explained by the three following reasons. First, we assume that the spatial model coefficients, empirically determined, remain unchanged throughout our simulations. Second, ground- ice content, an explanatory variable in our regression model, is also assumed constant over time. Third, our regression model accounts for only a fraction of the spatial variability in erosion, and may thus underestimate larger spatial changes to occur over time. Moreover, and independent from model design, local anomalies of the dynamical variables (i.e. local wave and thawing temperature exposure) are smaller in magnitude than their Arctic- mean increase. Therefore, our modelled changes in the spatial variability of erosion are small in comparison to its Arctic- mean increase. Nonetheless, our modelled spatial spread of erosion increases with time (Fig. 2g). The \(5^{\text{th}}\) - \(95^{\text{th}}\) percentile range of erosion rate distributions increases from 3.6 (0- 3.6) m/year in the historical period to 3.9 (0.9- 4.8) and 4.2 (1.4- 5.7) m/year in the SSP2- 4.5 and SSP5- 8.5 scenarios, respectively. Temporally resolved erosion rate observations are rare, often sparse in time, and only available at a relatively small number of locations [10]. Only with such observations, temporally resolved and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[82, 66, 904, 105]]<|/det|>
+at the pan-Arctic scale, would empirical models be able to better constrain the temporal evolution of spatial variability of coastal erosion.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 132, 608, 154]]<|/det|>
+## Spatial variability of organic carbon losses  
+
+<|ref|>text<|/ref|><|det|>[[115, 164, 905, 355]]<|/det|>
+The pan- Arctic OC loss from coastal erosion increases from 6.9 (1.5- 12.3) TgC year \(^{- 1}\) during the historical period to between 13.1 (6.4- 19.7) TgC year \(^{- 1}\) and 17.2 (9.0- 25.4) TgC year \(^{- 1}\) by the end of the century in the SSP2.4- 5 and SSP5- 8.5 scenarios, respectively (Fig. 3). For the present- day climate (i.e. the period for which erosion observations are available), we estimate a pan- Arctic OC loss from coastal erosion of 8.5 (3.3- 13.7) TgC year \(^{- 1}\) . Both our simulated present- climate mean and uncertainty range are comparable with previous estimates from observations [24, 33]. Our projections suggest a pan- Arctic OC flux increase by a factor of between 1.5 and 2.0 with respect to the present- day climate, or by a factor of between 1.9 and 2.5 by 2100 with respect to the historical period.  
+
+<|ref|>image<|/ref|><|det|>[[130, 371, 884, 608]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 624, 905, 727]]<|/det|>
+<center>Figure 3: Projected organic carbon loss. Changes in organic carbon released annually by coastal erosion according to observations-based estimates and in our model simulations for the historical period (1850-1950), current climate (according to observations from the ACD [33]) and at the end of the \(21^{\mathrm{st}}\) century (2081-2100) in the two future scenarios. The height of bars represent the total uncertainty of our projections, which we disentangle between ensemble spread, spatial and temporal erosion model components. Most of the uncertainties originate from the empirical estimates of the erosion model parameters (76-97%) and the smallest fraction to the ensemble spread (3-24%). </center>  
+
+<|ref|>text<|/ref|><|det|>[[115, 747, 905, 935]]<|/det|>
+The Laptev and East Siberian Seas (LESS, Fig.2a) together account for about three quarters of the pan- Arctic OC losses in our simulations, in accordance with observations- based estimates [24]. This also holds truth for future scenarios. The reason for the relatively high OC fluxes from the LESS coast is twofold. First, the region comprises coastal segments of extremely rapid erosion, often between 10 and \(20\mathrm{m / year}\) [8, 41]. Second, the LESS coast is dominated by Yedoma ice- complex deposits, where ground- ice concentration reaches more than \(80\%\) of soil volume [8, 42], and organic- carbon content is extremely high, reaching about \(5\%\) of weight [33]. From the LESS, we simulate a present- climate OC flux of 6.5 (2.4- 10.6) TgC year \(^{- 1}\) , comparable to the 2.9- 11.0 TgC year \(^{- 1}\) range estimated by Wegner et al. (2015) [24], and comprising the ACD
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[91, 66, 907, 343]]<|/det|>
+value of \(7.7 \mathrm{TgC}\) year \(^{- 1}\) . In an extensive campaign over the LESS continental shelf, Vonk et al. (2012)[23] determined that about \(20 \mathrm{TgC}\) year \(^{- 1}\) are buried in the LESS sediment, which would originate from a combination of coastal and seafloor erosion. Accounting for degradation before burial and assuming an equal contribution from coastal and subsea erosion, about 11 (7- 15) \(\mathrm{TgC}\) year \(^{- 1}\) would be released by coastal erosion alone. The LESS estimate of Vonk et al. (2012) [23] is \(43 - 57\%\) larger than other observations- based estimates [24] and about \(69\%\) larger than our present- climate modelled value. These differences are likely due to extensive and high- resolution sampling, allowing for more accurate upscaling [23]. However, the uncertainties associated with the contribution between coastal and subsea erosion comprehend our modelled range (their Table S6 [23]). Therefore, an underestimation from our side is not conclusive. From the LESS coast, we project an increase in OC fluxes from 5.3 (1.0- 9.6) \(\mathrm{TgC}\) year \(^{- 1}\) in the historical period to 9.6 (5.7- 13.4) \(\mathrm{TgC}\) year \(^{- 1}\) in the SSP2- 4.5 and 12.4 (7.8- 17.1) \(\mathrm{TgC}\) year \(^{- 1}\) in the SSP5- 8.5 scenarios by 2100, which translates to an increase by a factor of between 1.8 and 2.3.  
+
+<|ref|>text<|/ref|><|det|>[[91, 344, 907, 595]]<|/det|>
+The Beaufort Sea coast accounts for about half of the remaining fraction of pan- Arctic OC flux, releasing 0.9 (0.4- 1.4) \(\mathrm{TgC}\) year \(^{- 1}\) during the present climate in our simulations, in agreement with the 0.7 \(\mathrm{TgC}\) year \(^{- 1}\) estimates from the ACD [33], however larger than previous estimates of 0.2- 0.4 \(\mathrm{TgC}\) year \(^{- 1}\) [24] (Fig. 3). Hotspots of extreme erosion are also observed in the Beaufort Sea coast. Extensive field work has been recently carried out, especially in the Yukon coast region, showing increasing erosion rates and suggesting that the associated OC fluxes could have been previously underestimated [9, 22, 43- 45]. We project an OC flux increase from the Beaufort Sea coast from 0.7 (0.2- 1.2) \(\mathrm{TgC}\) year \(^{- 1}\) in the historical period to between 1.6 (0.9- 2.3) \(\mathrm{TgC}\) year \(^{- 1}\) and 2.3 (1.4- 3.1) \(\mathrm{TgC}\) year \(^{- 1}\) by 2100 in the SSP2- 4.5 and SSP5- 8.5 scenarios, respectively, translating to an increase by a factor of between 2.3 and 3.3. The remaining marginal Arctic Seas contribute with yearly OC fluxes at absolute amounts similar to those from Beaufort Sea in our projections, accounting for about \(12 - 14\%\) of the pan- Arctic totals.  
+
+<|ref|>text<|/ref|><|det|>[[91, 598, 907, 895]]<|/det|>
+Coastal erosion is estimated to sustain about one fifth of the total Arctic marine primary production at present- climate conditions [25]. Therefore, the projected additional OC loss could have a substantial impact on the Arctic marine biogeochemistry. However, the fate of the organic carbon released by Arctic coastal erosion is currently under active debate. Field work has shown that between about \(13\%\) and \(65\%\) of the OC released into the ocean by coastal erosion could settle in the marine sediment [44- 46], slowing down remineralization. In the sediment, organic matter degradation would then take place at millennial time scales [47]. However, in the shallow nearshore zone, resuspension driven by waves and storm activity increases the residence time of OC in the water column, and allows for more effective remineralization [48]. Moreover, partial degradation of the eroded material takes place before it enters the ocean, releasing greenhouse gases directly to the atmosphere [22, 23, 49]. The OC degradation time scale thus also depends on its transit time onshore [49]. It is therefore challenging to determine short- term impacts from the projected additional OC fluxes from coastal erosion, as large uncertainties still remain regarding pathways of OC degradation.  
+
+<|ref|>text<|/ref|><|det|>[[140, 899, 904, 916]]<|/det|>
+We partition the uncertainty sources in our projections between three sources: ensemble spread,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 66, 906, 299]]<|/det|>
+temporal, and spatial erosion model components (see Methods). Our erosion model contributes the most to the uncertainties in our simulations: from about \(76\%\) of the total uncertainty range in the historical period and up to \(97\%\) by the end of the century in SSP5- 8.5. The ensemble spread is responsible for the remaining \(24\%\) of the total uncertainty during the historical period, and for only \(3\%\) to \(6\%\) of the total range at the end of the future scenarios. The spatial component of the erosion model accounts for about half of the total range of uncertainties, on average, without significant changes in proportion over time. The fraction of uncertainties stemming from the temporal model component increases from about \(33\%\) of the total range in the historical period to about \(55\%\) by the end of the century in SSP5- 8.5 due to the increasing magnitude of the erosion drivers. The distribution of sources of uncertainties in our projections is qualitatively similar between the pan- Arctic and the regional totals.  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 324, 792, 346]]<|/det|>
+## Sensitivity of erosion and carbon losses to climate change  
+
+<|ref|>text<|/ref|><|det|>[[111, 355, 906, 824]]<|/det|>
+The sensitivity of Arctic coastal erosion to climate change increases over time in our simulations, and is tightly related with the Arctic amplification (AA) [12] after its onset. Arctic coastal erosion increases more rapidly in response to increasing global mean surface air temperature (SAT) in the future scenarios than it does in the historical period. Before the mid 1970s, neither global nor Arctic- mean SAT decadal trends are consistently significantly positive yet (Fig. 4a). During this period, the correlation between the Arctic- mean erosion rate and the Arctic- mean SAT is weak ( \(r = 0.26 \pm 0.29\) , mean \(\pm 2\sigma\) range, Fig. 4b). However, after the 1970s, correlations between erosion and Arctic SAT increase substantially (SSP2- 4.5: \(r = 0.68 \pm 0.18\) , SSP5- 8.5: \(r = 0.93 \pm 0.06\) , 2081- 2100 means), driven by the concurrent increasing trends. This turning point is also marked by the AA onset, when the Arctic SAT starts increasing at a faster pace than the global SAT, i.e. the AA factor is consistently larger than 1 (Fig. 4c). Therefore, the sensitivity of erosion to global SAT reflects the sensitivity of Arctic SAT to global SAT – quantified as the AA factor – after the AA onset, given the strong correspondence between erosion and the Arctic SAT at that time (Fig. 4d). The sharp increase of erosion sensitivity and the AA factor to their maximum values in the early 2000s is a signature from the so- called "hiatus" in global warming [50]. Global mean SAT stalls between the late 1990s and the early 2010s, while the erosion drivers continue to increase (Fig. S1b,c). Sensitivity values level off in the second half of the \(21^{\text{st}}\) Century, when global mean SAT trends decelerate. End- of- century sensitivities are lowest in the SSP2- 4.5 scenario, when Arctic SAT trends decrease sharply to reach the also consistently decreasing global SAT trends, and the AA factor approaches one. In order to avoid the effect of the warming hiatus, we quantify erosion sensitivity considering the historical period until before the AA onset, and during the last 50 years in the scenario simulations.  
+
+<|ref|>text<|/ref|><|det|>[[111, 827, 905, 930]]<|/det|>
+The sensitivity of the Arctic- mean erosion rate to global mean SAT increases significantly from \(0.18 \pm 0.31 \text{ m year}^{- 1} \text{°C}^{- 1}\) on average during the historical period until 1975, to at least double (between \(0.40 \pm 0.16\) and \(0.48 \pm 0.21 \text{ m year}^{- 1} \text{°C}^{- 1}\) ) during the second half of the \(21^{\text{st}}\) Century following the SSP2- 4.5 and SSP5- 8.5 scenarios, respectively. This translates to an increase in the sensitivity of OC losses to climate warming from \(1.4 \text{ TgC year}^{- 1} \text{°C}^{- 1}\) in the historical period
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[258, 149, 800, 707]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 714, 905, 857]]<|/det|>
+<center>Figure 4: Sensitivity to climate change. a: 20-year running trends of global and Arctic mean surface air temperature (SAT). b: Correlations between Arctic-mean erosion rates and Arctic mean SAT. c: The Arctic Amplification (AA) factor, expressed as regression coefficients of Arctic SAT changes on global SAT. The AA onset is defined when the AA factor is larger than 1. d: Sensitivity of Arctic-mean erosion rates to climate, expressed regression coefficients on global SAT. Running-window lengths are 20 years in all plots. Different window lengths show qualitatively similar results (not shown). The AA onset (dashed blue line) takes place in 1976, when the Arctic SAT increases at a faster pace than the global mean SAT, i.e. the AA factor is larger then 1. After the the 1970s, the AA factor is consistently significantly larger than 1, except for late 21st-century in the SSP2-4.5 scenario, when global and Arctic mean SATs deaccelerated and 20-year trends are momentarily similar. </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 66, 904, 106]]<|/det|>
+before until 1975, on average, to between 2.3 and \(2.8\mathrm{TgC}\) year \(^{- 1}\circ \mathrm{C}^{- 1}\) following the SSP2- 4.5 and SSP5- 8.5 scenarios, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[115, 110, 907, 597]]<|/det|>
+The sensitivity parameters are useful tools to assess the state of Arctic coastal erosion increase and the associated OC fluxes at intermediate states or policy- based targets of global warming. It must be noted, however, that the sensitivity parameters usually assume linear relationships between the forcing and outcome variables [51]. Similarly, in our erosion model, we assume that the linear combination of thermal and mechanical drivers of erosion provides us with first- order large- scale information on the time evolution of Arctic coastal erosion, associated with a range of uncertainties and scenarios of proportionality factors. Non- linear effects could emerge, for example, from earlier onsets of the storm season overlapping with longer- lasting positive temperatures into fall. We do not consider sea- level change in our projections. Adding sea- level change as a temporal driver of erosion would increase future erosion and the sensitivity parameters, if it increases proportionally faster than our thermal and mechanical drivers with respect to the historical period. We do not directly consider episodic water level changes due to storms, which are relevant for coastal abrasion and sediment transport. However, by using a global dynamical wave model, and integrating yearly wave exposure at the coastal- segment level, we do incorporate the effect of storms in our mechanical driver of erosion. Our erosion model, relatively simple in comparison with higher- resolution and process- based strategies [52- 57], does not intended to represent all processes, often of fine spatial scale (order of meters or less), associated with the erosion of the Arctic coast. Here, we empirically parameterize the role of the the main, first- order drivers of Arctic coastal erosion at larger- scales, compatible with the resolution and mechanisms represented in ESMs (order of tens or hundreds of kilometers). Future work on coastal erosion modelling is necessary to constrain our relatively large uncertainties. Nonetheless, our semi- empirical approach allows us to make Pan- Arctic projections of coastal erosion, its associated OC fluxes, and thus estimate the magnitude, timing and sensitivity of their increase to global warming.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 622, 263, 642]]<|/det|>
+## Conclusions  
+
+<|ref|>text<|/ref|><|det|>[[115, 655, 907, 929]]<|/det|>
+We present a semi- empirical model for coastal erosion to make \(21^{\mathrm{st}}\) - century pan- Arctic projections of erosion rates and associated organic carbon (OC) losses. Our model accounts for temporal and spatial variability of erosion, combining wave and thawing temperature exposure with ground- ice content as explanatory variables. With our approach, we are able to provide estimates of magnitude, timing and sensitivity of Arctic coastal erosion increase to climate change. The Arctic- mean erosion rate increases by a factor of between 2.2 and 2.9 from the historical period (1850- 1900) to the end of the \(21^{\mathrm{st}}\) Century following the SSP2- 4.5 and the SSP5- 8.5 scenarios, respectively. The associated pan- Arctic OC flux increases by a factor of 1.9- 2.5 at the same time, reaching up to 17.2 (9.0- 25.4, two standard- deviation range) \(\mathrm{TgC}\) year \(^{- 1}\) in the SSP5- 8.5 scenario. Our projections show that Arctic coastal erosion is very likely (at least \(90\%\) probability) to exceed its historical range of variability before end of the century, even in the intermediate- emission scenario. We estimate that the sensitivity of Arctic coastal erosion to climate also increases with time, following the Arctic amplification after its onset in the 1970s, due to the strong relationship between
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 66, 907, 277]]<|/det|>
+erosion and Arctic SAT at that time. During the second half of the century, one degree of global warming is associated with an increase of the Arctic- mean erosion by about 0.4- 0.5 m/year and 2.3- 2.8 TgC/year of associated OC carbon loss, equivalent to about 5- 8% of the present- climate OC yearly flux from the Arctic rivers into the Ocean. Arctic coastal erosion will increase more rapidly in the future in response climate change, roughly doubling in rates by 2100, and likely reaching values unseen before in the past century. Our projections allow future work to investigate the impact of Arctic coastal erosion on the permafrost- climate feedback, and the future evolution of the Arctic Ocean's ecosystems and its role as a global carbon sink. Moreover, our results should also inform policy makers on coastal conservation and socioeconomic planning at the pan- Arctic level, focusing on the sustainable future of Arctic coastal communities.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 302, 345, 322]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[92, 335, 907, 461]]<|/det|>
+D.M.N., M.D., P.O., T.I. and V.B. are funded by European Union's Horizon 2020 research and innovation programme under grant agreement number 773421 - project "Nunataryuk". D.M.N., P.P., A.B. T.I., J.B. and V.B. are funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy - EXC 2037 'CLICCS - Climate, Climatic Change, and Society' - Project Number: 390683824, contribution to the Center for Earth System Research and Sustainability (CEN) of Universität Hamburg.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 487, 352, 508]]<|/det|>
+## Competing Interests  
+
+<|ref|>text<|/ref|><|det|>[[117, 521, 468, 538]]<|/det|>
+The authors declare no competing interests.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 565, 378, 585]]<|/det|>
+## Authors' contributions  
+
+<|ref|>text<|/ref|><|det|>[[116, 598, 905, 658]]<|/det|>
+D.M.N, M.D., J.B. and V.B. conceived and designed the study. D.M.N., P.P., M.D., J.B. and V.B. designed the erosion model. D.M.N. and M.D. performed the Ocean wave simulations. All authors analyzed and discussed the results. All authors wrote and reviewed the paper.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 690, 238, 712]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 736, 171, 753]]<|/det|>
+## Data  
+
+<|ref|>text<|/ref|><|det|>[[116, 768, 360, 785]]<|/det|>
+Arctic coastal observations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 789, 905, 914]]<|/det|>
+We use the Arctic Coastal Dynamics (ACD) database [33] as our observational reference. The ACD compiles several sources of data and provides a list of variables for a total of 1314 coastal segments along the Arctic coast, including: long- term erosion mean rates, organic carbon concentration, soil bulk density, ground- ice fraction, mean elevation and length. From the 1314 segments, we take those classified as erosive and non- lithified, which excludes segments from the rocky coasts in Greenland and in the Canadian Archipelago and other segments that present stable or
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[103, 66, 905, 127]]<|/det|>
+aggrading dynamics. We also select segments containing excess ice, which excludes all the nonerosive segments from Svalbard, for example. We this work with a subset of 306 coastal segments in our analysis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[145, 152, 245, 168]]<|/det|>
+## Reanalysis  
+
+<|ref|>text<|/ref|><|det|>[[103, 173, 906, 277]]<|/det|>
+We take 2- meter air temperature and significant wave heights from ERA20C reanalysis [58] as empirical variables in our coastal erosion model. Data are taken in the same periods for which the erosion rates are provided in the ACD. The temperature and wave data have \(\sim 1.12^{\circ}\) (atmosphere) and \(1.5^{\circ}\) (waves) horizontal resolution. We assign the closest land grid cell in ERA20C from its atmospheric grid to ACD segments, and two rows of adjacent cells from the ocean grid.  
+
+<|ref|>sub_title<|/ref|><|det|>[[145, 301, 319, 318]]<|/det|>
+## Climate projections  
+
+<|ref|>text<|/ref|><|det|>[[102, 323, 907, 530]]<|/det|>
+To force our coastal erosion model, we use a 10- member ensemble of simulations from the Max Planck Institute Earth System Model (MPI- ESM) version 1.2 in its low- resolution configuration [59] performed for the Coupled Model Intercomparison Project phase 6 (CMIP6) [34]. In this configuration, the atmospheric component ECHAM6.3 has horizontal resolution of T63 ( \(\sim 200\) km), and 47 vertical levels. The oceanic component MPIOM1.6 uses the curvilinear grid GR1.5, which has mean horizontal resolution of \(\sim 150\) km and 40 vertical levels. We use the historical simulations (1850- 2014) and two future Shared Socioeconomic Pathway (SSP) scenarios for the \(21^{\mathrm{st}}\) century projections (2015- 2100), namely: the SSP2- 4.5 and the SSP5- 8.5, which represent a mid- range and a high- end emission scenario, respectively. This range of scenarios is realistic it terms of current cumulative \(\mathrm{CO}_2\) emissions [35].  
+
+<|ref|>sub_title<|/ref|><|det|>[[145, 556, 365, 572]]<|/det|>
+## Ocean wave simulations  
+
+<|ref|>text<|/ref|><|det|>[[103, 577, 906, 679]]<|/det|>
+We use the wave model WAM [60] to generate a 10- member ensemble of global waves for historical, SSP2- 4.5 and SSP5- 8.5 scenarios, forced by the MPI- ESM ensemble. In our setup, WAM has \(1^{\circ}\) grid resolution and is forced with daily sea- ice concentration (threshold of \(15\%\) to define open- water), 6- hourly 10- meter winds, and a realistic ETOPO2- based bathymetry as boundary conditions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[103, 707, 626, 727]]<|/det|>
+## Semi-empirical Arctic coastal erosion model  
+
+<|ref|>text<|/ref|><|det|>[[102, 741, 907, 929]]<|/det|>
+We present a simplified model for Arctic coastal erosion, compatible with the scales of Earth system models. Our model considers the dominant physical thermal and mechanical drivers of erosion, also referred to as thermal- abrasion (TA) and thermal- denudation (TD) [1]. The model is constrained to only simulate erosion at the presence of ground ice and at the absence of coastal sea ice. We use an empirical approach to quantify the relationship between the physical drivers, constraints and the erosion rates, by comparing the observations from the ACD with ERA20C reanalysis. The empirically estimated parameters are then applied to all coastal segments, which provides us with erosion rates in the pan- Arctic scale. Our model has yearly time resolution, and the spatial resolution follows the definitions of the ACD coastal segments.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 65, 905, 106]]<|/det|>
+The total erosion \(E(\mathrm{t},\mathrm{x})\) [m year \(^{- 1}\) ], defined in every year \(t\) and coastal segment \(x(lat,lon)\) , is given as a combination of a temporal and a spatial component.  
+
+<|ref|>equation<|/ref|><|det|>[[396, 128, 902, 150]]<|/det|>
+\[E(x,t) = \overline{{E}} (t) + \Delta E(x,t) \quad (1)\]  
+
+<|ref|>text<|/ref|><|det|>[[91, 161, 905, 286]]<|/det|>
+The temporal component represents the temporal evolution of the Arctic- mean erosion \(\overline{{E}} (t)\) [m year \(^{- 1}\) ]. The spatial component \(\Delta E(x,t)\) [m year \(^{- 1}\) ] represents local departures from the Arctic mean at every year and coastal segment, providing spatially distributed values of erosion. Hereafter, we use "Arctic mean", denoted by the overline, to refer to means along the Arctic coast. All data associated with ACD coastal segments are weighted by segment lengths in the computation of means.  
+
+<|ref|>sub_title<|/ref|><|det|>[[147, 310, 372, 327]]<|/det|>
+## The temporal component  
+
+<|ref|>text<|/ref|><|det|>[[91, 331, 905, 370]]<|/det|>
+The temporal component of our model is a linear combination of Arctic means of the thermal and mechanical drivers of erosion.  
+
+<|ref|>equation<|/ref|><|det|>[[384, 410, 902, 432]]<|/det|>
+\[\overline{{E}} (t) = a_{T D}\overline{{T}} (t) + a_{T A}\overline{{H}} (t) \quad (2)\]  
+
+<|ref|>text<|/ref|><|det|>[[91, 451, 905, 536]]<|/det|>
+The thermal driver of erosion is represented by Arctic- mean yearly- accumulated daily- mean positive 2- meter air temperatures \(\overline{{T}} (\mathrm{t})\) [°C day year \(^{- 1}\) ], also commonly known as positive degrees days or thawing- degree days. The mechanical driver of erosion is represented by Arctic- mean yearly- accumulated daily significant wave heights \(\overline{{H}} (\mathrm{t})\) [m day year \(^{- 1}\) ].  
+
+<|ref|>text<|/ref|><|det|>[[91, 537, 944, 620]]<|/det|>
+We empirically estimate the linear coefficients \(a_{T A}\) [m m \(^{- 1}\) day \(^{- 1}\) year] and \(a_{T D}\) [°C m \(^{- 1}\) day \(^{- 1}\) year] by scaling the Arctic- mean physical drivers, from ERA20C reanalysis, with the observed coastal erosion rates from the ACD. This is done for the reference time \(t_{obs}\) , during which observations are available.  
+
+<|ref|>equation<|/ref|><|det|>[[416, 652, 902, 735]]<|/det|>
+\[\begin{array}{l}{a_{T A} = q\frac{\overline{{E}}_{obs}}{\overline{{H}} (t_{obs})}}\\ {a_{T D} = (1 - q)\frac{\overline{{E}}_{obs}}{\overline{{T}} (t_{obs})}} \end{array} \quad (3)\]  
+
+<|ref|>text<|/ref|><|det|>[[91, 747, 905, 831]]<|/det|>
+We assume that the thermal and mechanical drivers \(a_{T D}\overline{{T}} (t)\) and \(a_{T A}\overline{{H}} (t)\) contribute in equal proportions to the Arctic- mean erosion during the reference time. We do that by setting the proportionality factor \(q\) to 0.5. We test the sensitivity of our results to this assumption by making scenarios with \(q = 0.1\) and \(q = 0.9\) (see Table S1 and Fig. S1a in the supplementary material).  
+
+<|ref|>sub_title<|/ref|><|det|>[[147, 854, 353, 871]]<|/det|>
+## The spatial component  
+
+<|ref|>text<|/ref|><|det|>[[91, 875, 905, 914]]<|/det|>
+The spatial component of our erosion model calculates local erosion anomalies with respect to the Arctic- mean temporal evolution, and consists of two multiple linear regression (MLR) models.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[98, 66, 907, 510]]<|/det|>
+We split the coastal segments in two groups by classifying them between extreme and non- extreme with respect to erosion, using \(2.5\mathrm{m / year}\) as a threshold ( \(\sim 90th\) percentile). We do not find a distinct separation between extreme and non- extreme segments in terms of geographical location (Fig. 2a), neither in terms of coastal morphology. Both groups show similar distributions of ground- ice content, mean cliff height, bathymetric profile, bulk density, as well as mean thermal and mechanical forcings derived from thawing temperature and ocean waves, for example (not shown). We test a comprehensive number of combinations of dynamical and geomorphological parameters as explanatory variables in MLR models, simultaneously maximizing goodness- of- fit and penalizing model complexity (Table S3). We fit MLR models using the usual Ordinary Least Square (OLS) method. The goodness- of- fit of models is assessed with the proportion of explained variance and root- mean squared error (RMSE). Since increasing the number of combined explanatory variables necessarily increases the model fit and may lead to overfitting, we penalize model complexity by assessing the changes in the Akaike Information Criterion (AIC) in parallel. The best performing combination of covariates is the one which maximizes correlation (or proportion of explained variance) and minimizes RMSE and AIC (Fig. S2). We train the spatial component of our erosion model only on those segments classified as "high quality" with respect to erosion data. We include medium- quality segments to train the model for the high- erosion case to increase our sample size and thus also statistical robustness. We validate each combination of regression coefficients with unseen data by performing a leave- one- out cross validation test. We use a Bootstrap approach with 10 thousand sampling iterations to obtain distributions of model coefficient estimates, and thus their associated uncertainties.  
+
+<|ref|>text<|/ref|><|det|>[[98, 514, 907, 767]]<|/det|>
+Three variables compose the best performing combinations: a) daily- mean thawing temperature exposure, expressed as the yearly- accumulated daily positive temperature divided by the number of positive- temperature days per year \(T_{day}[\mathrm{^{\circ}C}\) year \(^{- 1}\) ], b) daily- mean wave exposure, expressed as the yearly- accumulated daily significant wave heights divided by the number of open- water days per year \(H_{day}[\mathrm{m}\) year \(^{- 1}\) ], and c) ground- ice content \(\theta [\%\) of soil volume]. On one hand, combining ground- ice content with daily- mean wave exposure \((\theta +H_{day})\) explains about \(47\%\) of the observed spatial variance among non- extreme (2.5 m/year threshold) erosion segments \((r = 0.69\) \(9 - 95^{\mathrm{th}}\) - percentile range: \(r = 0.60 - 0.78\) , Fig. 2b, Fig. S3a). On the other hand, combining ground- ice content with the daily- mean thawing temperature exposure \((\theta +T_{day})\) explains about \(36\%\) of the variance among extreme- erosion segments \((r = 0.61\) \(9 - 95^{\mathrm{th}}\) - percentile range: \(r = 0.31 - 0.94\) Fig. 2c, Fig. S3a). The linear regression coefficients \(b\) obtained with the selected variable combinations are statistically significant \((p< 0.01)\)  
+
+<|ref|>equation<|/ref|><|det|>[[225, 797, 902, 852]]<|/det|>
+\[\Delta E(x,t) = \left\{ \begin{array}{ll}b_{\theta}\Delta \theta (x) + b_{H}\Delta H_{day}(x,t) & \mathrm{if} E_{obs}(x)< 2.5\mathrm{myear}^{-1}\\ b_{\theta}^{\prime}\Delta \theta (x) + b_{T}\Delta T_{day}(x,t) & \mathrm{otherwise} \end{array} \right. \quad (5)\]  
+
+<|ref|>text<|/ref|><|det|>[[98, 867, 905, 929]]<|/det|>
+Swapping the combinations and groups, that is, using \(\theta +H_{day}\) for the extreme and \(\theta +T_{day}\) for the non- extreme erosion segments, yields overall poorer fits (Fig. S3a,b) and less robust estimation of regression coefficients (Fig. S3c- e). We also test the sensitivity of these results to the choice
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[80, 65, 907, 170]]<|/det|>
+of the threshold to define extreme erosion. Allowing for an overlap between the extreme and non- extreme segments by lowering the threshold to \(2.0\mathrm{m / year}\) , for example, increases the robustness of the \(T_{day}\) regression coefficient estimate for the extreme group (Fig. S3d) by increasing the number of data points, and yields a similar fit to that of the higher threshold \((\theta +T_{day}\) in Fig. S3a,b) and also similar ground- ice coefficients \((\theta +T_{day}\) in Fig. 3Sc).  
+
+<|ref|>text<|/ref|><|det|>[[80, 174, 907, 255]]<|/det|>
+Finally, the total erosion is constrained to the open- water period, and set to zero whenever and wherever sea- ice concentration (SIC) is above \(15\%\) at the coast. Combining the temporal (Eq. 2) and spatial (Eq. 5) components into our total erosion model (Eq. 1), conditioned by open- water and the extreme- erosion threshold, our model assumes the complete form:  
+
+<|ref|>equation<|/ref|><|det|>[[112, 285, 972, 375]]<|/det|>
+\[E(x,t) = \left\{ \begin{array}{ll}a_{TD}\overline{T} (t) + a_{TA}\overline{H} (t) + \left\{ \begin{array}{ll}b_{\theta}\Delta \theta (x) + b_{H}\Delta H_{day}(x,t) & \mathrm{if} E_{obs}(x)< 2.5\mathrm{m / year}\\ b_{\theta}^{\prime}\Delta \theta (x) + b_{T}\Delta T_{day}(x,t) & \mathrm{if} E_{obs}(x)\geq 2.5 \end{array} \right. & \mathrm{if} SIC(x)< 15\% \\ 0 & \mathrm{if} SIC(x)\geq 15\% \end{array} \right. \quad (6)\]  
+
+<|ref|>sub_title<|/ref|><|det|>[[81, 402, 295, 423]]<|/det|>
+## Bias correction  
+
+<|ref|>text<|/ref|><|det|>[[80, 436, 907, 540]]<|/det|>
+Before forcing the erosion model with MPI- ESM data, we adjust the historical and scenario simulations for climate biases. The bias is removed between ERA20C data (used to estimate our model parameters) and MPI- ESM ensemble means at the coastal segments and reference periods from observations. The modelled distributions are shifted and scaled, so that their means and spread fit those of ERA20C at the reference time.  
+
+<|ref|>sub_title<|/ref|><|det|>[[81, 566, 374, 587]]<|/det|>
+## Organic carbon fluxes  
+
+<|ref|>text<|/ref|><|det|>[[80, 599, 906, 662]]<|/det|>
+We translate linear erosion rates into volumetric erosion rates \(\mathrm{E}_{vol}[\mathrm{m}^3 \mathrm{year}^{- 1}]\) , sediment fluxes \(\mathrm{S}[\mathrm{Kg} \mathrm{year}^{- 1}]\) , and carbon fluxes \(C_{flux}\) [Kg year \(^{- 1}\) ], considering the mean geometry and ground properties of each coastal segment.  
+
+<|ref|>equation<|/ref|><|det|>[[345, 696, 903, 767]]<|/det|>
+\[\begin{array}{c}{E_{vol}(x,t) = E(x,t)L(x)h(x)}\\ {S(x,t) = E_{vol}(x,t)(1 - \theta (x))\rho (x)}\\ {C_{flux}(x,t) = S(x,t)C_{conc}(x)} \end{array} \quad (7)\]  
+
+<|ref|>text<|/ref|><|det|>[[80, 782, 906, 844]]<|/det|>
+where \(L\) and \(h\) are the segments' mean length and elevation [m], \(\theta\) is the ground- ice content \([\% \mathrm{volume}]\) , \(\rho\) is the soil bulk density [Kg/m \(^3\) ], and \(C_{conc}\) is the organic carbon concentration \([\% \mathrm{weight}]\) . We integrate over the coastal segments:  
+
+<|ref|>equation<|/ref|><|det|>[[400, 864, 903, 897]]<|/det|>
+\[\overline{C}_{flux}(t) = \sum_{x}C_{flux}(x,t) \quad (8)\]  
+
+<|ref|>text<|/ref|><|det|>[[80, 904, 355, 920]]<|/det|>
+to obtain the total Arctic flux.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 63, 450, 85]]<|/det|>
+## Sensitivity to climate change  
+
+<|ref|>text<|/ref|><|det|>[[115, 96, 907, 393]]<|/det|>
+We estimate the sensitivity of the organic carbon release by Arctic coastal erosion to climate change following the approach of Friedlingstein et al. (2006) [51]; however, with a simplified set of tools. In their work, Friedlingstein et al. compare pairs of "coupled" and "uncoupled" simulations, where the increasing atmospheric \(\mathrm{CO_2}\) concentration either affects climate, or is neutral in terms of radiative effect. This pairwise comparison is necessary because the land- atmosphere and ocean- atmosphere carbon fluxes respond to changes in both climate and atmospheric \(\mathrm{CO_2}\) concentrations. Therefore, the difference between their coupled and uncoupled simulations provide the isolated effect of the \(\mathrm{CO_2}\) - induced changes in climate on carbon fluxes from the effect of the changing atmospheric \(\mathrm{CO_2}\) concentration. In our case, changes in atmospheric \(\mathrm{CO_2}\) alone do not induce any Arctic coastal erosion response, if not by its radiative effect. An uncoupled simulation, where \(\mathrm{CO_2}\) does not induce a change in climate, would not yield any change in the organic carbon released by Arctic coastal erosion. Therefore, we can estimate the sensitivity of the organic carbon release by Arctic coastal erosion to climate \(\gamma\) [TgC year \(^{- 1}\mathrm{^\circ C^{- 1}}\) ] by comparing changes in global mean surface temperature and the resulting changes in carbon fluxes from erosion.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 417, 816, 440]]<|/det|>
+## Probability and onset of emergence from the historical range  
+
+<|ref|>text<|/ref|><|det|>[[115, 450, 907, 620]]<|/det|>
+We define the yearly probability density distribution of a modelled variable \(\psi\) as the normal distribution \(N(t)\) at year \(t\) . The mean of \(N(t)\) is the ensemble mean and its standard deviation is the ensemble standard deviation (plus the standard deviation of the distribution of erosion model uncertainties in specific situations, made clear in the text). Similarly, the historical range of a modelled variable \(\psi\) is the normal distribution fitted to its average over the period 1850- 1950 \(N_{hist}\) . We calculate the area of distributions \(A_{hist} = \int N_{hist}d\psi\) and \(A(t) = \int N(t)d\psi\) to determine their overlap \(A_{hist}\cap A(t)\) . We define the probability of emergence from the historical range \(P(t)\) , i.e. the probability that \(N(t)\) be different from \(N_{hist}\) , as the fraction of \(A(t)\) that emerges from \(A_{hist}\) :  
+
+<|ref|>equation<|/ref|><|det|>[[348, 632, 902, 671]]<|/det|>
+\[P(t) = \frac{A(t) - A_{hist}\cap A(t)}{A(t)}\times 100[\% ] \quad (9)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 677, 905, 716]]<|/det|>
+We define the onset of emergence as the year when the ensemble mean is larger than \(\mu + 2\sigma\) from historical range \(N_{hist}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 742, 430, 763]]<|/det|>
+## Estimation of uncertainties  
+
+<|ref|>text<|/ref|><|det|>[[115, 776, 907, 879]]<|/det|>
+All ranges of uncertainties, except when clearly stated otherwise, are calculated with a Bootstrap method, which suits cases where the number of data is relatively small. From any vector \(\mathbf{X}\) of arbitrary length, a large number (i.e. 10 thousand) of vectors \(\mathbf{X}^{i}\) \((i = 1, 2, \dots 10\mathrm{k})\) is generated by sampling with replacement from \(\mathbf{X}\) . The uncertainty of any statistics of \(\mathbf{X}\) is estimated from the distribution of \(i\) realizations of the statistics obtained from \(\mathbf{X}^{i}\) .
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 63, 275, 86]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[110, 100, 909, 140]]<|/det|>
+[1] Aré, F. E. "Thermal abrasion of sea coasts (part I)". In: Polar Geography and Geology 12.1 (1988), pp. 1- 1. DOI: 10.1080/10889378809377343.  
+
+<|ref|>text<|/ref|><|det|>[[110, 147, 909, 185]]<|/det|>
+[2] Overeem, Irina et al. "Sea ice loss enhances wave action at the Arctic coast". In: Geophysical Research Letters 38.17 (2011). DOI: https://doi.org/10.1029/2011GL048681.  
+
+<|ref|>text<|/ref|><|det|>[[110, 192, 912, 250]]<|/det|>
+[3] Casas- Prat, Mercè and Wang, Xiaolan L. "Sea Ice Retreat Contributes to Projected Increases in Extreme Arctic Ocean Surface Waves". In: Geophysical Research Letters 47.15 (2020), e2020GL088100. DOI: https://doi.org/10.1029/2020GL088100.  
+
+<|ref|>text<|/ref|><|det|>[[110, 257, 907, 315]]<|/det|>
+[4] Barnhart, K. R., Overeem, I., and Anderson, R. S. "The effect of changing sea ice on the physical vulnerability of Arctic coasts". In: Cryosphere 8.5 (2014), pp. 1777- 1799. DOI: 10.5194/tc- 8- 1777- 2014.  
+
+<|ref|>text<|/ref|><|det|>[[110, 322, 907, 360]]<|/det|>
+[5] Barnhart, Katherine R. et al. "Mapping the future expansion of Arctic open water". In: Nature Climate Change 6.3 (2016), pp. 280- 285. DOI: 10.1038/nclimate2848.  
+
+<|ref|>text<|/ref|><|det|>[[110, 367, 905, 405]]<|/det|>
+[6] Jones, B M et al. "Increase in the rate and uniformity of coastline erosion in Arctic Alaska". In: Geophysical Research Letters 36.3 (2009). DOI: 10.1029/2008GL036205.  
+
+<|ref|>text<|/ref|><|det|>[[110, 412, 905, 470]]<|/det|>
+[7] Jones, Benjamin M et al. "A decade of remotely sensed observations highlight complex processes linked to coastal permafrost bluff erosion in the Arctic". In: Environmental Research Letters 13.11 (2018), p. 115001. DOI: 10.1088/1748- 9326/aaef471.  
+
+<|ref|>text<|/ref|><|det|>[[110, 477, 907, 535]]<|/det|>
+[8] Günther, F et al. "Observing Muostakh disappear: Permafrost thaw subsidence and erosion of a ground- ice- rich Island in response to arctic summer warming and sea ice reduction". In: Cryosphere 9.1 (2015), pp. 151- 178. DOI: 10.5194/tc- 9- 151- 2015.  
+
+<|ref|>text<|/ref|><|det|>[[110, 542, 905, 600]]<|/det|>
+[9] Irrgang, Anna M. et al. "Variability in Rates of Coastal Change Along the Yukon Coast, 1951 to 2015". In: Journal of Geophysical Research: Earth Surface 123.4 (2018), pp. 779- 800. DOI: https://doi.org/10.1002/2017JF004326.  
+
+<|ref|>text<|/ref|><|det|>[[110, 607, 905, 645]]<|/det|>
+[10] Jones B. M. Irrgang A. M., Farquharson A. M. et al. "Coastal Permafrost Erosion". In: Arctic Report Card 2020. Ed. by Frey K. E. et al. 2020. DOI: https://doi.org/10.25923/e47w-dw52.  
+
+<|ref|>text<|/ref|><|det|>[[110, 652, 905, 691]]<|/det|>
+[11] Stroeve, Julienne and Notz, Dirk. "Changing state of Arctic sea ice across all seasons". In: Environmental Research Letters 13.10 (2018), p. 103001. DOI: 10.1088/1748- 9326/aaed56.  
+
+<|ref|>text<|/ref|><|det|>[[110, 698, 905, 736]]<|/det|>
+[12] Serreze, MC et al. "The emergence of surface- based Arctic amplification". In: The Cryosphere 3.1 (2009), p. 11. DOI: https://doi.org/10.5194/tc- 3- 11- 2009.  
+
+<|ref|>text<|/ref|><|det|>[[110, 743, 905, 781]]<|/det|>
+[13] Cohen, J et al. "Recent Arctic amplification and extreme mid- latitude weather". In: Nature Geoscience 7.9 (2014), pp. 627- 637. DOI: 10.1038/ngeo2234.  
+
+<|ref|>text<|/ref|><|det|>[[110, 788, 905, 826]]<|/det|>
+[14] Biskaborn, Boris K. et al. "Permafrost is warming at a global scale". In: Nature Communications 10.1 (Dec. 2019), p. 264. DOI: 10.1038/s41467- 018- 08240- 4.  
+
+<|ref|>text<|/ref|><|det|>[[110, 833, 905, 890]]<|/det|>
+[15] Landrum, Laura and Holland, Marika M. "Extremes become routine in an emerging new Arctic". In: Nature Climate Change (2020), pp. 1- 8. DOI: https://doi.org/10.1038/s41558- 020- 0892- z.  
+
+<|ref|>text<|/ref|><|det|>[[110, 898, 905, 935]]<|/det|>
+[16] Notz, Dirk and Community, SIMIP. "Arctic Sea Ice in CMIP6". In: Geophysical Research Letters 47.10 (2020), e2019GL086749. DOI: https://doi.org/10.1029/2019GL086749.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[78, 68, 910, 107]]<|/det|>
+[17] Dobrynin, M., Murawsky, J., and Yang, S. "Evolution of the global wind wave climate in CMIP5 experiments". In: Geophysical Research Letters 39.17 (Sept. 2012). DOI: 10.1029/2012GL052843. 
+
+<|ref|>text<|/ref|><|det|>[[78, 115, 905, 150]]<|/det|>
+[18] Dobrynin, Mikhail et al. "Detection and attribution of climate change signal in ocean wind waves". In: Journal of Climate 28.4 (2015), pp. 1578-1591. DOI: 10.1175/JCLI-D-13-00664.1. 
+
+<|ref|>text<|/ref|><|det|>[[78, 161, 905, 214]]<|/det|>
+[19] Casas-Prat, Mercè and Wang, Xiaolan L. "Projections of Extreme Ocean Waves in the Arctic and Potential Implications for Coastal Inundation and Erosion". In: Journal of Geophysical Research: Oceans 125.8 (2020), e2019JC015745. DOI: https://doi.org/10.1029/2019JC015745. 
+
+<|ref|>text<|/ref|><|det|>[[78, 226, 905, 260]]<|/det|>
+[20] Schaefer, Kevin et al. "The impact of the permafrost carbon feedback on global climate". In: Environmental Research Letters 9.8 (2014), p. 085003. DOI: 10.1088/1748-9326/9/8/085003. 
+
+<|ref|>text<|/ref|><|det|>[[78, 272, 905, 306]]<|/det|>
+[21] Schuur, Edward AG et al. "Climate change and the permafrost carbon feedback". In: Nature 520.7546 (2015), pp. 171-179. DOI: https://doi.org/10.1038/nature14338. 
+
+<|ref|>text<|/ref|><|det|>[[78, 318, 905, 352]]<|/det|>
+[22] Tanski, George et al. "Rapid CO 2 release from eroding permafrost in seawater". In: Geophysical Research Letters 46 (2019), pp. 11244-11252. DOI: 10.1029/2019GL084303. 
+
+<|ref|>text<|/ref|><|det|>[[78, 364, 905, 417]]<|/det|>
+[23] Vonk, Jorien E et al. "Activation of old carbon by erosion of coastal and subsea permafrost in Arctic Siberia". In: Nature 489.7414 (2012), pp. 137-140. DOI: https://doi.org/10.1038/nature11392. 
+
+<|ref|>text<|/ref|><|det|>[[78, 429, 905, 482]]<|/det|>
+[24] Wegner, Carolyn et al. "Variability in transport of terrigenous material on the shelves and the deep Arctic Ocean during the Holocene". In: Polar Research 34.1 (2015). DOI: 10.3402/polar.v34.24964. 
+
+<|ref|>text<|/ref|><|det|>[[78, 495, 905, 548]]<|/det|>
+[25] Terhaar, Jens et al. "Around one third of current Arctic Ocean primary production sustained by rivers and coastal erosion". In: Nature Communications 12.169 (2021). DOI: https://doi.org/10.1038/s41467-020-20470-z. 
+
+<|ref|>text<|/ref|><|det|>[[78, 560, 905, 614]]<|/det|>
+[26] Koven, Charles D, Riley, William J, and Stern, Alex. "Analysis of permafrost thermal dynamics and response to climate change in the CMIP5 Earth System Models". In: Journal of Climate 26.6 (2013), pp. 1877-1900. DOI: https://doi.org/10.1175/JCLI-D-12-00228.1. 
+
+<|ref|>text<|/ref|><|det|>[[78, 626, 905, 679]]<|/det|>
+[27] Burke, E. J., Zhang, Y., and Krinner, G. "Evaluating permafrost physics in the Coupled Model Intercomparison Project 6 (CMIP6) models and their sensitivity to climate change". In: The Cryosphere 14.9 (2020), pp. 3155-3174. DOI: 10.5194/tc-14-3155-2020. 
+
+<|ref|>text<|/ref|><|det|>[[78, 691, 905, 725]]<|/det|>
+[28] Turetsky, Merritt R et al. "Carbon release through abrupt permafrost thaw". In: Nature Geoscience 13.2 (2020), pp. 138-143. DOI: https://doi.org/10.1038/s41561-019-0526-0. 
+
+<|ref|>text<|/ref|><|det|>[[78, 737, 905, 772]]<|/det|>
+[29] Pihl, Erik et al. "10 New Insights in Climate Science 2020 - a Horizon Scan". In: Global Sustainability 4 (2021), pp. 1-65. DOI: 10.1017/sus.2021.2. 
+
+<|ref|>text<|/ref|><|det|>[[78, 784, 905, 818]]<|/det|>
+[30] Fritz, Michael, Vonk, Jorien E., and Lantuit, Hugues. "Collapsing Arctic coastlines". In: Nature Climate Change 7.1 (2017), pp. 6-7. DOI: 10.1038/nclimate3188. 
+
+<|ref|>text<|/ref|><|det|>[[78, 830, 905, 864]]<|/det|>
+[31] Turetsky, Merritt R et al. Permafrost collapse is accelerating carbon release. 2019. DOI: https://doi.org/10.1038/d41586-019-01313-4. 
+
+<|ref|>text<|/ref|><|det|>[[78, 876, 945, 929]]<|/det|>
+[32] Nielsen, David Marcolino et al. "Coastal Erosion Variability at the Southern Laptev Sea Linked to Winter Sea Ice and the Arctic Oscillation". In: Geophysical Research Letters 47.5 (2020), e2019GL086876. DOI: 10.1029/2019GL086876.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[80, 68, 905, 120]]<|/det|>
+[33] Lantuit, Hugues et al. "The Arctic Coastal Dynamics Database: A New Classification Scheme and Statistics on Arctic Permafrost Coastlines". In: Estuaries and Coasts 35.2 (2012), pp. 383- 400. DOI: 10.1007/s12237- 010- 9362- 6.  
+
+<|ref|>text<|/ref|><|det|>[[80, 130, 905, 185]]<|/det|>
+[34] O'Neill, B. C. et al. "The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6". In: Geoscientific Model Development 9.9 (2016), pp. 3461- 3482. DOI: 10.5194/gmd- 9- 3461- 2016.  
+
+<|ref|>text<|/ref|><|det|>[[80, 181, 905, 236]]<|/det|>
+[35] Schwalm, Christopher R., Glendon, Spencer, and Duffy, Philip B. "RCP8.5 tracks cumulative \(\mathrm{CO_2}\) emissions". In: Proceedings of the National Academy of Sciences 117.33 (2020), pp. 19656- 19657. DOI: 10.1073/pnas.2007117117.  
+
+<|ref|>text<|/ref|><|det|>[[80, 243, 905, 298]]<|/det|>
+[36] Lantuit, Hugues et al. "Coastal erosion dynamics on the permafrost- dominated Bykovsky Peninsula, north Siberia, 1951- 2006". In: Polar Research 30. SUPPL.1 (2011). DOI: 10.3402/polar.v30i0. 7341.  
+
+<|ref|>text<|/ref|><|det|>[[80, 307, 905, 362]]<|/det|>
+[37] Lantz, Trevor C. and Kokelj, Steven V. "Increasing rates of retrogressive thaw slump activity in the Mackenzie Delta region, N.W.T., Canada". In: Geophysical Research Letters 35.6 (2008). DOI: https://doi.org/10.1029/2007GL032433.  
+
+<|ref|>text<|/ref|><|det|>[[80, 371, 905, 427]]<|/det|>
+[38] Lantuit, H. and Pollard, W.H. "Fifty years of coastal erosion and retrogressive thaw slump activity on Herschel Island, southern Beaufort Sea, Yukon Territory, Canada". In: Geomorphology 95.1 (2008), pp. 84- 102. DOI: https://doi.org/10.1016/j.geomorph.2006.07.040.  
+
+<|ref|>text<|/ref|><|det|>[[80, 435, 905, 491]]<|/det|>
+[39] Baranskaya, Alisa et al. "The Role of Thermal Denudation in Erosion of Ice- Rich Permafrost Coasts in an Enclosed Bay (Gulf of Kruzenstern, Western Yamal, Russia)". In: Frontiers in Earth Science 8 (2021), p. 659. DOI: 10.3389/feart.2020.566227.  
+
+<|ref|>text<|/ref|><|det|>[[80, 500, 905, 555]]<|/det|>
+[40] Farquharson, Louise M et al. "Temporal and spatial variability in coastline response to declining sea- ice in northwest Alaska". In: Marine Geology 404 (2018), pp. 71- 83. DOI: https://doi.org/10.1016/j.margeo.2018.07.007.  
+
+<|ref|>text<|/ref|><|det|>[[80, 564, 905, 600]]<|/det|>
+[41] Grigoriev, MN. "Coastal retreat rates at the Laptev Sea key monitoring sites". In: PANGAEA (2019). DOI: https://doi.org/10.1594/PANGAEA.905519.  
+
+<|ref|>text<|/ref|><|det|>[[80, 609, 905, 645]]<|/det|>
+[42] Fuchs, M. et al. "Carbon and nitrogen pools in thermokarst- affected permafrost landscapes in Arctic Siberia". In: Biogeosciences 15.3 (2018), pp. 953- 971. DOI: 10.5194/bg- 15- 953- 2018.  
+
+<|ref|>text<|/ref|><|det|>[[80, 654, 905, 710]]<|/det|>
+[43] Ramage, Justine L. et al. "Terrain controls on the occurrence of coastal retrogressive thaw slumps along the Yukon Coast, Canada". In: Journal of Geophysical Research: Earth Surface 122.9 (2017), pp. 1619- 1634. DOI: https://doi.org/10.1002/2017JF004231.  
+
+<|ref|>text<|/ref|><|det|>[[80, 719, 905, 775]]<|/det|>
+[44] Couture, Nicole J. et al. "Coastal Erosion of Permafrost Soils Along the Yukon Coastal Plain and Fluxes of Organic Carbon to the Canadian Beaufort Sea". In: Journal of Geophysical Research: Biogeosciences 123.2 (2018), pp. 406- 422. DOI: https://doi.org/10.1002/2017JG004166.  
+
+<|ref|>text<|/ref|><|det|>[[80, 784, 905, 820]]<|/det|>
+[45] Grotheer, H. et al. "Burial and Origin of Permafrost- Derived Carbon in the Nearshore Zone of the Southern Canadian Beaufort Sea". In: Geophysical Research Letters 47.3 (2020), e2019GL085897.  
+
+<|ref|>text<|/ref|><|det|>[[80, 829, 905, 865]]<|/det|>
+[46] Hilton, Robert G et al. "Erosion of organic carbon in the Arctic as a geological carbon dioxide sink". In: Nature 524.7563 (2015), pp. 84- 87. DOI: https://doi.org/10.1038/nature14653.  
+
+<|ref|>text<|/ref|><|det|>[[80, 874, 905, 930]]<|/det|>
+[47] Bröder, Lisa et al. "Bounding cross- shelf transport time and degradation in Siberian- Arctic land- ocean carbon transfer". In: Nature communications 9.1 (2018), pp. 1- 8. DOI: https://doi.org/10.1038/s41467- 018- 03192- 1.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[80, 66, 907, 125]]<|/det|>
+576 [48] Jong, Dirk et al. "Nearshore Zone Dynamics Determine Pathway of Organic Carbon From Eroding Permafrost Coasts". In: Geophysical Research Letters 47.15 (2020), e2020GL088561. DOI: https://doi.org/10.1029/2020GL088561.  
+
+<|ref|>text<|/ref|><|det|>[[80, 132, 905, 190]]<|/det|>
+579 [49] Tanski, George et al. "Permafrost Carbon and CO2 Pathways Differ at Contrasting Coastal Erosion Sites in the Canadian Arctic". In: Frontiers in Earth Science 9 (2021), p. 207. DOI: 10.3389/feart.2021.630493.  
+
+<|ref|>text<|/ref|><|det|>[[80, 198, 905, 255]]<|/det|>
+582 [50] Kosaka, Yu and Xie, Shang- Ping. "Recent global- warming hiatus tied to equatorial Pacific surface cooling". In: Nature 501.7467 (2013), pp. 403- 407. DOI: https://doi.org/10.1038/nature12534.  
+
+<|ref|>text<|/ref|><|det|>[[80, 263, 905, 321]]<|/det|>
+585 [51] Friedlingstein, Pierre et al. "Climate- carbon cycle feedback analysis: results from the C4MIP model intercomparison". In: Journal of climate 19.14 (2006), pp. 3337- 3353. DOI: https://doi.org/10.1175/JCLI3800.1.  
+
+<|ref|>text<|/ref|><|det|>[[80, 329, 905, 387]]<|/det|>
+588 [52] Hoque, Md. Azharul and Pollard, Wayne H. "Arctic coastal retreat through block failure". In: Canadian Geotechnical Journal 46.10 (2009), pp. 1103- 1115. DOI: https://doi.org/10.1139/T09- 058.  
+
+<|ref|>text<|/ref|><|det|>[[80, 395, 905, 453]]<|/det|>
+591 [53] Ravens, Thomas M. et al. "Process- Based Coastal Erosion Modeling for Drew Point, North Slope, Alaska". In: Journal of Waterway, Port, Coastal, and Ocean Engineering 138.2 (2011), pp. 122- 130. DOI: 10.1061/(asce)ww.1943- 5460.0000106.  
+
+<|ref|>text<|/ref|><|det|>[[80, 461, 905, 519]]<|/det|>
+594 [54] Barnhart, Katherine R. et al. "Modeling erosion of ice- rich permafrost bluffs along the Alaskan Beaufort Sea coast". In: Journal of Geophysical Research: Earth Surface 119.5 (2014), pp. 1155- 1179. DOI: 10.1002/2013JF002845.  
+
+<|ref|>text<|/ref|><|det|>[[80, 527, 905, 585]]<|/det|>
+597 [55] Thomas, Matthew A. et al. "Geometric and Material Variability Influences Stress States Relevant to Coastal Permafrost Bluff Failure". In: Frontiers in Earth Science 8 (2020), p. 143. DOI: 10.3389/feart.2020.00143.  
+
+<|ref|>text<|/ref|><|det|>[[80, 593, 905, 650]]<|/det|>
+600 [56] Frederick, Jennifer et al. "A thermo- mechanical terrestrial model of Arctic coastal erosion". In: Journal of Computational and Applied Mathematics (2021), p. 113533. DOI: https://doi.org/10.1016/j.cam.2021.113533.  
+
+<|ref|>text<|/ref|><|det|>[[80, 659, 905, 716]]<|/det|>
+603 [57] Rolph, R. et al. "ArcticBeach v1.0: A physics- based parameterization of pan- Arctic coastline erosion". In: Geoscientific Model Development Discussions 2021 (2021), pp. 1- 26. DOI: 10.5194/gmd- 2021- 28.  
+
+<|ref|>text<|/ref|><|det|>[[80, 725, 905, 762]]<|/det|>
+606 [58] Poli, Paul et al. "ERA- 20C: An atmospheric reanalysis of the twentieth century". In: Journal of Climate 29.11 (2016), pp. 4083- 4097. DOI: 10.1175/JCLI- D- 15- 0556.1.  
+
+<|ref|>text<|/ref|><|det|>[[80, 770, 905, 828]]<|/det|>
+608 [59] Mauritsen, Thorsten et al. "Developments in the MPI- M Earth System Model version 1.2 (MPI- ESM1.2) and Its Response to Increasing CO2". In: Journal of Advances in Modeling Earth Systems 11.4 (2019), pp. 998- 1038. DOI: https://doi.org/10.1029/2018MS001400.  
+
+<|ref|>text<|/ref|><|det|>[[80, 836, 905, 873]]<|/det|>
+611 [60] The WAMDI Group. "The WAM Model—A Third Generation Ocean Wave Prediction Model". In: Journal of Physical Oceanography 18.12 (1988), pp. 1775- 1810.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 433, 150]]<|/det|>
+CoastalErosionSuplement20210608. pdf
+
+<--- Page Split --->

preprint/preprint__7e4e872719150214d05666e45e752bb79babca9f83dcb6625d3b51b87c87d710/images_list.json

@@ -0,0 +1,92 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 | snCED-seq for post-fixed tissues overview. The workflow of snCED-seq for post-fixed tissues includes single nuclei isolation by CED and HED method with the snRandom-seq method used in this study. The steps from nucleus extraction to targeted sequencing are shown. In contrast to HED, the nuclei prepared with CED were morphologically intact without leakage of RNA molecular.",
+    "footnote": [],
+    "bbox": [
+      [
+        105,
+        155,
+        891,
+        528
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2 | Quality control of nuclei prepared by cryogenic enzymatic dissociation. (A, B) Image of trypan blue-stained (A) and DAPI-stained (B) nuclei isolated from FFPE mouse brain by CED before cell encapsulation, respectively. Scale bar, \\(50 \\mu \\mathrm{m}\\) ; (C) Nuclei yield at different proteinase K concentrations; (D, E) Bar plots showing the relationship between the nuclei numbers of the PFA-fixed hippocampus with the dissociation time was isolated by HED (C) and CED (D) methods; (F) Nuclear yield per gram of hippocampal tissue, using CED and mechanical homogenization, respectively; (G, H) RNA integrity number (RIN) (F) and total RNA yield of RNA extracted from nuclei of three forms of tissue; (I) Representative peak values of amplified cDNA in different groups. \\(\\mathrm{n} = 3\\) technical replicates, and bars show mean \\(\\pm \\mathrm{s.d}\\) (C-H).",
+    "footnote": [],
+    "bbox": [
+      [
+        108,
+        90,
+        890,
+        636
+      ]
+    ],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 | Validation of snCED-seq quality control data. (A-C) Number of nuclei (A), UMIs per nuclei (B) and genes (C) per nuclei detected in fresh frozen, PFA-fixed and FFPE samples; (D) Gene detection per nuclei comparison of our data (>10,000 nuclei) with mouse tissues (5795 (kidney), 4287 (liver), 6732 (heart) and 3774 (testis) nuclei) by snRandom-seq \\(^{10}\\) , mouse brain (7031) by snFFPE-seq \\(^{9}\\) and breast (5721) by snPATHO-seq \\(^{8}\\) ; (E, F) Percentage of mitochondrial (E) and ribosomal (F) genes; (G) Saturation analysis of snCED-seq based on the different samples; (H) Percentage of reads mapped to different genomic regions under different conditions; (I) Counts of different RNA biotypes detected in FFPE brains. (J) The Pearson's correlation coefficient (R) of the normalized gene expressions between technical replication samples and post-fixed/fresh samples.",
+    "footnote": [],
+    "bbox": [
+      [
+        87,
+        324,
+        910,
+        678
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 | snCED-seq distinguishes major brain-cell types. (A) Cell map of mouse hippocampus. UMAP of 150,507 single-nucleus RNA profiles from hippocampi of fresh frozen, PFA-fixed and FFPE samples. colored by cluster; (B) Dot plot of the average expressions of top two markers in each of the 21 clusters. The color bars indicate the gene",
+    "footnote": [],
+    "bbox": [
+      [
+        92,
+        535,
+        866,
+        840
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 | snCED-seq distinguishes major cell types and shows disease-cells in the 5XFAD brains. (A) Overview of the experimental strategy. (B) Cell map of mouse hippocampus in WT and AD by supervised clustering with reference \\(^{25}\\) . UMAP of 62,000 single-nucleus RNA profiles from hippocampi of 5-month-old male mice, three WT and three 5xFAD (AD); colored by cluster. (C) Heat map showing expression of specific markers in all cell types, identifying each cluster in B. Expression level (color scale) of marker genes across clusters and the percentage of cells expressing them (dot size). (D) The frequency of each cluster in every sample. (E) The percentage of cell types in AD and WT. AD1 sample was screened. (F) DEG counts for each cell type The intensity of the blue colour was proportional to entry values. (G) The odds ratios of DEGs and AD-disease genes in every cluster. The dot size expresses cells association with the AD disease.",
+    "footnote": [],
+    "bbox": [
+      [
+        93,
+        80,
+        905,
+        655
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Fig. 6 | Characterization of the DAMs in AD. (A) UMAP plot of re-clustered microglia identifying 4 sub-clusters. (B) Cell map of mouse hippocampus in WT (Left) and AD (Right). (C) Average scaled expression of the top-10 upregulated disease-specific DEGs in split by sample (Left) and cluster (Right). (D) Venn diagram of DEGs in microglia with AD disease gene sets. (E) Heat map of intersection genes expressed in Microglia sub-clusters. (F) Violin plot of genes in subcluster 3 highly expressed in AD. (G) Heat map of amyloid-related genes expression in sub-clusters. (H) DAM genes specificity high expression in Micro 2 and Micro 3. (I-J) The genes associated with disease-related function or pathway were highly expressed in AD (I) and all clusters (J, K). \\(\\mathrm{n} = 3\\) biologically independent mouse brain samples per genotype; Color scheme of heat maps shows row max and row min, which",
+    "footnote": [],
+    "bbox": [
+      [
+        96,
+        202,
+        884,
+        720
+      ]
+    ],
+    "page_idx": 19
+  }
+]

preprint/preprint__7e4e872719150214d05666e45e752bb79babca9f83dcb6625d3b51b87c87d710/preprint__7e4e872719150214d05666e45e752bb79babca9f83dcb6625d3b51b87c87d710.mmd

@@ -0,0 +1,500 @@
+
+# snCED-seq: High-fidelity cryogenic enzymatic dissociation of nuclei for single-nucleus RNA-seq of FFPE tissues  
+
+Xiangwei Zhao xwzhao@seu.edu.cn  
+
+State Key Laboratory of Bioelectronics, School of Biological Science & Medical Engineering, Southeast University  
+
+# Yunxia Guo  
+
+1 State Key Laboratory of Digital Medical Engineering, School of Biological Science & Medical Engineering, Southeast University, Nanjing 210096, China. 2 Department of Anesthesiology, The Affiliated Junjie Ma  
+
+3 CAS Key Laboratory of Computational Biology, Shanghai Institute of Nutrition and Health, University of Chinese Academy of Sciences, Chinese Academy of Sciences, Shanghai 200031, China. 4 Department  
+
+# Ruicheng Qi  
+
+CAS Key Laboratory of Computational Biology, Shanghai Institute of Nutrition and Health, University of Chinese Academy of Sciences, Chinese Academy of Sciences, Shanghai 200031, China.  
+
+# Xiaoying Ma  
+
+State Key Laboratory of Digital Medical Engineering, School of Biological Science & Medical Engineering, Southeast University, Nanjing 210096, China.  
+
+# Jitao Xu  
+
+State Key Laboratory of Digital Medical Engineering, School of Biological Science & Medical Engineering, Southeast University, Nanjing 210096, China.  
+
+# Kaiqiang Ye  
+
+State Key Laboratory of Digital Medical Engineering, School of Biological Science & Medical Engineering, Southeast University, Nanjing 210096, China.  
+
+# Yan Huang  
+
+State Key Laboratory of Digital Medical Engineering, School of Biological Science & Medical Engineering, Southeast University, Nanjing 210096, China.  
+
+# Xi Yang  
+
+State Key Laboratory of Digital Medical Engineering, School of Biological Science & Medical Engineering, Southeast University, Nanjing 210096, China.  
+
+# Guang-Zhong Wang
+
+<--- Page Split --->
+
+Shanghai Institute of Nutrition and Health, Chinese Academy of Sciences https://orcid.org/0000- 0001- 6432- 8310  
+
+## Article  
+
+Keywords: Formalin- fixed and paraffin- embedded, cryogenic enzymatic dissociation, nuclei, Alzheimer's Disease.  
+
+Posted Date: October 21st, 2024  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 5197301/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 May 2nd, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 59464- 0.
+
+<--- Page Split --->
+
+# snCED-seq: High-fidelity cryogenic enzymatic dissociation of nuclei for single-nucleus RNA-seq of FFPE tissues  
+
+Yunxia Guo \(^{1,2\dagger}\) , Junjie Ma \(^{3,4\dagger}\) , Ruicheng Qi \(^{3,\dagger}\) , Xiaoying Ma \(^{1,\dagger}\) , Jitao Xu \(^{1}\) , Kaiqiang Ye \(^{1}\) , Yan Huang \(^{1}\) , Xi Yang \(^{1}\) , Guang- zhong Wang \(^{3}\) , and Xiangwei Zhao \(^{1,\ast}\)  
+
+\(^{1}\) State Key Laboratory of Digital Medical Engineering, School of Biological Science & Medical Engineering, Southeast University, Nanjing 210096, China. \(^{2}\) Department of Anesthesiology, The Affiliated Hospital of Yangzhou University, Yangzhou University, Yangzhou, China. \(^{3}\) CAS Key Laboratory of Computational Biology, Shanghai Institute of Nutrition and Health, University of Chinese Academy of Sciences, Chinese Academy of Sciences, Shanghai 200031, China. \(^{4}\) Department of Laboratory Medicine of The First Affiliated Hospital & Liangzhu Laboratory, Zhejiang University School of Medicine, Hangzhou, China.  
+
+\*Correspondence: xwzhao@seu.edu.cn (X.Z.).  
+
+\(\dagger\) These authors contributed equally to this work.
+
+<--- Page Split --->
+
+## Abstract  
+
+Profiling cellular heterogeneity in formalin- fixed paraffin- embedded (FFPE) tissues is key to characterizing clinical specimens for biomarkers, therapeutic targets, and drug responses. Recent advancements in single- nucleus RNA sequencing (snRNA- seq) techniques tailored for FFPE tissues have demonstrated their feasibility. However, isolation of high- quality nuclei from FFPE tissue with current methods remains challenging due to RNA cross- linking. We, therefore, proposed a novel strategy for the preparation of high- fidelity nuclei from FFPE samples, cryogenic enzymatic dissociation (CED) method, and performed snRandom- seq (snCED- seq) for polyformaldehyde (PFA) - fixed and FFPE brains to verify its applicability. The method is compatible with both PFA- based and FFPE brains or other organs with less hands- on time and lower reagent costs, and produced 10 times more nuclei than the homogenate method, without secondary degradation of RNA, and maximized the retention of RNA molecules within nuclei. snCED- seq shows 1.5- 2 times gene and UMI numbers per nucleus, higher gene detection sensitivity and RNA coverage, and a minor rate of mitochondrial and ribosomal genes, compared with the nuclei from traditional method. The correlation gene expression of nucleus from the post- fixed and the frozen sample can be up to \(94\%\) , and the gene expression of our nuclei was more abundant. Moreover, we applied snCED- seq to cellular heterogeneity study of the specimen on Alzheimer's Disease (AD) to demonstrate a pilot application. Scarce Cajal Retzius cells in older mice were robustly detected in our data, and we successfully identified two subpopulations of disease- associated in astrocytes, microglia and oligodendrocytes, respectively. Meanwhile, we found that most cell types are affected at the transcriptional level by AD pathology, and there is a disease susceptibility gene set that affects these cell types similarly. Our method provides powerful nuclei for snRNA- seq studies for FFPE specimens, and even helps to reveal multi- omics information of clinical samples.  
+
+Keywords: Formalin- fixed and paraffin- embedded; cryogenic enzymatic dissociation; nuclei; Alzheimer's Disease.
+
+<--- Page Split --->
+
+## Introduction  
+
+High- throughput single- cell/ nuclei RNA sequencing (scRNA/snRNA- seq) methods have revolutionized the entire- field of biomedical research 1- 3. scRNA/snRNA- seq have been highly successful at disease mechanisms, discovering biomarkers to help stratify patients, and identifying novel therapeutic targets as well as determining the impact of drugs. However, fresh/frozen specimen procurement is not a standard clinical and diagnostic practise in most institutions, and fresh/frozen samples cannot be obtained for certain sample types. Routine formalin- fixed paraffin- embedded (FFPE) tissues are the most common archival specimens, constituting a vast and valuable patient material bank for clinical history 4. Inevitably, the irreversible modifications caused by formalin fixation on macromolecules in FFPE samples always make it challenging for molecular biology applications. The studies have made great progress in transcription profiling in FFPE samples by optimal RNA extraction methods 5, 6 or spatial in situ profiling 7. What's more, the combinations of scRNA- seq and spatial technologies have been applied to FFPE tissues [3]. Currently, three methods have been posted, snPATHO- Seq 8, snFFPE- seq 9, and snRandom- seq 10, provided optimized methods to isolate single intact nuclei from FFPE tissues to perform snRNA- seq, which demonstrates the feasibility of snRNA- Seq in FFPE tissues and unlocks a dimension of these hard- to- use samples. Accurate transcriptomic characterization of each cell in clinical FFPE specimens is believed to provide a better understanding of cell heterogeneity and population dynamics, thereby improving accurate diagnosis, treatment, and prognosis of human disease. With the development of snRNA- seq techniques for FFPE samples, there is growing interest in the use of the vast archives of samples for diagnostic purposes.  
+
+The application of snRNA- seq in FFPE samples is premised on obtaining superior nuclei. However, isolation of intact and high- quality nuclei remains challenging due to RNA crosslinking, modification, and degradation caused by formaldehyde fixation. The strategies of nuclei preparation for FFPE tissues are longstanding and can date back to the last century, but previous applications are only limited to DNA content 11, fluorescence in situ hybridization (FISH) 12, 13, genome- wide association studies (GWAS) 14, 15, and chromatin accessibility profiling 16. Specifically, nuclei were dissociated by hyperthermia of biological tissue sections in protease solution, a technique that is sensitive to heating time and easily destroys the nuclear membrane, resulting in the loss of nuclear morphology. Moreover, prolonged exposure to enzyme buffers may increase the permeability of the nuclear membrane, resulting in RNA molecule leakage and adversely affecting snRNA- seq experiments conducted in droplets. The current state- of- the- art snRNA- seq for FFPE samples uses a mechanical homogenization method that is suitable for frozen samples before, and combined with a hyperthermic enzyme dissociation approach for nuclei preparation 8- 10. However, the homogenization of formaldehyde- fixed tissue poses challenges, leading to the presence of debris in the resulting nuclei suspension, which necessitates multiple filtration steps. This, in turn, affects the yield of nuclei and may result in the loss of smaller nuclei. However, the presence of tissue debris remains a challenge, introducing a higher amount of ribosomal RNA (rRNA), which can affect sequencing data quality. Therefore, the
+
+<--- Page Split --->
+
+acquisition of high- quality nuclei from PFA- fixed or FFPE samples will be an important basis for transcriptome study of clinical samples.  
+
+To overcome these challenges, we proposed CED method, an efficient and high- fidelity method to extract nuclei from FFPE tissues, and combine it with full- length and total RNA snRNA sequencing for post- fixed brains. We reported side- by- side comparison of nuclei between CED and conventional methods, as well as among fresh frozen, PFA- fixed and FFPE tissues, to validate the robustness of snCED- seq in FFPE samples. Although we have optimized the conventional method, the nuclei obtained by CED method outperformed those reported method in terms of RNA integrity, nuclei numbers, number of gene and UMI per nuclei and richness of gene expression. Next, we used snCED- seq to more than 60,000 single nuclei from AD hippocampus, and our resource provides a comprehensive cellular heterogeneity analysis of AD mice by characterizing the total transcriptome at single- cell resolution. We provided evidence to show that snCED- seq is a reliable platform for analysis the transcriptomic profiles of FFPE tissues of degenerative neurological diseases, and we believe that it also has the potential to unlock the largely untapped other archives of biological material found in pathology archives, paving the way for clinical applications.  
+
+## Results  
+
+## Overview of the cryogenic enzymatic dissociation of nuclei for post-fixed tissues  
+
+The acquisition of high- fidelity nuclei is a prerequisite for the research and application of snRNA- seq for FFPE samples, and also a key factor for its full mining of transcriptional information. Since the last century, nuclear preparation of FFPE tissues requires enzymatic dissociation at high temperatures (HED) worldwide and are limited to non- transcriptomic applications. We converted the idea of traditional protocols of preparing nuclei from FFPE tissues, the factors that we deemed pertinent to affect transcriptome analysis, such as dissociation temperature, reagent and time. We established a completely new method of nucleus preparation for post- fixed (paraformaldehyde fixed (PFA- fixed) and FFPE) tissues, cryogenic enzymatic dissociation (CED) strategy. For this method, sarcosyl was used instead of SDS (Sodium n- Dodecyl Sulfate) or Triton X- 100 as an anionic surfactant to participate in the nuclei preparation, which was more friendly to the nuclear membrane than the cell membrane, and became the preferred component for nucleus isolation in CED method. Moreover, proteinase K (PK) was used to digest proteins of tissue to minimize background contamination. Our CED method eliminates the need for ultracentrifugation through a sucrose cushion, nor any filtration procedure, maximizing product retention, increasing nucleation rates, and free of nuclear membranes and cytoplasmic contamination. Most importantly, the entire nucleus preparation process was carried out at low temperature, which protects the nuclear membrane and maximally retains the RNA molecules within nucleus, providing high- fidelity nucleus for snRNA- seq research of FFPE samples. In addition, by adjusting the experimental parameters, CED method is not only suitable for tissue slides, but also has good compatibility with FFPE blocks, which is more in line with the application needs of
+
+<--- Page Split --->
+
+snRNA- seq in disease research. Next, the full- length total RNAs within nuclei of frozen, PFA- fixed and FFPE brains were captured by random primers for snRNA- seq (snCED- seq), and the main workflow of snCED- seq was shown in Fig. 1.  
+
+![](images/Figure_1.jpg)
+
+<center>Fig. 1 | snCED-seq for post-fixed tissues overview. The workflow of snCED-seq for post-fixed tissues includes single nuclei isolation by CED and HED method with the snRandom-seq method used in this study. The steps from nucleus extraction to targeted sequencing are shown. In contrast to HED, the nuclei prepared with CED were morphologically intact without leakage of RNA molecular. </center>  
+
+The nuclei drived from FFPE brains prepared by CED method had intact morphology, good dispersion, high purity without agglomeration (Fig. 2A). Confirmation of the integrity and dispersion of nuclear morphology was also verified using epifluorescence microscopy (Fig. 2B). Representative images of nuclei isolated from the hippocampus of three biological replicates showed much less debris and a size- distributions were centered around 6- 8 \(\mu \mathrm{m}\) (Extended Data Fig. 1A), slightly smaller than normal frozen brain nuclei \(^{17}\) , presumably due to the tissue being fixed. Perhaps, CED method without cumbersome filtering procedures, tiny nuclei could be preserved. Statistics showed that at least a million levels of nuclei were obtained from each pair of hippocampi (Extended Data Fig. 1A, bottom). The recent snRNA- seq techniques for FFPE tissues based on random primer capture \(^{10}\) or gene probe capture \(^{8}\) require the input of nearly one million nuclei to ensuring the output of about 10,000 nuclei. Our CED method strategy can effectively circumvent the shortcomings of the current two mainstream nuclear preparation strategies, and can export nuclei stably without introducing more impurities and destroying the nuclear membrane.
+
+<--- Page Split --->
+
+We optimized the proteinase K (PK) concentration in the nuclear dissociation system as measured by morphology and count of nuclei, and found that the optimal concentration for HED was \(0.4 \mathrm{mg / mL}\) , while higher was required in the CED for mouse brain, which was due to the reduced enzyme activity at low temperature (Fig. 2C). The nuclei count gradually decreased with the extension of dissociation time at \(37^{\circ} \mathrm{C}\) (Fig. 2D), but was not observed in our method (Fig. 2E) and with the intact morphology throughout. Since the release and disappearance of nuclei occur simultaneously during enzyme dissociation, the traditional HED method will force the preferentially obtained nuclei to digest in enzyme solution, or damage the nuclear membrane, affecting nuclear yield, and this method was very sensitive to reaction time, increasing the burden on the experimenter. In addition, the CED method could obtain more than 100,000 nuclei per gram of hippocampal tissue, which was more than 10 times that of commercially nuclear extraction kits based on mechanical methods (Fig. 2F and Extended Data Fig. 2I). Finally, given the clinical demand for snRNA- seq in a variety of organs, we also dissociated the nuclei of multiple organs (Soybean size) by CED method, including heart, liver, spleen, lung, stomach, intestines, kidney, and pancreas (Extended Data Fig. 2). We observed strong applicability of this approach to multiple organs, except to the heart and lung. Especially in spleen, intestines and kidney, with the tens of millions of nuclei numbers, and despite their abundance and dense arrangement, they remained independent, intact, and unaggregated (Extended Data Fig. 2C, F and G). Despite the lower fitness of CED method in the heart and lung, outperformed the mechanical homogenization (Extended Data Fig. 2H). Moreover, conventional wisdom suggests that nuclei are not freeze- thaw friendly, forced to improve the experimenter's awareness of time control. Notably, nuclear envelope rupture and aggregation did not occur in nuclei isolated by CED even after one month of dry ice or storage at \(- 80^{\circ} \mathrm{C}\) . This property breaks the restriction that the nucleus cannot be cryopreserved.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Fig. 2 | Quality control of nuclei prepared by cryogenic enzymatic dissociation. (A, B) Image of trypan blue-stained (A) and DAPI-stained (B) nuclei isolated from FFPE mouse brain by CED before cell encapsulation, respectively. Scale bar, \(50 \mu \mathrm{m}\) ; (C) Nuclei yield at different proteinase K concentrations; (D, E) Bar plots showing the relationship between the nuclei numbers of the PFA-fixed hippocampus with the dissociation time was isolated by HED (C) and CED (D) methods; (F) Nuclear yield per gram of hippocampal tissue, using CED and mechanical homogenization, respectively; (G, H) RNA integrity number (RIN) (F) and total RNA yield of RNA extracted from nuclei of three forms of tissue; (I) Representative peak values of amplified cDNA in different groups. \(\mathrm{n} = 3\) technical replicates, and bars show mean \(\pm \mathrm{s.d}\) (C-H). </center>  
+
+## Without damage of RNA molecules in the nucleus from CED method  
+
+The morphological of the nucleus ensures the independence of single nucleus data, while the quality of RNA molecules in the nucleus can ensure the high- quality output of snRNA- seq, which is also one of most important factors affecting its application in transcriptome research. PFA fixation of cells
+
+<--- Page Split --->
+
+induces cross- linking between nucleic acids and proteins, whereas the preparation of FFPE samples requires hours of high temperature wax immersion, both result in RNA damage. How to avoid secondary damage to RNAs during nuclear preparation is crucial for snRNA- seq. We extracted RNA from nuclei to verify the harmlessness of CED method on RNA molecules. We first investigated how to extract RNA molecules from the nuclei of fixed brains. The conventional commercial RNA extraction kits are obviously not suitable for the nuclei of PFA- fixed and FFPE tissues. We combined two lysis systems which suitable for fresh or frozen tissue to extract RNA from cross- linked nuclei. The effects of different heating conditions and proteinase K concentrations on RNA integrity (RIN) and RNA yield were tested using Drop- seq buffer and commercial RNA extraction kits. We found that PFA cross- linking was effectively reversed by incubation at \(56^{\circ}\mathrm{C}\) for \(15\mathrm{min}\) in standard Drop- seq lysis buffer (Extended Data Fig. 1B), significantly shortening the heating time compared to the reported \(^{18 - 20}\) . PK has been reported to increase RNA yield \(^{18,19}\) , but our results shown that the PK concentration has little effect on RIN and RNA yield (Extended Data Fig. 1C). In addition, we performed the same experimental exploration on lysis systems of other high- throughput sequencing platforms, although comparable amount of RNA could also be obtained, RIN values were low (2- 4). This means that the standard Drop- seq lysis buffer can be used directly as lysate for FFPE nuclei at \(56^{\circ}\mathrm{C}\) .  
+
+The results showed that the CED method had almost no damage to the RNA molecules compared with the HED method. The RIN values of nuclei were basically consistent with the RIN values of PFA- fixed tissues [PFA(4T) vs PFA- fixed section] (Extended Data Fig. 1D), but far higher than that of the nuclei prepared by HED method [PFA(37T)], and even the RIN values of FFPE [FFPE(4T)] nuclei was higher than that of PFA(37T) (Fig. 2F). Then, the cDNA libraries were generated from multiple tissues to truly reflect the quality of polyA_RNA. The major peak size of cDNA for both PFA(4T) and FF(0h) was above 1200bp, while around 800bp for PFA(37T), which was even lower than that of FFPE(4T) (Fig. 2H), which again confirmed that CED method was less damaging to RNA molecules in the nucleus than the HED method. Notably, the RNA yield of nuclei isolated by CED method was consistent with that of freshly frozen (FF) samples, but much higher than that of traditional methods (Fig. 2G), which might be a key reason for limiting the application of snRNA- seq for post- fixed samples. We found that free RNA penetrates into the enzyme solution during high- temperature dissociation process, resulting in the reduction of the amount of RNA in the nucleus (Extended Data Fig. 1E), which we concluded to be a fatal shortcoming of the conventional method. It has been reported that \(3\times\) or \(5\times\) SSC proved to be a good medium for the prevention of cellular RNA degradation, but we found that that the two types of buffers had almost the same effect on RNA molecules (Extended Data Fig. 1F). In conclusion, CED method can minimize the damage to the nuclear morphology and RNA molecules of post- fixed brains.  
+
+## Validation the nuclei quality derived from CED method by snRNA-seq  
+
+We employed droplet- based snRNA- seq technology to capture total RNA (M20 Genomics) and poly(A) RNA ( \(10\times\) Genomics). snRNA- seq was performed on mouse hippocampus samples with three
+
+<--- Page Split --->
+
+treatment conditions: (1) fresh frozen tissue- the nuclei were extracted by mechanical homogenization, capture by poly(T) [FF \((10\times)]\) and random primers [FF(M20)]; (2) PFA- fixed tissue - the nuclei were prepared by HED [PFA(37T)] and CED method [and PFA(4T)] for snRandom- seq; (3) The nuclei of FFPE tissue were dissociated by CED method and capture by random primers [FFPE(4T)/FFPE]. With snRNA- seq data of frozen tissue for reference, the influence of high and low temperature dissociation on sequencing data was evaluated to verify the fidelity of nuclei obtained by the CED method, and also to evaluate the applicability of CDE method in FFPE samples. Before microfluidic encapsulation, the nuclei were imaged to confirm single nucleus morphology and counted, and the results returned that the nuclei numbers of all samples were about million level. snRandom- seq necessitate a substantial amount of input material, millions of nuclei fully meet the requirement of nuclear detection rate of 10,000. After barcoding and amplification, the fragment size of the cDNA from FF(M20) main peaked about at 700 bps (Extended Data Fig. 3A). While the main peak of cDNA from PFA(4T) between 300 and 1000bps, and longer than PFA(37T), which might be stem from RNA degradation (Extended Data Fig. 3A). In addition, the next- generation sequencing (NGS) library with equal input of cDNA showed the lowest library for PFA(37T) (Extended Data Fig. 3B). The amount of cDNA and NGS library from PFA(4T) was slightly higher than that in FFPE(4T) and FF(M20), but much higher than PFA(37T), indicating that CED method effectively blocked the leakage of nuclear RNA and almost maintained the true level RNA molecules within the nucleus (Extended Data Fig. 3B). In fact, we have optimized the HED method to work well for snRNA- seq, which has improved its investigability in transcriptomics, but the results are still unsatisfactory. In short, the nuclei prepared by the CED method from fixed or paraffin- embedded samples are more suitable for the research of snRNA- seq.  
+
+## Performance of nuclei from FFPE tissues in snRNA-seq  
+
+We identified 150,507 high- quality unique nucleus barcodes using the barcode- gene rank plot, with clearly separated of nuclei from background noise, and an average of more than 10,000 nuclei were identified in every sample (Fig. 3A). Gene and UMI count distribution showed that the total UMIs and genes in FF(M20) and PFA(4T) were significantly \((p< 0.05)\) higher than in PFA(37T) (Extended Data Fig. 4A, B), which confirmed that CED method could maximize the retention of RNA molecules. But the total gene numbers detected in all samples were comparable, all above 30,000 (Extended Data Fig. 4C). In addition, snRNA- seq captured a mean of 1835, 2725, 1347, 2013, 1847 genes and 4189, 11100, 4996, 8759, 7072 UMIs in single nucleus by sequencing average \(\sim 27\mathrm{k}\) , \(19\mathrm{k}\) , \(13\mathrm{k}\) , \(17\mathrm{k}\) , \(12\mathrm{k}\) reads per nucleus for FF(10x), FF(M20), PFA(37T), PFA(4T) and FFPE(4T) samples, respectively (Fig. 3B, C). The number of genes and UMIs in FF(M20) was higher than FF(10x), which benefit from the principle of capturing full length and transcripts by random primers. Moreover, the gene and UMI counts per nuclei in PFA(4T) were slightly lower than FF(M20), but about 1.5 to 1.75 times higher than PFA(37T), and even about 2- 2.5 times in individual samples, and their numbers in FFPE(4T) were also higher than that in FF(10x) and PFA(37T) (Fig. 3B, C). The saturation analysis showed that FF(M20) and PFA(4T) had the highest sensitivity, followed by FFPE(4T), with 4000 to 5000 detected genes per nuclei, respectively, at a sequencing depth of 30,000 trimmed reads per nuclei, and both exhibited a
+
+<--- Page Split --->
+
+higher gene detection rate than PFA(37T) (Fig. 3G). We next compared our data with other reported results (Fig. 3D and Extended Data Fig. 4D). The genes detected per nuclei in snCED-seq datasets of brains comparable with other parenchymatous organs, despite inherently lower RNA abundance in brains (Extended Data Fig. 4D). The number of genes detected in snRNA-seq reached saturation between 100k and 150k uniquely aligned reads per nuclei (Extended Data Fig. 4D). Beyond that, lower rates of mitochondrial and ribosomal genes in PFA(4T) and FFPE samples than others, almost 0, indicating that the CED nuclei were pure without cytosolic contamination (Fig. 3E, F). Unlike FF(10x), samples captured by random primers exhibited homogeneous coverage across the body of protein- coding, but with a slight bias toward the 3'- end due to the extra addition of oligo(dT) primer in reverse transcription (Extended Data Fig. 3E). However, PFA(37T) had more 3' bias, possibly related to the greater fragmentation of RNA within nuclei (Extended Data Fig. 3E).  
+
+![](images/Figure_3.jpg)
+
+<center>Fig. 3 | Validation of snCED-seq quality control data. (A-C) Number of nuclei (A), UMIs per nuclei (B) and genes (C) per nuclei detected in fresh frozen, PFA-fixed and FFPE samples; (D) Gene detection per nuclei comparison of our data (>10,000 nuclei) with mouse tissues (5795 (kidney), 4287 (liver), 6732 (heart) and 3774 (testis) nuclei) by snRandom-seq \(^{10}\) , mouse brain (7031) by snFFPE-seq \(^{9}\) and breast (5721) by snPATHO-seq \(^{8}\) ; (E, F) Percentage of mitochondrial (E) and ribosomal (F) genes; (G) Saturation analysis of snCED-seq based on the different samples; (H) Percentage of reads mapped to different genomic regions under different conditions; (I) Counts of different RNA biotypes detected in FFPE brains. (J) The Pearson's correlation coefficient (R) of the normalized gene expressions between technical replication samples and post-fixed/fresh samples. </center>  
+
+In our snRNA sequencing experiment of PFA- fixed and FFPE brains, less than \(10\%\) uniquely aligned reads were mapped to exons and intergenic regions, and with more reads mapped to introns
+
+<--- Page Split --->
+
+(Fig. 3H). In contrast, frozen samples had a higher proportion of exons (FF(M20): \(\sim 13\%\) and FF(10x): \(\sim 21\%\) ) (Fig. 3H). We suspected that the nuclei of frozen samples were prepared by the homogenization method, which was more susceptible to the cytoplasm of pollution. By comparison, higher coverage of intronic regions in the post- fixed groups, especially in PFA(4T) and FFPE(4T) (Fig. 3H), suggesting that our nuclei had little cytosolic contamination, with higher fidelity. The higher proportion of introns might lead to more accurate RNA velocity measurements across differentiation trajectories \(^{21}\) . Abroad spectrum of RNA biotypes was detected, and protein- coding genes were the most highly detected biotype across all groups, but also other biotypes (Extended Data Fig. 5A). Unexpectedly, a substantial amount of full transcripts were detected in all group samples, especially in FF(10x) (Extended Data Fig. 5A), used 10x Chromium Single Cell 3' Solution, which is consistent with our previous bulk RNA- seq analysis \(^{22}\) . Contrary to our previous knowledge, we speculated that perhaps there is a wider range of A- capped non- coding RNA molecules within the nucleus. However, at least our data show that extensive and ought to exist non- coding genes can be detected in the FFPE nuclei prepared by CED method (Fig. 3I).  
+
+Next, gene expression correlation analysis was performed on our data. To prove the repeatability of our method, duplicate samples were sequenced independently, and a high correlation (Pearson R: 0.99, \(\mathrm{p}< 2.2\mathrm{e} - 16\) ) of gene expression profiles across random batches were seen in PFA(37T), PFA(4T) and FFPE(4T) groups (Fig. 3J), indicating the robustness of nuclei from fixed/FFPE samples. We then analyzed the correlation of gene expression between fixed/FFPE and frozen samples. Consistently, the total RNA profiles of fixed/FFPE and FF(M20) samples displayed a good correlation (PearsonR : \(>0.9\) , \(\mathrm{p}< 2.2\mathrm{e} - 16\) ), more genes were underexpressed in PFA(37T) group, but not observed in PFA(4T) and FFPE(4T) samples (Fig. 3J). A poor correlation between fixed/FFPE and FF(10x) (PearsonR: \(\sim 0.7\) , \(\mathrm{p}< 2.2\mathrm{e} - 16\) ) (Extended Data Fig. 4F), and higher gene expression in fixed samples, which stem from differences in technique. In addition, the correlation between FFPE(4T) and PFA(4T) was as high as 0.99, reaching the within- group level. Compared with PFA(37T), it was only 0.95, and the gene expression was higher in FFPE(4T) samples (Extended Data Fig. 4F). These results suggest that nuclei from the CED method behave more similarly to frozen samples.  
+
+## Cell heterogeneity analysis in PFA-fixed and FFPE tissues  
+
+We next compared the cell types identified in all group samples at single- cell resolution. Unsupervised clustering of the above filtered high- quality single brain nucleus profiles, by merging the data of PFA- fixed, FFPE samples and frozen samples. By merging the data of all batch of samples, we obtained a robust cell clustering by UMAP (Uniform Manifold Approximation and Projection), and the low similarity cellular landscapes between FF and fixed/FFPE samples before batch (Extended Data Fig. 6A). Batch- based processing resulted in integrated UAMP profiles revealed over 21 distinct clusters (Fig. 4A and Extended Data Fig. 6B). All clusters could be further annotated based on classical known cell- type markers (Fig. 4B), and 11 major cell types were identified with cell- specific genes reliably mapped on the corresponding clusters (Fig. 4B and Extended Data Fig. 6C). Most of the recommended terms in mouse hippocampus samples were identified, including excitatory neuron
+
+<--- Page Split --->
+
+(Ex1- 8), inhibitory neuron (Inh1- 4), Interneuron (Inter_N), astrocytes (AST), Oligodendrocytes (Oligo), Oligodendrocyte progenitor cells (OPC), Microglial (Micro) and Cajal Retzius cells (CRC) (Fig. 4A). Besides the known cell types, we also respectively annotated choroid plexus cells (CPC) marked by \(Prlr\) , which are rarely detected in reported data (Fig. 4A). We suspect that our nuclei were more abundant and contained more cell types.  
+
+Subsequently, we analyzed the proportion of cells types across all groups. As expected, the proportion of cells differed between 3' and random primer capture techniques, mainly in AST, Oligo, and Endo cells (Fig. 4Cand Extended Data Fig. 6D). However, the similarly cell proportion between frozen and post- fixed samples in our datasets was seen (Fig. 4C). We surmise that we shortened the time of enzymatic dissociation of the samples at high temperatures, thereby retaining most of the cell types in PFA(37T) group. However, the experimentalists are required to be experienced, otherwise resulting in poor batch. Despite this, CPC cells were severely lost in PFA(37T) samples, but a considerable cells were detected in both PFA(4T) and FFPE samples (Extended Data Fig. 6D), indicating that CED detected more scarce cells than conventional methods. In addition, higher number of cell clusters was obtained at a resolution of 0.1 than other reported results \(^{23 - 25}\) . Therefore, we counted the cluster numbers under different resolution and found that absolute advantage in PFA(4T) samples, and even a higher cluster number of FFPE sample than PFA(37T) group (Extended Data Fig. 6E). But the number of clusters reached a comparable level in all samples when resolution at 1.0 (Extended Data Fig. 6E). We reconfirmed the above inference that the nucleus prepared by CED method may bring more cellular heterogeneity information and has the potential to recognize more cell types.  
+
+![](images/Figure_4.jpg)
+
+<center>Fig. 4 | snCED-seq distinguishes major brain-cell types. (A) Cell map of mouse hippocampus. UMAP of 150,507 single-nucleus RNA profiles from hippocampi of fresh frozen, PFA-fixed and FFPE samples. colored by cluster; (B) Dot plot of the average expressions of top two markers in each of the 21 clusters. The color bars indicate the gene </center>
+
+<--- Page Split --->
+
+expression level, the bluer the color, the higher the expression level. The bubble diameter indicates the proportion of expression of the gene in that cell cluster, and the larger the diameter, the stronger the specific expression. (C) Number of nuclei (Left) and proportion (Right) of annotated cell types of all samples by snCED- seq; (D) The number of differentially expressed genes (DEGs) between each comparison group; Red and blue indicate up- regulation and down- regulation, respectively.  
+
+## snCED-seq revealed cell diversity and heterogeneity in FFPE hippocampal from AD mice  
+
+To validate the promise of our nuclei for the research of brain diseases, we applied snCED- seq on the FFPE hippocampus of AD and matched wild type (WT) mice (Fig. 5A) to explore the specific- cell state changes of AD samples. After nuclei with over or under expression of genes were filtered out, snCED- seq identified 62,000 true nuclei in the FFPE brains, and with approximately zero mitochondrial and ribosomal genes in all samples (Extended Data Fig. 7A). Unsupervised clustering of the single nucleus revealed 19 distinct clusters at resolution of 0.1 (Extended Data Fig. 7B). The main cell types of AD and WT hippocampus could be identified based on the known cell- type markers, including Ex1- 6 (Hs6st3, Pdzrn3), Inh1- 4 (Gad1, Gad2), AST (Slc1a2), Oligo (Mbp, Mobp), Micro (Dock2), OPC (Vcan), Endo (Flt1, Mecom), Smooth muscle cell (SMC, Ebf1). Olfactory ensheathing glia (OEG, Bnc2) and CRC (Cdh4, Reln) with minimal number of cells were also identified (Extended Data Fig. 7C). Abundant cell types could be identified in our data, but differences in cell proportions compared with previous data from frozen hippocampal of 7- month- old mice [2], such as an increased proportion of most neuronal cells in the AD model, but AD- related cells (Ast, Micro, Oligo, OPC) and vascular- related cells (Endo, SMC) were absent (Extended Data Fig. 7F). The AST proportion gradually increased with the increase of age of AD \(^{25}\) , and a decrease in the proportion of AST of AD mice has also been reported \(^{23}\) . We speculate that it might be due to inconsistent methods of nuclear preparation or age of AD mice.  
+
+To identify cell types more accurately and precisely, a reference of the published snRNA- seq data of frozen AD hippocampal \(^{25}\) was used for supervised clustering using Approximate Nearest Neighbors Oh Yeah (Annoy) (Fig. 5B). \(90.7\%\) of the FFPE data and the reference data were predicted to be high predicted scores, and only a minority of the cells had a score less than 0.8 (Extended Data Fig. 7D, E). However, we could also infer cell types from the distribution of low- predicted cells in the UAMP map, such as Ast and ExN cells (Extended Data Fig. 8A). The gene and UMI numbers in all cell types from the high predicted score was higher, and the other quality control data was also better (Extended Data Fig. 8B). In addition, the ExN.IEGs cells from the reference data were successfully detected in our data, but in a small proportion, and overlapped with ExN.CA1.1 cells, so we included them in the ExN.CA1.1 (Extended Data Fig. 8C). Finally, we determined the atlas of supervised clustering, and identified 22 clusters covering 11 cell types, including 9 Ex cells (ExN, ExN.CA1.1- 1.3, ExN.CA3.1- 3.3, ExN.DG and ExN.sub), 4 Inh cells (GABAergic.1- 4, GABA1- 4), and 9 non- neuronal cells (Fig. 5B, 5C). We observed a disproportion of cells in the AD1 that did not conform to conventional wisdom (Fig. 5D), but the proportion of diseased cells before and after removal of the AD1 sample was barely affected (Fig. 5E and Extended Data Fig. 7G). Overall, the cell types in the
+
+<--- Page Split --->
+
+reference data were all detectable in our FFPE nuclei, with Ex cells accounted for the largest proportion (52 %), followed by Ast (21 %), Oligo (11 %), GABA (6 %), OPC (4 %) and Micro (3 %), and the other cells accounted for about 1 % respectively (Extended Data Fig. 8D).  
+
+The proportions of cells obtained by both clustering methods were similar (Extended Data Fig.7F, G). We observed that most cells coincided with unsupervised clustering by marker gene comparison, with ExN.DG corresponding to Ex.neuron6 and GABA4 corresponding to Inh.neuron2 (Fig. 5C and Extended Data Fig. 7C). However, the annotation of some cells changed. For example, we merged Ast1- 2 in the reference data into AST, and labeled SMCs and OEGs in unsupervised clustering as pericytes_ Per and Fibroblasts_Fib, respectively (Fig. 5C and Extended Data Fig. 8C). OECs are a glial cell between Schwann cells and oligo, which have the functions of neurotrophic, inhibition of gliosis, scar formation and sheath formation, and can provide a suitable microenvironment for axon growth and strong migration characteristics. It has been reported that OECs transplantation reduced amyloid burden in amyloid precursor protein transgenic mouse model 26. OECs injected into the hippocampus of AD mice can improve the learning and memory ability and increase the activity of mitochondrial cytochrome oxidase in the hippocampal CA1 region, which has an obvious therapeutic effect on AD. This is consistent with our results that OECs in AD undergo loss (Extended Data Fig. 7F). Notably, CR cells are only present in our data other than reference data (Extended Data Fig. 7D). The number of CR cells decreases with brain development, and a handful of CR cells can still be detected in the hippocampus of old mice 27. Since the dominant advantage of the CED strategy to maximize the retention of nuclei of FFPE tissues, relatively few CR cells distributed in the hippocampus were efficiently enriched, and detected by snRNA- seq. We then observed that the proportion of all glial and vascular- associated and other nonneuronal cells were reduced in AD, compared with WT (Fig. 5E). Although the proliferation of Ast and Micro cells is deemed to be the cellular changes of AD disease, but the frozen hippocampal snRNA- seq data reported by Regev, except for Micro and Endo cells, the remaining proportion changes of non- neural cells are consistent with ours 25. Also, in the cortical data, Ast was in a status of missing, Micro was the only cell type that increased in AD, and the rest were in a stable state. But in the hippocampus, Ast, OPC and vascular cells were reduced in AD, the proportion of Micro increased, and Oligo remained almost unchanged 23. The similar results further demonstrate the reliability of our data.  
+
+In brief, more abundant cell types can be detected in our data, and provide the superior nuclei for omics research in brain diseases. We next used these cells to characterize the heterogeneity of AD- disease cells and the perturbing nature of perturbation of gene expression.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Fig. 5 | snCED-seq distinguishes major cell types and shows disease-cells in the 5XFAD brains. (A) Overview of the experimental strategy. (B) Cell map of mouse hippocampus in WT and AD by supervised clustering with reference \(^{25}\) . UMAP of 62,000 single-nucleus RNA profiles from hippocampi of 5-month-old male mice, three WT and three 5xFAD (AD); colored by cluster. (C) Heat map showing expression of specific markers in all cell types, identifying each cluster in B. Expression level (color scale) of marker genes across clusters and the percentage of cells expressing them (dot size). (D) The frequency of each cluster in every sample. (E) The percentage of cell types in AD and WT. AD1 sample was screened. (F) DEG counts for each cell type The intensity of the blue colour was proportional to entry values. (G) The odds ratios of DEGs and AD-disease genes in every cluster. The dot size expresses cells association with the AD disease. </center>  
+
+## Multidimensional identification of AD disease-specific cells  
+
+To reveal AD- associated cells, we compared levels of gene expression in nuclei isolated from AD
+
+<--- Page Split --->
+
+versus WT individuals by cell type, and identified 8026 unique differentially expressed genes (DEGs) that implicated all major cell types, and \(90\%\) of DEGs were overexpressed genes (Fig. 5F). Neurons showed a strong signature of activation, \(79\%\) of DEGs in Ex (Ex.CA3 predominated) and \(96\%\) in GABA (GABA2 predominated) neurons were overexpression, whereas DEGs of non- neuronal cells were almost \(100\%\) upregulated (log2FC \(>0.25\) , \(p < 0.01\) ) (Extended Data Fig. 9A and Supplementary Table 1). Indicating that most genes in AD nuclei were in an activated state. Both up- and downregulated DEGs were highly cell type specific, \(62\%\) of DEGs in neurons, whereas DEGs in non- neuronal populations were substantially smaller, probably owing to reduced power in lower- abundance cell types \(^{28}\) . Furthermore, vascular cells (Endo and Epend) also showed no less differential changes than glial cells (Fig. 5F and Extended Data Fig. 9A). These contrasting observations on the number and dominant directionality of DEGs reveal a heterogeneous response to AD between cell types- a recurrent theme that will be observed throughout the study.  
+
+The vast majority of DEGs (50%) were perturbed only in a single cell type, which indicates that these perturbations are strongly cell- type specific (Extended Data Fig. 9B). But a thimbleful of genes was highly expressed in \(82\%\) cell types, such as Magi2, Cadm2, Grm7, Adgrl3, Ctna2, Ctnnd2, Camta1, Dgki, Drc1, Lsamp, Mbd5, Nrg3, Ppfla2 and Prkce (Supplementary Table 2). Rn18s- rs5 and Malat1 genes were under- expressed in most cells. Among them, Magi2 has been reported to be associated with AD phenotypes \(^{29}\) and is considered a potential "new" candidate locus in the etiology of divergent AD, which was involved in the regulation of protein degradation, apoptosis, neuron loss, and neurodevelopment \(^{30}\) . We speculate that these genes preferentially undergo perturb changes in expression in AD pathology, which may be therapeutic targets associated with AD disease. Overall, these results of our snRNA- seq for FFPE brains indicate that all major cell types are affected at the transcriptional level by AD pathology. Finally, we evaluated whether AD- associated variants are enriched in genomic regions with genes whose expression pattern is cell- type- specific. Fisher test enrichment scores of each cell type- specific DEGs and AD risk genes were calculated, and AD risk variants were found to be associated with genes from Micro, OPC, Ast and Oligo cells, and were also significantly \((p < 0.05)\) enriched in GABA1 and GABA2 (Fig. 5G).  
+
+The multi- dimensional analyses results showed that nuclei prepared by CED method could identify disease- specific cell types counterpart to frozen samples by snRNA- seq, and vascular cells, which have been less studied in AD, also surfaced. Next, the traditional AD- associated cells (Micor, Ast, Oligo) were used for further analysis firstly.  
+
+## Microglial heterogeneity analysis associated with AD-related traits  
+
+Using the single- cell resolution feature, we sub- clustered Micro cells of AD and WT mice, and identified four subpopulations (Micro0- 3) (Fig. 6A). The micro2 was representative cells only in 5xFAD mouse, but micro0 and micro1 were mainly distributed in WT (Fig. 6B). A scanty of 15 micro3 cells were distributed independently in the UMAP atlas, and with the 1 to 2 ratio in AD and WT (Fig. 6B), indicating that Micro cells are significantly affected by AD pathology and the disease- induced differences result in two nonoverlapping cellular states. We observed that the top 10 up- regulated
+
+<--- Page Split --->
+
+DEGs in AD were highly expressed only in Micro2 and Micro3, implying specific disease changes for these cells (Fig. 6C). Micro's DEGs overlapped with the AD disease gene set, and 11 genes were identified (Fig. 6D). These genes were mainly expressed in Micro2 and Micro3, and individual genes were up-regulated in Micro1 (Fig. 6E). Hence, we determined Micro2 is a DAM (diseases-associated microglial, DAM). Although Micro3 cells accounted for less than \(1\%\) of the total, disease genes were specifically highly expressed in them, such as the \(\beta\) - amyloid precursor protein related gene (App) (Fig. 6E). We found that the expression of these genes was higher in Micro3 of AD mice (Fig. 6F), implying that five Micro3 cells in AD were also DAM cells, indicating that our nuclei are highly cellular heterogeneous and more suitable for the application of transcriptomics in diseases.  
+
+To demonstrate the accuracy of the DAM cells we identified, we employed multi- channel data for validation. The expression of DAM genes in human cerebral cortex was first verified in our data (Fig. 6G). The encoded proton divalent cation transporter Slc11a1[203], which regulates ion homeostasis and has pleiotropic effects on proinflammatory responses, was expressed only in Micro3. The zinc efflux transporter gene Tmem163 and immune cell adaptor gene Skap1 were expressed only in Micro2, which were specific for AD brain (Fig. 6G). SKAP1 is an immune- cell adaptor that couples T- cell receptors to the "inside- out" signaling pathway of LFA- 1- mediated T- cell adhesion. Studies have reported that Skap1- deficient mice are highly resistant to collagen- induced arthritis, which is a new potential target for therapeutic intervention of autoimmune and inflammatory diseases \(^{31}\) . Thus, the high expression of Skap1 in Micro2, lose its anti- inflammatory resistance, which is promising to be a new checkpoint for studying the mechanism of AD disease. Then, DAM genes in cortical Micro of 7- month- old 5XFAD mice were also used to verify the accuracy of our DAM identification \(^{23}\) . The results showed that Axl, Lgf1, and Csfl, which were upregulated in AD cortex, were all highly expressed in AD hippocampus compared with WT, but the homeostatic genes, such as P2ry12, Crybb1, Tmem119 and Cx3cr1, were under expressed in AD (Fig. 6H, Left). However, Trem2 was under- expression in AD of our data, whereas Csfl, a gene reported Trem2- dependent upregulation, does not appear to be affected by Trem2 deficiency in the AD hippocampus. Similar to the above result, the reported DAM genes were mainly highly expressed in Micro2 and Micro3, while most of the down- regulated genes were more prominently expressed in Micro1 (Fig. 6H, Right). The expression profiles of reported disease- pathway genes associated with AD in microglia subclusters were also analyzed. The antioxidant defense system is essential for cell survival in the central nervous system, and oxidative stress dysfunction is associated with neurodegenerative diseases \(^{32}\) . Therefore, we first analyzed the genes involved in the regulatory pathway of oxidative stress- induced neuronal death. We found that these genes were highly expressed in AD mice (Fig. 6I), such as the amyloid gene (App), the ubiquitin protein ligase gene (Prkn), and the oxidation resistance 1 gene (Oxr1). Analogous results were presented again, that all the genes related to this pathway were significantly \((p < 0.05)\) highly expressed in Micro2 and Micro3 (Fig. 6J). In addition, genes involved in the B- cell receptor signaling pathway, regulation of neuronal apoptotic processes, stress- activated protein kinase signaling cascade, regulation of GTPase activity, and immune response- activation signaling pathways were all significantly overexpressed in the AD hippocampus (Fig. 6K).
+
+<--- Page Split --->
+
+Our snRNA- seq data from FFPE hippocampal nuclei prepared by CED method can identified two DAM cells, in which Micro2 was the proliferating DAM in AD, and DAM- signature gene expression was independent of Trem2 expression. Moreover, Micro3 is also affected by AD development and exists in AD independent of Micro2. In addition, the combined analysis of multiple data indicated a strong robustness of our nuclei, which will be verified several times in subsequent analyses.  
+
+![](images/Figure_6.jpg)
+
+<center>Fig. 6 | Characterization of the DAMs in AD. (A) UMAP plot of re-clustered microglia identifying 4 sub-clusters. (B) Cell map of mouse hippocampus in WT (Left) and AD (Right). (C) Average scaled expression of the top-10 upregulated disease-specific DEGs in split by sample (Left) and cluster (Right). (D) Venn diagram of DEGs in microglia with AD disease gene sets. (E) Heat map of intersection genes expressed in Microglia sub-clusters. (F) Violin plot of genes in subcluster 3 highly expressed in AD. (G) Heat map of amyloid-related genes expression in sub-clusters. (H) DAM genes specificity high expression in Micro 2 and Micro 3. (I-J) The genes associated with disease-related function or pathway were highly expressed in AD (I) and all clusters (J, K). \(\mathrm{n} = 3\) biologically independent mouse brain samples per genotype; Color scheme of heat maps shows row max and row min, which </center>
+
+<--- Page Split --->
+
+represents relative expression of each gene among AD and all sub-clusters.  
+
+## Similar disease-related transcriptional changes occur in astrocytes and oligodendrocytes  
+
+Next, we identified five unique AST transcriptomically defined clusters characterized by high expression of Luzp2 (Ast0), Rgs6 (Ast1), Kcnip4 (Ast2), Cdh4 (Ast3), and Rnf213 (Ast4) (Extended Data Fig. 10A, B). The DEGs of AST subclusters between AD and WT were evaluated, and most DEGs were upregulated in AD (Extended Data Fig. 10C). We identified DEGs that were unique to single or combinations of AST subclusters (Extended Data Fig. 11A) and evaluated these gene sets by GO analysis (Extended Data Fig. 11B). We observed that the specific DEG numbers (Extended Data Fig. 12A) and their enriched GO terms (Extended Data Fig. 11B) were largest in AST2, and the greatest change (log2FC) in specific expression (Extended Data Fig. 10C). In addition, when comparing the top- 10 up-/downregulated DEGs by cluster and disease state, few conserved/common transcriptomic changes were found across all AST subpopulations but instead found highly cluster- specific transcriptomic changes based on disease state (Extended Data Fig. 10D). The perturbation of gene expression changes in Ast2 and Ast4 was the most prominent (Extended Data Fig. 10D), and was provisionally defined as DAA (Disease- associated astrocytes). The DEGs that were significantly highly expressed were exactly the marker genes of Ast2 (Extended Data Fig. 10 B- D and 11C). Despite the overall absence of Ast cells in AD, Ast2 was highly enriched in 5xFAD mice, and Ast4 also exhibited slight cellular proliferation (Extended Data Fig. 10E), consistent with the pathological features of astrogliosis in AD. DAA genes, Kcnip4, Erc2, Nrg3, Nrxn3, and Csmd1 were notably highly expressed in Ast2 ofAD mice compared with other subclusters (Extended Data Fig. 10G). We suspected that these DAA genes were primed activated in the hippocampus, and were preferentially activated during disease induction to dominate cell state changes.  
+
+DAA cells of Gfap- high state in the report 25, its upregulated genes were highly and unique expression in Ast4, such as Gfap, Aqp4, C4b, and the gene encoding a serine protease inhibitor linked to increased amyloid accumulation (Serpina3n) and encoding a lysosomal cysteine protease involved in proteolytic processing of amyloid precursor protein, Ctsb (Cathepsin B) (Extended Data Fig. 10F). Morover, a set of genes including those involved in endocytosis (Vim), complement cascade (Osmr) and senescence (Ggta1) were also overexpression in AST4, confirming our AST4 as a DAA (Extended Data Fig. 10F). Gsk3b (glycogen synthase kinase 3β gene), Psen1 (presenile factor gene), Bdnf (brain- derived neurotrophic factor), and AD risk gene Sorl1 (encoding endosomal recycling receptor gene) and App, associated with AD pathological pathways were also highly expressed in our two DAA cells (Extended Data Fig. 10H- I). Then, we examined the expression levels of RNA signatures from bulk datasets, only ischemic related genes (Mcao) and inflammation related genes (Lps) were overexpressed in Ast2, but downregulated in Ast4 (Extended Data Fig. 11D).  
+
+Following the approach of AST, we also characterized the Oligo subpopulations (Extended Data Fig. 13A). Six Oligo subclusters were identified, and Oligo2 was characterized by Kcnip4, Nrg3, Csmd1 and Grin2a, which was consistent with Ast2 (Extended Data Fig. 13B). Moreover, the top up- regulated genes in AD were similarly distributed in the Oligo2 (Extended Data Fig. 13D), and its unique DEG
+
+<--- Page Split --->
+
+and GO term numbers were most than other sub- cells (Extended Data Fig. 13C and 14A), while the down- regulated DEGs were mainly derived from Oligo5 (Extended Data Fig. 14C). Unlike Miro and AST, comparable cell proportions of Oligo subclusters between AD and WT (Extended Data Fig. 13E), which was consistent with conventional cognition. We also observed that AD- related genes were overexpressed in Oligo2 (Extended Data Fig. 13G, H). Strikingly, marker genes of Oligo5, such as Gpc5, Ntm, Rora and Nrxn1, were also highly expressed in Ast4, which was specifically expressed by pathologically (Extended Data Fig. 13I). Among them, Gpc5 (Glypican 5) was the susceptibility gene for inflammatory demyelinating diseases \(^{33}\) . Ntm was involved in the negative regulation of neuronal projection development and acts upstream or within cell adhesion. Downregulation of Rora inhibits glioma proliferation through NF- \(\kappa\) B signaling pathway \(^{34}\) , and its regulatory effect was lost when Ast4 and Oligo5 are upregulated. Nrxn1 (Neurophin 1) was a cell adhesion molecule that plays a key role in establishing and maintaining synaptic connections, and its abnormal expression has been implicated in schizophrenia \(^{35}\) . Next, we analyzed the expression of marker genes from AD- pathology- associated Oligo \(^{28}\) in Oligo0- 5, and most of genes were more significantly expressed in the Oligo2 and Oligo5 (Extended Data Fig. 13J). In particularly, Qdpr, Nlgn1, Lama2 and Fchsd2, closely related to AD- pathology genes reported, were highly expressed in Oligo5 (Extended Data Fig. 13J).  
+
+Finally, to further confirm the accuracy of DAA and DAO (Disease- associated Oligodendrocytes) identification, we collected functional terms associated with AD pathology of previously reported, and analyzed the enrichment of these disease functions in Ast and Oligo subpopulations (Extended Data Fig. 12 and Supplementary Table 3). The results showed that almost all functions were enriched in the Ast2, including lipids, glial cell regeneration, endocytosis, NF \(\kappa\) B, endothelial cell differentiation and cognition (Extended Data Fig. 12A). Ast4 cells, however, were enriched with relatively independent functional sets, including functions in the regulation and regulation of growth, response to oxygen levels, and autophagy (Extended Data Fig. 12A). Moreover, GO terms enriched in Ast2 were also enriched in Oligo2 with stronger significance and enrichment index (Extended Data Fig. 12B). And the DEGs of Oligo2 were also enriched in autophagy, apoptosis, mRNA regulation and myelination (Extended Data Fig. 12B).  
+
+In a nutshell, the characteristics of DAA and DAO were similar in our snRNA- seq data of FFPE samples. Oligo2 and Ast2 had the same specific expression genes and transcription differences, while Ast4 and Oligo5 have similar transcription characteristics. We conjectured that a group of disease- susceptible gene sets caused similar transcriptional changes in different cell types, which in turn affected the occurrence and progression of AD.  
+
+## Integration of astrocytes and oligodendrocytes from multiple datasets  
+
+Given the abundance of high- quality, well- powered AD sample AST and Oligo snRNA- seq datasets in the literature, we next sought to determine whether we could resolve the same transcriptomic differences previously reported, and in turn demonstrate the availability of our nucleus. We evaluated AST and Oligo subtypes in each individual dataset and compared them with ours. We compared five AST clusters (G0- G4) in the Grubman dataset \(^{36}\) , four AST clusters (M0- M3) in the Mathys dataset \(^{28}\) ,
+
+<--- Page Split --->
+
+and 7 AST clusters (Z0- Z9) in the Zhou dataset \(^{23}\) , and 9 AST clusters (L0- L8) in the Liddelow dataset \(^{37}\) were integrated with our AST0- 4 for analysis, separately (Extended Data Fig. 11E). Similar analysis was performed on Oligo subtypes (Extended Data Fig. 14D). Using our AST and Oligo subpopulation profiles as a reference, we identified sub- cells that were also recognizable in the individual datasets. Although a complete one- to- one correspondence was not possible, we still observed that AST and Oligo subtypes were analyzed in the individual data, and disease- associated cells (Ast2, Ast4, Oligo2 and Oligo5) were clearly identified in all datasets, especially in Mathys and Liddelow and Multidatasets. For example, AST2 was highly correlated with G0, G3, M3, M4, L3- 6 cells, while AST4 was more correlated with G1, G4, M0, L7, L8 (Extended Data Fig. 11E). In contrast, AST0, Oligo0 and Oligo1 showed poor agreement in these datasets. In conclusion, the results of multi- channel data integration analysis of our AST and Oligo subclusters confirmed the aforementioned argument that the diversity of distribution detected in multiple frozen samples could be detected in our data, again demonstrating that our nucleus has cellular diversity, which lays the foundation for the study of disease heterogeneity.  
+
+## Transcriptional similarities in different disease-specific cell types  
+
+Although there are minimal transcriptional changes in neurons and other cells in AD cortex \(^{23}\) , and a recent work also focused only on astrocytes and oligodendrocytes \(^{38}\) . But the reported data of AD shows that all major cell types are affected by AD pathology at the transcriptional level \(^{28}\) , which was consistent with our results. In our snRNA- seq data of FFPE tissues, the Micro, GABA1, OPCs, AST, Ex.Neu, Oligo, Ex.CA3.1, GABA.2 cells were more perturbed by AD (Extended Data Fig. 9A). Moreover, more than 400 up- regulated DEGs were being in two vascular related cells, Epend and Endo, which even exceeded AST (Extended Data Fig. 9A). To test the previous conjecture that there is a disease susceptibility gene- set with consistent transcriptional differences in different cell types. We performed differential analysis of gene expression for all cell types perturbed by AD to explore the transcriptional similarities of disease- related cell types. (Extended Data Fig. 15A).  
+
+We observed that only the DEGs of Micro were cell- specific, while the most significant DEGs of the remaining glial cells and vascular cells were highly heterogeneous, and the top DEGs of neuronal cells also overlapped strongly (Extended Data Fig. 15B and Supplementary Table 4). Since the Log2FC of two vascular cells were too large to annihilating the information of the other cells, we present them independently (Extended Data Fig. 15C). The positional candidate or therapeutic marker genes, include the immune- related hub genes (Fgf13 \(^{39}\) and Etl4 \(^{40}\) ), the anti- inflammatory gene (Myo1e) \(^{41}\) , the multichannel transmembrane tonic transporter gene (Ank), and the cadherin- related protein gene (Ctma3) \(^{42}\) showed the greatest transcriptional changes only in micro cells (Extended Data Fig. 15B). However, Kcnip4, Grin2a, and Lrp1b were among the most differentially transcribed in other nonneuronal cells. The gene encoding Kv channel interacting protein 4 (Kcnip4) was a candidate gene for attention deficit hyperactivity disorder \(^{43}\) . The inability of the Kcnip4 isoform to interact with the secretase complex leads to increased secretion of beta- amyloid enriched in the more toxic Aβ- 42 species \(^{44}\) . And it also has been reported that Kcnip4 interacts with presenilin, and the presenilin gene
+
+<--- Page Split --->
+
+is associated with early- onset familial AD \(^{45}\) . Sleep deprivation (SD) could increase the risk of AD, and N- methyl- D- aspartate receptors (NMDAR) is an important cognitive regulator. Specific knockdown of hippocampal astrocytic Grin2a (the gene encoding the NMDAR subunit GluN2A) aggravated SD- induced cognitive decline, elevated \(\mathrm{A}\beta\) , and attenuated the SD- induced increase in autophagy flux \(^{46}\) . Most of these conclusions were based on the results of immunofluorescence staining, while our snRNA- seq data showed exactly the opposite, the Grin2a gene was not only highly expressed in Ast, but also positively expressed in most of the cell types associated with AD (Extended Data Fig. 15B). The low- density lipoprotein receptor- associated protein 1B (LRP1B) can interact with APP and regulate its processing to \(\mathrm{A}\beta^{47}\) . In summary, the transcriptional profiling of all cell types closely associated with AD reconfirmed our previous hypothesis that a single disease- susceptible gene set causes similar transcriptional changes in different cell types.  
+
+## Discussion  
+
+In this study, we developed a strategy for high- quality nuclear preparation from FFPE tissues by enzymatic dissociation of archived sample at low temperature without any tedious filtration step, which therefore provides a critical advance to profile single nuclei transcriptome from low- quality biological samples of PFA- fixed or FFPE tissues of. Meanwhile, we performed snRNA- seq on frozen, PFA- fixed and FFPE brains using \(10\times\) Genome and snRandom- seq technologies, and performed head- to- head comparison. To prove our method, we performed validation and obtained promising results. snRNA- seq was performed on frozen samples using \(10\times\) Genome and snRandom- seq technologies to eliminate platform differences, and the data of frozen samples was used as the gold standard for reference. We used snRandom- seq to perform snRNA- seq on PFA- fixed samples to compare the nucleus preparation strategies, and explored the applicability of snRNA- seq on FFPE samples as well as its application performance in brain diseases. The CED method and snCED- seq represents a significant advance in single- nuclei sequencing, enabling researchers to retrospectively select samples from a large paraffin sample bank, and facilitating mechanistic studies of brain disease samples that are difficult to obtain clinically.  
+
+Molecular biological application of FFPE tissues has always been challenging due to the chemical cross- linked and low- quality RNA. Although researchers have gradually become aware of the potential for obtaining expression profiles of individual cells or nuclei FFPE tissues, the approaches of extracting or isolating high- quality nuclei remains challenging. The acquisition of nuclei is one of the important conditions for snRNA- seq of FFPE samples, and its quality directly determines transcriptome analysis. The preparation methods of FFPE sample nucleus are longstanding, and are mainly divided into two categories, hyperthermal enzymatic dissociation strategies and mechanical extraction strategies. Enzymatically obtained nuclei are unhealrd of in transcriptomics studies. In fact, prolonged high temperature treatment resulted in secondary RNA degradation of FFPE samples, and prolonged exposure of nuclei to the enzyme buffer may increase the permeability of the nuclear membrane, leading to RNA molecules leakage and adversely affecting snRNA- seq experiments performed in droplets. The mechanical homogenization strategy was less damaging to RNA molecules
+
+<--- Page Split --->
+
+within nucleus. However, tissue homogenization for fixed and FFPE tissues, becomes increasingly challenging due to molecular cross- linking within the nuclei. Firstly, effectively removing excessive tissue debris poses difficulties and leads to severe contamination of the snRNA- seq data \(^{9}\) ; Moreover, the high proportion of rRNA requires additional removal processes when employing total RNA protocols \(^{21,48}\) . In our early experiments, the large amount of tissue debris interfered with accurate identification of nuclei due to the molecular cross- linking introduced by formaldehyde fixation, which requires a complicated debris removal process, thus affecting the yield of nuclei and losing smaller nuclei. Currently, the method employed in high- throughput snRNA- seq platforms rely on a combination of enzyme dissociation and homogenate \(^{10}\) . Despite the optimization of nuclear suspension and RNA quality within the nucleus, their own shortcomings have not been dismissed. Furthermore, all current methods for preparing nuclei from FFPE samples focus on tissue sections (5- 100 \(\mu \mathrm{m}\) ), while disease research often involves the tissue blocks.  
+
+Due to the characteristics of dissociation and digestion of nuclei in the preparation process of enzyme dissociation, the traditional high- temperature method makes the nuclei prepared first be digested in the dissociation solution, or the nuclear membrane is damaged, which affects the nuclear yield and is very sensitive to the reaction time, increasing the burden on the experimenter. Our nuclei were obtained by enzymatic hydrolysis of molecularly cross- linked tissues with a single step at low temperature, and without ultracentrifugation through a sucrose cushion and any filtration procedures, thereby maximizing product retention and nucleation rates. Taking a pair of mouse hippocampus as an example, the number of nuclei prepared by CED method was about 10 times that of the traditional method, and CED method can better enrich the small diameter nuclei missed by the traditional method. Most importantly, our CED method can effectively protect the nuclear membrane and maximally retains the nuclear molecules, providing high- fidelity nucleus for snRNA- seq research. For the latest snRNA- seq technology based on random primer capture \(^{10}\) or gene probe capture \(^{8}\) , it is necessary to input nearly one million nuclei on the premise of ensuring the output of about 10,000 nuclei. Our CED method effectively avoids the current two major nucleus preparation strategies, and can export the nucleus stably without introducing more impurities and damaging the nuclear membrane. The nuclei prepared by our CED method could be successfully preserved or transported on dry ice, which we speculated might be due to the fact that the permeability of the nuclear membrane was not damaged. In addition, our CED method has good applicability to a variety of organs, such as brain, liver, kidney, pancreas, spleen tissues, but slightly poor compatibility with heart and lung, although the yield of nuclei was still higher than that of the mechanical method. Heart has a complex cellular composition, mainly including myocardial tissue, nerve tissue, Purkinje fiber, connective tissue, epithelial tissue, etc. Similarly, the main connective component of the lung is composed of connective tissue, which is rich in collagen fibers, elastic fibers, reticular fibers. Connective tissue and cell relatively dense structure greatly increases the difficulty of the heart and lung tissue, and the choice of operating conditions and enzymes need further adjustment.  
+
+The excellent performance of the CED method was maintained for both PFA- fixed and FFPE tissues in our benchmarking effort. Compared with the HED high- throughput snRNA- Seq database of the PFA- fixed samples, snCED- seq outperforms well in various perspectives, supported by the genes and transcripts per nuclei, percent of mitochondrial and ribosomal genes, gene detection sensitivity, gene expression correlation with frozen samples, especially in gene expression richness. High- quality and
+
+<--- Page Split --->
+
+high- sensitivity snRNA- seq data from post- fixed (PFA- fixed and FFPE) specimens by snCED- seq allows to identify rare cell populations. We further provide a detailed map of cell- type- specific of FFPE samples from AD and WT mice, which highlights the predominance of gene expression richness in our nuclei. Multiple disease- related subpopulations have been successfully identified, and the DAM has transcriptional independence, while the transcriptional similarity between DAA and DAO subpopulations. There is even a population of genes (Kcnip4, Grin2a, Lrp1b, ect.) that are in the waiting state of activation, priority in different cells by the interference of the disease. In short, nuclei from CED method have excellent in revealing cellular heterogeneity, which contributes to the precision diagnosis and treatment to human disease.  
+
+Overall, this study proposed a novel method for the preparation of high- fidelity nuclei from post- fixed samples, which outperforms the traditional method in various aspects, and has good compatibility with a variety of FFPE organs. The application of FFPE samples in AD was also investigated, and found that our nuclei have great potential for uncovering disease cellular heterogeneity. The simple experimental protocols and comprehensive transcriptomic information from the FFPE tissues described in this study are expected to enable snCED- seq to large- scale applications in basic and clinical researches in the future. Our nuclear preparation strategy lays the foundation for revealing the transcriptomics and even multi- omics information of FFPE samples.  
+
+## Methods  
+
+## Ethical statement  
+
+The study was approved by the animal ethical and welfare committee of Zhongda Hospital Southeast University (approval numbers: 20200104005). All procedures were conducted following the guidelines of the animal ethical and welfare committee of SEU. All applicable institutional and/or national guidelines for the care and use of animals were followed.  
+
+## Experimental model  
+
+Male wildtype (WT) C57BL6/J mice (8 weeks of age) were ordered from Qinglongshan Animal Farm, Nanjing, China. AD and their control mice were purchased from Jiangsu Huachuang sinoPharmaTechCo., Ltd, Taizhou, China. Five- month- old heterozygous 5xFAD transgenic mice (on a C57/BL6 background) co- overexpress mutant forms of human amyloid precursor protein associated with familial AD, the Swedish mutation (K670N/M671L), the Florida mutation (1716V), the London mutation (V717I) and carry two FAD mutations (M146L and L286V) people PSEN1. The expression of both transgenes is regulated by the mouse neurospecific regulatory element Thy1 promoter to drive transgene overexpression in the brain. Throughout the study, all mice in each experiment were nontransgenic littermates from the same mouse colony.  
+
+All the mice were single- housed under standard laboratory conditions, including a 12h light/darkcycle, temperatures of \(25^{\circ}\mathrm{C}\) with \(40\%\) humidity, with free access to mouse diet and water. The animals were anesthetized with \(500\mathrm{mg / kg}\) tribromoethanol (Sigma, Saint Louis, MO, USA) and were killed by cervical dislocation. After the animals were sacrificed, hippocampi of brain were isolated. Fresh frozen (FF) tissues were obtained by quickly frozen in liquid nitrogen; PFA- fixed (PFA)
+
+<--- Page Split --->
+
+tissues were prepared by adding PFA (4 %) to the hippocampus and fixed for \(20\mathrm{h}\) at \(4^{\circ}\mathrm{C}\) ; FFPE samples were prepared by dehydration the fixed hippocampus twice in \(70\%\) , \(90\%\) and \(100\%\) ethanol, respectively, and then clearing with xylene solution for \(15\mathrm{min}\) , twice, followed by paraffin embedding for \(2\mathrm{h}\) ( \(62^{\circ}\mathrm{C}\) ). Frozen samples and PFA- fixed samples were stored in a \(- 80^{\circ}\mathrm{C}\) , and FFPE samples were stored at \(4^{\circ}\mathrm{C}\) .  
+
+## Single nuclei isolation from PFA-fixed and FFPE tissues  
+
+For frozen samples: the nuclei prepared by the homogenization method by Singleron Biological Tech Co. for \(10\mathrm{x}\) Genomics snRNA- seq and M20 Genomics for snRandom- seq.  
+
+For PFA fixed samples: the tissue was washed three times with \(1\mathrm{mL}\) PBS \((1\times ,\mathrm{pH} = 7.4)\) and cut into \(1\mathrm{mm}^3\) pieces in a \(2\mathrm{mL}\) enzyme- free centrifuge tube, adding \(1\mathrm{mL}\) dissociation buffer \((1.5\mathrm{mg / mL}\) Protease K, TE buffer, \(\mathrm{pH} = 8\) , \(0.5\%\) Sarkosyl), and shaking at low temperature overnight. The supernatant was centrifuged in a \(1.5\mathrm{mL}\) tube, centrifuged at \(10000\mathrm{rpm}\) for \(10\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) , and discarded supernatant. Then washed nuclei with \(1\mathrm{mL}\) pre- cold PBS \((1\times ,\mathrm{pH} = 7.4)\) twice, centrifuged at \(10000\mathrm{rpm}\) for \(10\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) . Finally, the nuclei were resuspended in \(200\mathrm{uL}\) of nuclear store buffer \((1\times \mathrm{PBS},0.2\mathrm{U / mLRNaseInhibitor})\) and stored at \(- 80^{\circ}\mathrm{C}\) .  
+
+For FFPE samples: the intact hippocampal tissue was trimmed out of the FFPE blocks with a sterilized scalpel and placed in a \(2\mathrm{mL}\) tube, and washed thrice with \(1.5\mathrm{mL}\) xylene for \(2\mathrm{h}\) at \(4^{\circ}\mathrm{C}\) to remove the paraffin. The samples were gently rehydrated by immersing the samples in a graded series of ethanol solutions, starting with pure \(100\%\) ethanol and ending with \(50\%\) ethanol \((100\% \times 2\) , \(95\%\) , \(70\%\) , \(50\% \times 1\) ) for \(1\mathrm{h}\) , then washed twice with pre- cold water. The steps of nuclei prepared was same as PFA tissue. An aliquot of nuclei was stained with DAPI \((4',6\) - diamidino- 2- phenylindole) staining solution, loaded on a hemocytometer and observed under an inverted fluorescence microscope. Eligible nuclei were stored on dry ice and sent to M20 Genomics.  
+
+## Library Construction and Sequencing  
+
+For frozen samples: Isolated nuclei were subjected to droplet- based \(3^{\prime}\) end massively paralleling using Chromium Single Cell \(3^{\prime}\) Reagent Kits per the manufacturer's instructions (10x Genomics).  
+
+For PFA- fixed and FFPE samples: In the single- cell transcriptome sequencing experiments of this study, we utilized the VITAcruizer single- cell preparation instrument DP400 (Cat #E20000131, M20 Genomics) to achieve droplet generation, single- cell partition and encapsulation, and nucleic acid capture. The VITApolite high- throughput FFPE single- cell transcriptome kit (Cat #R20121124, M20 Genomics) was employed for pre- library sample processing, single- cell library construction, and purification. Experimental procedures were conducted following the perspective kit and instrument manuals The main workflow is outlined below.  
+
+Nuclei were removed from dry ice and thawed at \(4^{\circ}\mathrm{C}\) , and the qualified single nuclei were subjected to snRNA- seq processing according to a previously published snRandom- seq protocol \(^{10}\) . Random primers were then added for the reverse transcription of total RNA inside the cell nuclei. Subsequently, the generated cDNA fragments were ligated to adaptors inside the cell nuclei. The reverse- transcribed
+
+<--- Page Split --->
+
+single- nucleus suspension and reagents, along with barcode beads containing cell barcodes and UMIs, were mixed. The mixture underwent encapsulation, capturing and barcoding using the VITAcruizer DP400 instrument. The resulting product underwent extension, resulting in barcoded cDNA strands. Following that, PCR amplification was conducted using the cDNA as the template. The purified products after cDNA amplification were then used to constructed a standard next- generation sequencing library. The constructed single- cell library contained P5 and P7 adapters and was sequenced on a Novaseq 6000 sequencer (Illumina) with 150 bp paired- end reads.  
+
+## Data analysis for C57BL6/J mice  
+
+## Preprocessing of snRNA-seq data.  
+
+Mus_musculus. GRCm39.109 reference genome was downloaded from ensemble database. Then we used STARsolo module in STAR (2.7.10a) with default parameters to generate the gene expression matrix and filter the valid nuclei. The Seurat v4.2 was applied for the major downstream analysis. Before we started downstream analysis, there are some filtering metrics to guarantee the reliability of each data.  
+
+detected in fewer than 3 cells were filtered to avoid cellular stochastic events. We deleted mitochondrial genes after the quality control, the left genes used for downstream analysis. For the cell part, we set different filter standards for each dataset according to the UMI and gene numbers distribution to filter low quality cells. Finally, we got 23248 genes and 142661 cells as the expression matrix to do downstream analysis in the method comparison part.  
+
+## Clustering and Cell Annotation  
+
+After quality control, unsupervised clustering was performed using Seurat v4.2. A series of preprocessing procedures including normalization, variance stabilization and scaling data, were performed in an R function 'SCTransform'based on regularized negative binomial regression. Then, we selected 2000 highly variable genes to integrate all sequencing libraries using 'FindIntegrationAnchors' and 'IntegrateData' functions, followed by the regression of technical noise. Principal component analysis (PCA) was performed using integrated output matrix, and the reasonable principal component (PC) numbers was chosen using the 'JackStraw' function. And we chose the top 30 significant PCs for downstream cluster identification and visualization. Clusters were defined based on 'FindClusters' function with resolution from 0.1 to 1 with 0.1 as seperation. Uniform Manifold Approximation and Projection (UMAP) was used for the final dimension reduction and visualization. Based on the cluster results with resolution equal to 0.2, we next used 'FindAllMarkers' function with MAST algorithm. We ranked the marker genes according to the p- value and log2 fold change (log2 FC) within each cluster and searched top genes in Cell Marker database \(^{49}\) and Panglao DB \(^{50}\) databases to annotate cell types of clusters.  
+
+## Differential expression analysis  
+
+Within each cluster, we calculated differentially expressed genes (DEGs) between 2 different conditions by using 'FindMarkers' function. we used 'MAST' setting as well and the Benjamini- Hochberg procedure to adjust p value. Then we set threshold q_ adjust \(< 0.05\) , absolute value of log2
+
+<--- Page Split --->
+
+FC \(>0\) to filter DEGs. The DEGs functional enrichment analysis based on Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) was applied by an R package ClusterProfile v4.10.0 using a hypergeometric test and corrected for multiple hypothesis by FDR.  
+
+## Data analysis for AD and WT mice  
+
+## Preprocessing of snRNA-seq data.  
+
+Mus_musculus. GRCm39.109 reference genome was downloaded from ensemble database. Then we used STARsolo module in STAR (2.7.10a) with default parameters to generate the gene expression matrix and filter the valid nuclei. The Seurat v4.4.0 was applied for the major downstream analysis. Before starting the downstream analysis, we used four filtering metrics to ensure the reliability of the data. (1) The detected genes were discarded to less than 3 cells to avoid cell random events; (2) Remove nuclei with mitochondrial gene expression percentage \(>10\%\) to exclude apoptotic cells; (3) Remove UMI \(>30000\) cells; (4) Remove cells outside the range of 300 to 5,000 genes. After filtering cells and genes based on the above metrics, we further use Doublet Finder V2.0 with default parameters to predict and remove potential doublet in each sample. Only cells that have passed a rigorous multi- step quality control regimen are considered for downstream analysis. Thus, 23,248 genes and 63,789 nuclei were retained in AD part.  
+
+## Clustering and cell Annotation  
+
+After quality control, we aligned the data from different batches using the SCTransform \(^{51}\) integration workflow in Seurat with default settings. We identified high- resolution clusters (resolution \(= 0.1\) ) using the Seurat functions FindNeighbors and FindClusters (Leiden clustering algorithm) based on the first 30 principal components. To annotate cell types within this dataset, we employed two distinct approaches: (1) manual annotation using previously published databases Cellmarker 2.0 and and Panglao DB databases, and (2) projecting annotations onto the cells analyzed in this study by integrating the clustering results with the dataset from Habib et al. \(^{25}\) using the Seurat functions FindTransferAnchors and TransferData. Combining these methods yielded detailed and reliable cell cluster information. We also excluded certain less reliable cell types (e.g., Ex.neuron2 and Ex.IEG). During the subsequent cell proportion analysis, we observed a significant imbalance in the cell proportions of the AD1 sample, with neurons comprising the majority and non- neuronal cells only accounting for \(4\%\) . Thus, we deemed this sample unreliable and excluded it from further analysis. Ultimately, we retained 23,248 genes and 52,569 cells.  
+
+For the subcluster analysis of astrocytes, oligodendrocytes, and microglia, we similarly utilized the Seurat functions FindNeighbors and FindClusters on the top 30 principal components, setting the resolution to 0.1.  
+
+## Differential expression analysis  
+
+Within each cluster, we used the "FindMarkers" function in the Seurat package to detect DEGs between AD and WT conditions. We applied the "MAST" setting and controlled the false discovery rate (FDR) using the Benjamini- Hochberg procedure. We set thresholds of \(|\mathrm{avg\_log2FC}| > 0.25\) and p_val_adj \(< 0.05\) to filter DEGs, identifying both upregulated and downregulated genes in AD relative
+
+<--- Page Split --->
+
+to WT for each cluster.  
+
+## Enrichment analysis  
+
+The functional enrichment analysis of Differentially Expressed Genes (DEGs) based on GO biological processes and the KEGG was conducted using the R package ClusterProfile v4.11.1. The analysis employed a hypergeometric test and corrected for multiple hypotheses using the False Discovery Rate (FDR). For enrichment between gene sets, the testGeneOverlap function from the R package GeneOverlap v1.34.0 was utilized to perform Fisher's exact test, identifying overlaps among different gene sets.  
+
+## Comparison with external data sets  
+
+To compare our data with external datasets, we collected single- cell data from studies by Grubman, Mathys, Zhou, Liddelow, and others. Specifically, we gathered the top markers for astrocytes, oligodendrocytes, and microglia from these studies. We then analyzed the relative expression levels of these markers in our own dataset to identify corresponding AD- related cell subtypes mentioned in these publications. This comparative approach allowed us to validate our findings and highlight specific cell types associated with Alzheimer's disease in our study.  
+
+## Acknowledgements  
+
+We thank all members of Xiangwei Zhao and Guangzhong Wang's lab for their help and suggestions, and the team of M20 Genomics for their help with snRNA- seq experiments. I would like to thank J.J. Ma and R.C. Qi for their support. This work was conducted in the context of National Natural Science Foundation of China (81827901, 82361138570).  
+
+## Author contributions  
+
+Y.X.G. conceived and developed CED method and snCED- seq. Y.X. Guo optimized the protocol. Y.X.G. conducted the experiments. J.J.M. and R.C.Q. performed analyzed the data. Y.X.G. visualized the results. Y.X.G. prepared the figures. M20 Genomics prepared libraries and performed next- generation sequencing. J.T.X. contributed to the mouse brain tissue sampling. Y.X.G. wrote the manuscript with contributions from all authors. X.W.Z. and G.Z.W. supervised the project. All authors had read and agreed to the published version of the manuscript.  
+
+## Competing interests  
+
+The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.  
+
+## Data availability  
+
+The single nuclei RNA sequencing data that support the findings of this study have been deposited in the National Center for Biotechnology Information (NCBI) with accession number PRJNA1117576.  
+
+## References
+
+<--- Page Split --->
+
+1. Shalek, A.K. et al. Single-cell transcriptomics reveals bimodality in expression and splicing in immune cells. Nature 498, 236-240 (2013).  
+
+2. Jaitin, D.A. et al. Massively Parallel Single-Cell RNA-Seq for Marker-Free Decomposition of Tissues into Cell Types. Science 343, 776-779 (2014).  
+
+3. Rozenblatt-Rosen, O., Stubbington, M.J.T., Regev, A. & Teichmann, S.A. The Human Cell Atlas: from vision to reality. Nature 550, 451 (2017).  
+
+4. Gnanapragasam, V.J. Unlocking the molecular archive: the emerging use of formalin-fixed paraffin-embedded tissue for biomarker research in urological cancer. Bju International 105, 274-278 (2010).  
+
+5. Amini, P. et al. An optimised protocol for isolation of RNA from small sections of laser-capture microdissected FFPE tissue amenable for next-generation sequencing. Bmc Molecular Biology 18, 22 (2017).  
+
+6. Maliga, Z. et al. Micro-region transcriptomics of fixed human tissue using Pick-Seq. Preprint at https://biorxiv.org/content/early/2021/03/18/431004. (2021).  
+
+7. Liu, Y. et al. High-plex protein and whole transcriptome co-mapping at cellular resolution with spatial CITE-seq. Nat Biotechnol 41, 1405-1409 (2023).  
+
+8. Vallejo, A.F. et al. snPATHO-seq: unlocking the FFPE archives for single nucleus RNA profiling. Preprint at https://doi.org/10.1101/2022.08.23.505054 (2022).  
+
+9. Chung, H. et al. SnFFPE-Seq: towards scalable single nucleus RNA-Seq of formalin-fixed paraffin embedded (FFPE) tissue. Preprint at https://doi.org/10.1101/2022.08.25.505257 (2022).  
+
+10. Xu, Z. et al. High-throughput single nucleus total RNA sequencing of formalin-fixed paraffin-embedded tissues by snRandom-seq. Nat Commun 14, 2734 (2023).  
+
+11. Hedley, D. Method for analysis of cellular DNA content of paraffin-embedded pathological material using flow cytometry. J. Histochem Cytochem 31 (1983).  
+
+12. Thomas Liehr, Claussen Uwe & Erich, G. Nucleus extraction from single mounted tissue sections. Genet. Anal-Biomol. E. 15, 65-69 (1999).  
+
+13. Paternoster, S.F. et al. A New Method to Extract Nuclei from Paraffin-Embedded Tissue to Study Lymphomas Using Interphase Fluorescence in Situ Hybridization. Am. J. Pathol. 160, 1967-1972 (2002).  
+
+14. Schutte, B., Reynders, M.M., Bosman, F.T. & Blijham, G.H. Flow cytometric determination of DNA ploidy level in nuclei isolated from paraffin-embedded tissue. Cytometry 6, 26-30 (1985).  
+
+15. Martelotto, L.G. et al. Whole-genome single-cell copy number profiling from formalin-fixed paraffin-embedded samples. Nat. Med. 23, 376-385 (2017).  
+
+16. Chin, H.G. et al. Universal NicE-seq for high resolution accessible chromatin profiling for formaldehyde fixed and FFPE tissues. Preprint at https://doi.org/10.1101/2020/04/27/064592. (2020).  
+
+17. Maitra, M. et al. Extraction of nuclei from archived post-mortem tissues for single-nucleus sequencing applications. Preprint at 10.21203/rs.3.pex-905/v3 (2020).  
+
+18. Phan, H.V. et al. High-throughput RNA sequencing of paraformaldehyde-fixed single cells. Nat Commun 12, 5636 (2021).  
+
+19. Thomsen, E.R. et al. Fixed single-cell transcriptomic characterization of human radial glial diversity. Nat Methods 13, 87-93 (2016).  
+
+20. Alles, J. et al. Cell fixation and preservation for droplet-based single-cell transcriptomics. BMC
+
+<--- Page Split --->
+
+Biol 15, 44 (2017).  
+
+21. Salmen, F. et al. High-throughput total RNA sequencing in single cells using VASA-seq. Nat Biotechnol 40, 1780-1793 (2022).  
+
+22. Guo, Y. et al. Transcriptomic profiling of nuclei from paraformaldehyde-fixed and formalin-fixed paraffin-embedded brain tissues. Analytica Chimica Acta 1281, 341861 (2023).  
+
+23. Zhou, Y. et al. Human and mouse single-nucleus transcriptomics reveal TREM2-dependent and TREM2-independent cellular responses in Alzheimer's disease. Nat. Med. 26, 131-142 (2020).  
+
+24. Guo, Y. et al. Defining Specific Cell States of MPTP-Induced Parkinson's Disease by Single-Nucleus RNA Sequencing. Int. J. Mol. Sci. 23, 10774-10794 (2022).  
+
+25. Habib, N. et al. Disease-associated astrocytes in Alzheimer's disease and aging. Nat. Neurosci. 23, 701-706 (2020).  
+
+26. Cao, J.L., Liu, J.Y., Xiao, J., Jiang, X.R. & Huang, H.Y. Olfactory ensheathing cells transplantation reduced amyloid burden in amyloid precursor protein transgenic mouse model. Acta Anatomica Sinica 41, 823-826 (2010).  
+
+27. Fan, W.J., Cheng, W.J., Niu, Y.L., Li, M.S. & Deng, J.B. Normal development of Cajal-Retzius cells in mouse hippocampus and their changes in APPswe transgenic mice. Acta Anatomica Sinica 41, 211-218 (2010).  
+
+28. Mathys, H. et al. Single-cell transcriptomic analysis of Alzheimer's disease. Nature 570, 332-337 (2019).  
+
+29. Shen, L., Thompson, P.M., Potkin, S.G., Bertram, L. & Saykin, A.J. Genetic analysis of quantitative phenotypes in AD and MCI: imaging, cognition and biomarkers. Brain Imaging & Behavior 8, 183-207 (2014).  
+
+30. Potkin, S.G. et al. Hippocampal Atrophy as a Quantitative Trait in a Genome-Wide Association Study Identifying Novel Susceptibility Genes for Alzheimer's Disease. PLoS ONE 4, 6501 (2009).  
+
+31. Smith, X., Taylor, A. & Rudd, C.E. T-cell immune adaptor SKAP1 regulates the induction of collagen-induced arthritis in mice - ScienceDirect. Immunology Letters 176, 122-127 (2016).  
+
+32. Yixing, Davies, Kay, E., Oliver & Peter, L. The antioxidant protein Oxr1 influences aspects of mitochondrial morphology. Free Radic Biol Med 95, 255-267 (2016).  
+
+33. Shin et al. Putative association of GPC5 polymorphism with the risk of inflammatory demyelinating diseases. Journal of the Neurological Sciences Official Bulletin of the World Federation of Neurology 335, 82-88 (2013).  
+
+34. B, Y.J.A. et al. MiR-18a-downregulated RORA inhibits the proliferation and tumorigenesis of glioma using the TNF-α-mediated NF-κB signaling pathway. EBioMedicine 52, 102651 (2020).  
+
+35. Uzunhan, T.A. & Ayaz, A. Homozygous exonic and intragenic NRXN1 deletion presenting as either West syndrome or autism spectrum disorder in two siblings. Clinical Neurology and Neurosurgery 214, 107141 (2022).  
+
+36. Grubman, A. et al. A single-cell atlas of entorhinal cortex from individuals with Alzheimer's disease reveals cell-type-specific gene expression regulation. Nat Neurosci 22, 2087-2097 (2019).  
+
+37. Liddelow, S.A. Neurotoxic reactive astrocytes are induced by activated microglia. Nature 541, 481-487 (2017).  
+
+38. Sadick, J.S. et al. Astrocytes and oligodendrocytes undergo subtype-specific transcriptional changes in Alzheimer's disease. Neuron 110, 1788-1805 (2022).
+
+<--- Page Split --->
+
+39. Zhao, K. et al. Integrated analysis and identification of hub genes as novel biomarkers for Alzheimer's disease. Frontiers in Aging Neuroscience 14, 1-14 (2022).  
+
+40. Kamphuis et al. Isolation of glia from Alzheimer's mice reveals inflammation and dysfunction. Neurobiology of Aging 35, 2746-2760 (2014).  
+
+41. Rangaraju, S. et al. Identification and therapeutic modulation of a pro-inflammatory subset of disease- associated-microglia in alzheimer's disease. Molecular Neurodegeneration 13, 24-49 (2018).  
+
+42. Busby, V.L., Goossens, S., Nowotny, P. & Lovestone, S. α-T-Catenin Is Expressed in Human Brain and Interacts With the Wnt Signaling Pathway But Is Not Responsible for Linkage to Chromosome 10 in Alzheimer's Disease. NeuroMolecular Medicine 5, 133-146 (2004).  
+
+43. Weissflog, L., Scholz, C.-J., Jacob, C.P., Nguyen, T.T. & Reif, A. KCNIP4 as a candidate gene for personality disorders and adult ADHD. European neuropsychopharmacology 23, 436-447 (2012).  
+
+44. Massone, S. et al. RNA polymerase III drives alternative splicing of the potassium channel-interacting protein contributing to brain complexity and neurodegeneration. Journal of Cell Biology 193, 851-866 (2011).  
+
+45. Morohashi & Y. Molecular Cloning and Characterization of CALP/KChIP4, a Novel EF-hand Protein Interacting with Presenilin 2 and Voltage-gated Potassium Channel Subunit Kv4. Journal of Biological Chemistry 277, 14965-14975 (2002).  
+
+46. Ros-Bernal, F. et al. Neuroinflammation and Parkinson's Disease. Springer New York (2014).  
+
+47. Judy et al. The Low Density Lipoprotein Receptor-related Protein 1B Retains β-Amyloid Precursor Protein at the Cell Surface and Reduces Amyloid-β Peptide Production. Journal of Biological Chemistry 279, 29639-29646 (2004).  
+
+48. Sang, G. et al. High throughput detection of variation in single-cell whole transcriptome through streamlined scFAST-seq. Preprint at https://doi.org/10.1101/2023.03.19.533382 (2023).  
+
+49. Zhang, X. et al. CellMarker: a manually curated resource of cell markers in human and mouse. Nucleic Acids Res. 47, 721-728 (2018).  
+
+50. Oscar et al. PanglaoDB: a web server for exploration of mouse and human single-cell RNA sequencing data. J. Biol. Data. 2019, 1-9 (2019).  
+
+51. Hafemeister, C. & Satija, R. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Cold Spring Harbor Laboratory 20, 296 (2019).
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+SupplementaryMaterial.pdf SupplementaryTable1.xlsx SupplementaryTable2.xlsx SupplementaryTable3.xlsx SupplementaryTable4.csv
+
+<--- Page Split --->

preprint/preprint__7e519e008d15c7e10d8d8f8234fc8066280b0cc87c77ab82d8b32e00f8df0572/images_list.json

@@ -0,0 +1,122 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. Schematic description of gradient BMG. (a) Schematic representation of one dominant shear band in uniform MG. (b) Schematic diagram of proposed GMG. (c) Proposed shear band deflection mechanism in a GMG during the uniaxial compression process.",
+    "footnote": [],
+    "bbox": [
+      [
+        171,
+        87,
+        830,
+        491
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Generation and Characterization of the GMG. (a) Schematic of the GMG by means of CTC. (b) Density values as a function of holding time for the treated MGs. (c) Cryogenic thermal cycling procedure from a high temperature (323 K) to liquid nitrogen temperature (77 K), together with the waiting time, \\(t\\) , at both the maximum and the minimum temperatures. (d) Variation of average hardness value along the distance from the center. (e) The top shows the method for hardness measurements along various circles. The bottom shows the t150 sample for TEM. (f)-(h) TEM images with the corresponding selected-area-diffraction patterns (SAED) of the edge, middle, and center regions (labeled A, B, C) for the t150 sample. (i) Corresponding radial distribution functions calculated from the SAED of the edge, middle, and center regions (labeled A, B, C) in the t150 sample.",
+    "footnote": [],
+    "bbox": [
+      [
+        149,
+        108,
+        843,
+        551
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. Enhanced mechanical properties, fracture surface change and shear band deflection. (a) Compressive stress-strain curves for the as-cast and treated MGs. (b) Variation of the plastic strain with the structural gradient. (c)-(g) Lateral morphologies of the fractured as-cast and gradient MG samples. (h)-(l) The corresponding 3D contours of the fracture surfaces. (m) Height variation profiles of the middle dash line (h) along the shear band plane for as-cast and gradient MG samples. (n)-(s) SEM surface morphologies at positions 1-3 in the as-cast (c) and 4-6 in the t150 sample (g).",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        125,
+        850,
+        608
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4. Investigation of the deformation behavior of GMG by MD simulations. (a) Atomic free volumes as the functions of the position along the Y-direction in as-cast and gradient MGs. Inset shows an illustration of the design of GMG. (b) Representative stress-strain results for the as-cast and gradient MGs during compression along the X-direction. (c) The spatial distribution of atomic Mises strain (d) Rotation angle of the as-cast and gradient MGs at \\(9\\%\\) compression strain. (e) The variations of the rotation angle along lines 1-3 in (d). (f) Schematic illustration of the STZ percolation mechanism in GMG.",
+    "footnote": [],
+    "bbox": [
+      [
+        164,
+        80,
+        830,
+        700
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        55,
+        100,
+        940,
+        700
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        55,
+        60,
+        925,
+        680
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        62,
+        60,
+        920,
+        722
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        66,
+        53,
+        768,
+        771
+      ]
+    ],
+    "page_idx": 23
+  }
+]

preprint/preprint__7e519e008d15c7e10d8d8f8234fc8066280b0cc87c77ab82d8b32e00f8df0572/preprint__7e519e008d15c7e10d8d8f8234fc8066280b0cc87c77ab82d8b32e00f8df0572.mmd

@@ -0,0 +1,256 @@
+
+# Extra plasticity governed by shear band deflection in gradient metallic glasses  
+
+Yao Tang ( \(\boxed{\pm}\) tangyao@zju.edu.cn) International Center for New- Structured Materials (ICNSM), Zhejiang University  
+
+Haofei Zhou Zhejiang University https://orcid.org/0000- 0001- 9226- 9530  
+
+Xiaodong Wang Zhejiang University  
+
+Qingping Cao Zhejiang University  
+
+Dongxian Zhang Zhejiang University  
+
+Wei Yang Zhejiang University  
+
+Jian- Zhong Jiang Zhejiang University  
+
+## Article  
+
+Keywords: Controlled Structural Gradients, Heat Treatment Engineering Protocol, Cryogenic Thermal Cycling  
+
+Posted Date: April 5th, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 366951/v1  
+
+License: © \(\circledast\) 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 April 19th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29821- 4.
+
+<--- Page Split --->
+
+# Extra plasticity governed by shear band deflection in gradient metallic glasses  
+
+## Abstract  
+
+Inspired by gradient materials in nature, advanced engineering components with controlled structural gradients have attracted significant research interest due to their exceptional combinations of properties. However, it remains challenging to generate structural gradients that penetrate through bulk materials, which is essential for achieving enhanced mechanical properties in metallic materials. Here, we propose a heat treatment engineering protocol to realize a controllable structural gradient in bulk metallic glasses (BMGs). By adjusting the holding time of cryogenic thermal cycling, a series of BMGs with gradient- distributed free volume contents from internal to external can be synthesized. Both mechanical testing and atomistic simulations demonstrate that the spatial gradient can endow BMGs with extra plasticity without sacrificing their ultrahigh strength. Such an enhanced mechanical property is governed by the gradient- induced deflection of shear deformation that fundamentally suppresses the unlimited shear localization on a straight plane that would be expected in BMGs without such a gradient.
+
+<--- Page Split --->
+
+Advances in modern science and technology continue to impose more stringent requirements for engineering materials, including exceptional strength and toughness. Unfortunately, these two properties are almost mutually exclusive in monolithic materials \(^{1,2}\) . Obtaining optimal mechanical performance is always a compromise, one which can be achieved by optimizing the microstructure through empirical design. Notably, the introduction of structural gradients can overcome the strength- ductility trade- off in metallic materials and give rise to high- performance functionalities \(^{3 - 8}\) . Concerning such gradients, nature provides a rich source of inspiration. Many natural materials have highly sophisticated structures with complex gradient designs that possess extremely impressive combinations of properties significantly surpassing those of their constituents \(^{9 - 13}\) . In view of the gradient structures of natural materials, exploring structural gradients to enhance the properties of engineering materials has generated strong interest. Typical examples are widely- exploited gradient metals with nano- grained (NG) \(^{14}\) or nano- twinned (NT) structures. \(^{15}\) In contrast to conventional homogeneous coarse- grained (CG) materials, the deformation mechanism of gradient nanostructured (GNS) materials is often heterogeneous and is regulated and constrained by the gradient structure. Also, structural gradients typically cause stress gradients and even activate new dislocation structures \(^{8}\) . Nevertheless, current GNS materials are limited to a few, pure face- centered- cubic metals and typical alloys. For example, Mg alloys can be strengthened by introducing a gradient nanograined structure while this strategy is unable to provide large ductility in Mg alloys. Recently, an Mg- based nano dual- phase metallic glass (NDP) coated on a gradient nano- grained Mg alloy showed enhanced ductility and yield strength compared to the base alloy \(^{16}\) . The success of this design strategy of combining heterogeneous metallic glass (MG) and gradient nanograined structure provides us a motivation to extend the principles of structural gradients to amorphous systems in designing ‘intrinsic’ gradient MGs (GMGs). Indeed, MGs with extraordinary physical and biomaterial properties have recently been developed \(^{17}\) , but severe brittleness holds one major weakness that precludes the wide application of MGs. The introduction of spatial gradients may offer a promising solution for tuning deformation behavior and enhancing the plasticity of MGs.  
+
+Over the past several years, various fabrication methods have been applied to develop structural gradients in engineering materials. The fabrication methods can be divided into two categories: bottom- up methods, including physical and chemical deposition \(^{18}\) , layer- by- layer assembly \(^{19}\) , and three- dimensional printing \(^{20}\) ; and top
+
+<--- Page Split --->
+
+down methods including surface mechanical treatment methods<sup>21- 23</sup>, laser shock peening<sup>24</sup>, and roll bonding<sup>25</sup>. Despite their widespread use in engineering design, these methods suffer from marked constraints. Bottom- up methods are generally only feasible for making thin films or microscopic samples. Existing top- down methods, on the other hand, have limits for the range of bulk gradient materials. For instance, surface mechanical treatments always produce a limited volume fraction of gradients only near the surface, or they generate a negligible degree of structural gradients along the gradient direction. All of the aforementioned issues limit our ability to achieve a gradient throughout the bulk metallic glass (BMG) samples. It is essential to develop new strategies and practical methods to design and fabricate gradient BMGs to tailor their mechanical properties.  
+
+In this paper, we report a design strategy and a set of simple fabrication methods to produce the GMG in bulk form by introducing a controllable spatial gradient of the free volume content. Through experiments and MD simulations, we demonstrate that the excellent performance of GMG can be attributed to its "shear band deflection" capability that arises from its intrinsic gradient structure. A notable difference in the local free volume defects the angle of the shear band initiation and propagation. Using model heterogeneous materials, we discuss the atomic- scale origin of the observed variations in the SB dynamics and the angle with changing structural state. The design strategy offers a simple and yet versatile method to improve the mechanical properties of BMGs and, more importantly, to design new generations of high- performance structural materials.  
+
+## Results  
+
+The design strategy for gradient BMG. The design strategy for the GMG structure is proposed in Fig. 1. The plastic deformation of uniform BMGs is through shear localization into narrow bands (Fig. 1a). Such localization often leads to the running away of one dominant shear band, eventually leading to catastrophic failure and macroscopic brittle behavior<sup>26</sup>. The shear band plane in BMGs occurs along an angle at which the corresponding effective shear stress is maximized, which suggests the important influence of normal stress on the shear plane<sup>27,28</sup>. The normal stress effect on deformation in MG lies in the principle of atomistic friction, as embodied in the Mohr- Coulomb criterion:
+
+<--- Page Split --->
+
+\[\tau_{y} = \tau_{0} - \alpha \sigma_{n}\]  
+
+where, \(\tau_{y}\) is the effective shear yield stress, \(\tau_{0}\) is a constant, and \(\alpha\) is an effective coefficient of friction that controls the strength of the normal stress effect<sup>29,30</sup>.  
+
+We propose that the plasticity of BMGs can be enhanced through a gradient design of the microstructure, with the free volume concentration increasing from the outer to the inner part of the cylindrical BMG specimen (Fig. 1b). Changes in the free volume content and bonding conditions of structural units are expected to modify friction coefficient \(\alpha\) and therefore shear band angle \(\theta\) . Figure 1c shows schematic illustrations of the development of shear bands in such a gradient BMG specimen. The primary shear band initiates at the upper- left surface with a relatively lower content of free volumes, corresponding to a relatively larger friction coefficient \(\alpha\) . Taking normal stress into consideration, the effective shear yield stress is maximized at shear band angle \(\theta\) for the local hard region. As the shear band progresses toward the central soft region of the specimen, the increasing value of free volume concentration alters the normal stress effect on the shear band, inducing a gradual increase in the shear band angle. The shear band is thus deflected by the structural gradient of the BMG. Afterwards, as the shear band propagates from the center to the lower- right surface, free volume concentration declines, leading to a gradual decrease in the shear band angle and a reversed deflection pattern of the shear band. In short, the deflected shear band path avoids the, otherwise straight, transecting shear band across the whole sample, which provides a promising route for improving the plastic deformability of BMGs.  
+
+Characteristics of the GMG. To test this design strategy, we selected a relatively brittle \(\mathrm{Zr_{58}Cu_{22}Fe_{8}Al_{12}}\) MG to construct the gradient BMG structure. We used cryogenic thermal cycling between a certain temperature (323 K) and cryogenic temperature (77 K) to prepare the gradient samples. Figure 2a shows a detailed description of the cryogenic thermal cycling treatment apparatus used to introduce a gradient rejuvenation into the cylindrical MG samples. When the sample was first heated at a certain temperature and then cooled to a low temperature, the rejuvenation behavior came first, which is mainly related to the quasi- localized vibrations of atoms in the flow unit surrounded by the elastic matrix. The sample must be held for a sufficiently long time to enhance its rejuvenation effect. With a long holding time at high temperature, the atoms within flow units will move cooperatively and reversibly on a large scale, and
+
+<--- Page Split --->
+
+resulting in a fast relaxation in turn. Therefore, it can be expected that dynamic rejuvenation or relaxation behaviors vary at internal and external parts of the sample during their evolution with time. Controlling the holding time can induce gradient rejuvenation (free volume content) processes.  
+
+Figure 2b shows the results of density measurements of the as- cast sample and the treated samples. The treated samples refer to ten- cycled samples using the same thermal cycling process but with different holding times, as shown in Fig. 2c. The density of the treated samples is lower than that of the as- cast samples, which suggests a relatively large free volume content. The density of the treated sample decreases with the increased holding time but seemingly saturates when \(t\) is larger than about 60 s. Figure 2d shows the variation of hardness across the diameter on a cross- section of the \(2\mathrm{mm}\) cylindrical as- cast and treated samples (the top of Fig. 2e shows the method for hardness measurements along various circles, in which 8 indentations were performed to acquire an average harness value at each circle, together with 80 indentations from the center to the edge). For the t6 sample (treated with a 6 s holding time), the hardness exhibits little decrease from a distance of \(0.3\mathrm{mm}\) to the center. Notably, a gradient of the hardness value can be detected for the t25 sample. From the edge to the center, the hardness value of the t25 sample tends to decrease from 500 HV to 485 HV, respectively. In particular, a more obvious hardness- value gradient can be seen for the t70 and t150 samples. The gradient hardness values suggest that this method can provide a potentially low- cost manufacturing process for the scalable production of GMGs.  
+
+Transmission electron microscopy (TEM) analyses were performed to characterize the amorphous structure at different positions of the t150 GMG sample (bottom of Fig. 2e). Specifically, Figure 2f- h display the TEM images of edge, middle, and center regions (labeled A, B, and C) for the t150 sample. One can clearly see a grain- like microstructure with a dark- bright contrast in the sample at the edge. Of significant interest, for the sample in the middle, the sizes of the dark and bright regions are enlarged to 2–3 nm. The characteristic length of the inhomogeneous microstructure reaches 5 nm in the sample at the center. Considering that our TEM results show an increasing heterogeneity with a decreasing distance from the center, the hypothesis suggested by Ketov et al. provides a reasonable explanation for the gradient amorphous microstructures<sup>31</sup>. Ketov et al. attribute the rejuvenation effects to intrinsic non- uniformity of the glass structure, which gives a non- uniform coefficient of thermal expansion. The population and intensity of soft spots (dark), with lower elastic stiffness
+
+<--- Page Split --->
+
+and higher CTE, increases with cycling. Hence, the brighter contrast in our TEM images at the center region may result from a lower density zone, while the edge region has a relatively high density in our treated samples. Following the corresponding selected area electron diffraction (SAED) patterns (insets of Fig. 2f- h), we confirm that the observed microstructures are all structurally amorphous. The radial distribution functions (RDFs, Fig. 2i), which were calculated from the SAED patterns of three parts in the t150 sample, have differences in their peak positions. It was found that the first peak position in RDFs is shifted to higher r values from the edge to the center of the treated sample, i.e., the average atomic bond distances increased from edge to center. All observed experimental results - i.e., at the center with a reduction in hardness, larger bright contrast in TEM images, and larger average atomic bond distance - reveal an enhancement of free volume in the center part. Therefore, gradient structure (or rejuvenation) indeed occurs from edge to center in the treated samples.  
+
+Mechanical behavior and deformation mechanisms of GMG. To demonstrate the enhanced plastic deformability of gradient BMGs, we compared the engineering compressive stress- strain curves of the gradient BMG samples with the as- cast sample (Fig. 3a). The as- cast sample undergoes negligible plastic strain before fracture, typical of the strong- and- brittle behavior reported in the literature. Of importance, plasticity strongly increased significantly without the expense to the strength in the gradient samples, reaching maximum when the holding time was 70 s and 150 s. For comparison, Fig. 3b summarizes variations of measured plastic strain as a function of the structural gradient. The structural gradient is a parameter that quantifies the structure difference of cryogenically treated samples as the change in hardness per unit thickness along the gradient direction. As illustrated in Fig. 3b, we observed an increase in plastic strain with the increase of the structural gradient. The samples with a structural gradient of 16 HV/mm exhibited the largest plastic strain, about four times that of the as- cast sample. The above observations strongly suggest that a substantial plasticity increment can be achieved solely by introducing the structural gradient.  
+
+To uncover the physical mechanisms underlying ductility enhancement in the GMGs, we explored the key structural parameters that are affected by the structural gradient. Figure 3c- g display the lateral morphologies of the as- cast and GMG samples obtained by SEM after the final failure. One dominating primary shear band plane, a typical failure mode observed in brittle MGs, exists in the as- cast sample (Fig. 3c). In
+
+<--- Page Split --->
+
+contrast, the GMG samples demonstrate interesting fracture surface morphology. For the t6 sample (Fig. 3d), the fracture surface looks somewhat uneven. A small bulge on the fracture surface can be seen for the t25 sample. For the t70 and t150 samples (Fig. 3f and g), some humps can be clearly observed in the center region of the fracture surface. These fracture surfaces in Fig. 3c-g indicate that the shear band plane deflects obviously from the original shear planes when the shear bands propagate into the center of the GMG samples. To accurately characterize these fracture surfaces, three- dimensional (3- D) profiles of the fracture surfaces are also displayed in Fig. 3h- l. Ridges running parallel to the shear band plane direction in the middle (see line in Fig. 3h) are connected by a set of transverse ridges. It can be seen that the ridges in the fracture surface of the as- cast and t6 samples are almost flat. For the t25 sample, the branching of the ridges as well as their meandering in different directions as the shear plane front advances can be noted. An obvious hump can be observed in the height variation along the shear band plane from top to bottom of the fracture surfaces in the t70 and t150 samples. The shear band plane looks like a non- uniform surface with a bent shape, which means that the fracture angle changes at a particular stage during shear band propagation. Distributions of ridge heights, obtained from 2- D profiles of fracture surfaces along the dash line (Fig. 3h), are shown in Fig. 3m. The relative homogeneous as- cast sample only shows small height fluctuation. In contrast, the GMG samples exhibit marked height differences, i.e. lower on both sides and higher in the middle. From the calculated fracture angle at each point along this variation line, the fracture angle in the center is larger than that in the outer part, which suggests that shear band deflection indeed occurs during deformation. It is apparent that this is direct evidence that a novel non- uniform deformation mechanism in GMG occurred. The structural gradient exerts a direct influence on the shear banding, and the deflection of the shear band is thought to be closely related to the variation of the free volume content. In the edge region near the fracture origin of samples, a larger \(\alpha\) value might be expected, and consequently a larger effect of normal stress. In this case, the more difficult is the expected movement among structural units. The material also shows brittle behavior - hence the small angle of the shear band plane. On the other hand, a smaller \(\alpha\) might be associated with an increase in free volume content by gradient rejuvenation in the center part, leading to easier plastic flow and a larger angle of shear band plane. Our SEM images of the fracture surface morphology verify the above- mentioned analyses. Fig. 3n- s clearly show the morphologies of fracture as- cast and
+
+<--- Page Split --->
+
+t150 samples at different positions. The three positions of the as- cast sample exhibit typical viscous, river- like patterns along the shear direction with very narrow spacing. Distinct surface morphologies developed in the center regions of the two samples, which indicates a difference in the local free volume content. A clear vein pattern was observed in the center of typical t150 BMGs (position 5) in Fig. 3r. All these fracture morphologies reveal a novel deformation mechanism related to the controlled variation of free volume content in GMGs.  
+
+Atomistic mechanism of shear band deflection. Beyond these encouraging experimental results, we examined atomic- level details of the observed shear band deflection process in GMGs. Microscopically, the shear- banding process is controlled by shear transformation zone (STZ) percolation<sup>32</sup>. Owing to spatial and temporal confinement, detailed characterization of STZs in GMGs is impossible to probe experimentally. The molecular dynamic simulation method offers a powerful approach to explore the fundamental characteristics of shear band deflection and derive an atomistic description of the deflection mechanism. A typical gradient \(\mathrm{Cu}_{65}\mathrm{Zr}_{35}\) MG consisting of two regions with disparate amounts of free volume is created by randomly removing \(2\%\) of atoms in the right half of the simulated MG box - as shown in the inset in Fig. 4a, the right region (blue) of the gradient model is soft. Here, we selected Cu- Zr system as a prototype because of its high quality of potential, which has been well developed and frequently applied<sup>33</sup>, while no potential is available for \(\mathrm{Zr}_{58}\mathrm{Cu}_{22}\mathrm{Fe}_{8}\mathrm{Al}_{12}\) MG. Figure 4a depicts the variation of atomic volume along the Y- direction in the as- cast and gradient MG models. As- cast MG has almost no free volume fluctuation due to its relatively homogeneous structure, whereas GMG displays an obvious larger atomic volume value in the right part. Figure 4b depicts the calculated compressive stress- strain curves of the as- cast and gradient MGs. The calculated stress- strain curve of the relatively uniform as- cast MG with less free volume exhibits a yield strength and a distinct stress drop after about \(7\%\) strain, indicating substantial shear localization. In contrast, the stress drop is much less pronounced in the GMG. More interestingly, the GMG exhibits enhanced average flow stress in the \(10\% - 15\%\) strain region relative to the as- cast MG, which indicates enhanced plasticity in the GMG. This can be also deduced from the plastic deformation of as- cast and gradient MGs at \(9\%\) strain. As shown in Fig. 4c, the configurations of the shear bands observed in a GMG are very different from those in a relative uniform as- cast MG. For the as- cast MG, the shear
+
+<--- Page Split --->
+
+band almost penetrated the entire sample along the maximum shear plane with a rough straight line. In contrast, the propagation appeared to be changed when it penetrated the gradient- transition region in the GMG. It is apparent that the shear band angle becomes significantly larger in the right soft region due to the presence of a large free volume and a small friction coefficient. These results, obtained by simulations, suggest that gradient free volume content plays a vital role in deflecting the propagation of shear bands in the region with changed Mises strain distribution, ultimately giving rise to the improved plasticity of the glass, which is roughly consistent with the experimental observations.  
+
+What is more, the shear band is not only related to the percolation of STZs, it is also related to the consecutive activation of STZs based on successive strong strain (triggering STZs) and rotation fields (vortex- like) \(^{34 - 36}\) . Thus, the deflection behavior of shear banding can be greatly affected by non- uniform stress/strain fields. Variation of the free volume content across the gradient- transition region not only perturbs the strain distribution but also changes the rotation fields, which can be obtained by analyzing the rotational part of the deformation gradient tensor. As shown in Figure 4d, for the ascast MG, two rotation fields around the STZ can be clearly identified. The direction of rotation is clockwise (white color), along with the shear front, and anti- clockwise (black color) if one moves perpendicular to it. More specifically, the clockwise rotation fields are strongly connected and concentrated in one shear band, whereas the anti- clockwise rotation fields are nearly uniformly distributed throughout the whole sample. These observations are consistent with the STZ- vortex mechanism \(^{34}\) proposed by Sopu et al. They reported that STZs can induce collective vortex- like motions in the shear front. The vortex- like motions in turn act as a medium, triggering the activation of successive STZs, and finally cause the rapid propagation of the shear band. For the GMG, the rotation fields around the STZ in the left part are similar to those observed in the ascast MG. However, the rotation fields around STZs clearly change in the right soft region. STZ percolation follows a new specific direction with a larger angle. Previously accumulated clockwise rotation fields are discontinuous within the single shear band. Meanwhile, new clockwise STZ- vortex sequences are activated with a deviation from the maximum shear band plane. The anti- clockwise rotation fields become weaker and decrease around the main shear band. To better visualize the change in the rotation fields, the rotation fields across the shear bands at different positions in the GMG are depicted in Fig. 4e, which shows the variation in rotation angle corresponding to lines
+
+<--- Page Split --->
+
+1- 3 in Fig. 4d. Due to the accumulation of the clockwise STZ- vortex in the shear band, the symmetric profiles for the uniform as- cast MG (line 1) show a flat section with positive rotation angles. With the increased distance from the center of the shear band, the rotation angle significantly decreases to negative values, which corresponds to the anti- clockwise rotation fields. The anti- clockwise rotation fields obviously weaken in the gradient- transition region (line 2). The rotation angle displays only minor negative values away from the central shear band and varies between 6 degrees and 0 degrees. In the soft region (line 3), the rotation angle is almost the same in this line, and the anti- clockwise rotation fields almost vanish. The variation of the rotation fields across the shear band indicates that the STZ vortex- like motion mechanism characteristic in the GMG must be perturbed. This is due to the modification of the local strain field around the STZ caused by the gradient structure.  
+
+## Discussion  
+
+Figure 4f schematically shows the STZ percolation mechanism under compressive strain in GMG. In the left hard region, since the rotation field shows a quadrupolar- like distribution around the STZs when subjected to compressive stress, the compressive strain is oriented along the y- axis, while the tensile strain is oriented along the x- axis. The activated STZ can perturb the surrounding STZ by generating strong, clockwise rotation strain fields. These clockwise rotation strain fields can compress the STZ located at the top right of the vortex while stretching the STZ located at the bottom left of the vortex. Meanwhile, this activated STZ can also be perturbed by the anti- clockwise rotation strain fields generated by surrounding STZs. In this state, the anti- clockwise rotation strain fields can also drive the movement of the vortex with a special angle. In this way the following STZ vortex will be activated and the shear band will percolate in this direction. When the STZ vortex moves to the soft region, the weakening of the atomic bonding will perturb the strain field around the STZ. The tensile strain fields will become larger, which will govern the vortex to be closer to the tensile strain direction. By the same reasoning, the influence from the anti- clockwise rotation strain fields will become weak. Hence, the angle of the STZ percolation path changes to a larger value than that in the hard region. The non- activated STZ will be aligned with the new local strain fields and the shear band will percolate in alignment with this new angle - ultimately, the shear band deflected as we observed experimentally. Hence, our model demonstrates that local strain fields can be effectively tuned by the
+
+<--- Page Split --->
+
+free volume gradient, and thus the STZ- vortex motion and shear banding behavior can be controlled. It is still desirable to optimize the gradient structure and strain fields to more effectively control the propagation of shear bands.  
+
+In summary, we have proposed a novel gradient BMG design strategy through cyclic heat treatment. The resulting substantial bulk scale structural gradient allows for extra plastic strain without sacrificing ultrahigh strength. Both experimental and computational evidence has demonstrated the importance of covering the whole structure with a tunable free volume gradient for the deflection of shear deformation. We have shown that the deflection of the shear band is accompanied by a gradient change in the vortex motion of the shear transformation zone, which is regulated by the variation of the non- uniform strain fields in the GMG. Our research highlights the potential of creating gradient structured engineering materials with high strength and plastic deformability.  
+
+## Methods:  
+
+Materials preparation. MGs with the atomic components of \(\mathrm{Zr_{58}Cu_{22}Fe_{8}Al_{12}}\) were prepared by arc melting of a mixture of pure elements (99.9% purity) in an argon atmosphere and injection casting into a copper moldwith a diameter of \(2\mathrm{mm}\) . For cryogenic thermal cycling, one cycle consisted of dipping the sample into liquid nitrogen for a certain period of time (6 s, \(25\mathrm{s}\) , \(70\mathrm{s}\) , and \(150\mathrm{s}\) ), followed by transferring it into hot water ( \(T = 323\mathrm{K}\) ) for the same time (6 s, \(25\mathrm{s}\) , \(70\mathrm{s}\) , and \(150\mathrm{s}\) ). All the samples were treated for 10 cycles.  
+
+Characterization and mechanical testing. Density measurements, based on the Archimedes method, were conducted using a high precision balance with an accuracy of \(\pm 0.01\mathrm{mg}\) . They were repeated at least 15 times to ensure data reliability. Vickers microhardness was measured on the cross- section using a Matsuzawa MMT- X indentation machine with a load of \(1.96\mathrm{N}\) and holding for \(15\mathrm{s}\) . The microstructure of the samples was examined by TEM using a Cs- corrected FEI Titan G2 80- 200 ChemiSTEM. Specimens for TEM were cut from the bulk MG specimens and thinned using a dual- beam FIB system (FEI Helios Nano- Lab 600i). Uniaxial compression tests were performed in an Instron 5982 mechanical testing machine at a strain rate of \(1\times 10^{- 4}\mathrm{s}^{- 1}\) at room temperature. The specimens, \(2\mathrm{mm}\) in diameter and \(4\mathrm{mm}\) in height with two ends, were carefully polished to ensure parallelism. The fracture features and surface morphology of all the samples after compression failure were investigated by
+
+<--- Page Split --->
+
+scanning electron microscopy on a Zeiss Sigma500 field emission gun (FEG) SEM. The height maps of the fracture surface of all the samples were measured by the Dektak XT profile measurement.  
+
+Molecular dynamics simulations. Molecular dynamics simulations were performed using a Large- scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) \(^{37}\) with newly- developed Cu- Zr embedded atom method (EAM) potential \(^{38}\) . A cubic box containing 2213750 atoms was constructed with three- dimensional (3D) periodic boundary conditions (PBCs) as an initial model. The atoms were randomly arranged, and the system was melted by maintaining a high temperature of 2000 K for 8 ns with zero external pressure under the NPT ensemble. The melted model was then quenched at a constant cooling rate of \(10^{11} \mathrm{~K} / \mathrm{s}\) from 2000 K to 50 K and relaxed at 50 K for 2 ns to equilibrate the structure. The final size of the sample was about \(120 (\mathrm{x}) \times 60 (\mathrm{y}) \times 5 (\mathrm{z}) \mathrm{nm}^{3}\) . Afterwards, the gradient MG was prepared by randomly removing \(2\%\) of atoms in the right half of the model (along the Y direction). Once the gradient structure was generated, it was relaxed for 50 ps. The obtained gradient MG was further compressed along the X- direction at 50 K under a strain rate of \(1 \times 10^{8} \mathrm{~s}^{- 1}\) . The atomic- scale shear banding process was monitored by the atomic shear strain, \(\eta^{\mathrm{Mises39}}\) and rotation field by Ovito \(^{40}\) .  
+
+## References:  
+
+[1] Ritchie, R. O. The conflicts between strength and toughness. Nat Mater 10, 817–822 (2011). [2] Launey, M. E. & Ritchie, R. O. On the fracture toughness of advanced materials. Adv. Mater. 21, 2103–2110 (2009). [3] Wu, X. et al. Extraordinary strain hardening by gradient structure. Proc. Natl. Acad. Sci. U.S.A. 111, 7197 (2014). [4] Wei, Y. J. et al. Evading the strength- ductility trade- off dilemma in steel through gradient hierarchical nanotwinsmade pioneering contributions to the mechanics of fracture, dislocations, and stress effects in solid materials. Nat. Commun. 5, 3580 (2014). [5] Chen, W. et al. Mechanically- induced grain coarsening in gradient nano- grained copper. Acta Mater. 125, 255 (2017). [6] Zhao, S. T. et al. Generating gradient germanium nanostructures by shock- induced
+
+<--- Page Split --->
+
+amorphization and crystallization. Proc. Natl. Acad. Sci. U.S.A. 114, 37 (2017).  
+
+[7] Lu, K. Making strong nanomaterials ductile with gradients. Science 345, 1455 (2014).  
+
+[8] Li, X. Y. et al. Mechanical properties and deformation mechanisms of gradient nanostructured metals and alloys. Nat. Rev. Mater. 5, 706 (2020).  
+
+[9] Liu, Z. Q. et al. Functional gradients and heterogeneities in biological materials: Design principles, functions, and bioinspired applications. Prog. Mater. Sci. 88, 467 (2017).  
+
+[10] Espinosa, H. D. et al. Merger of structure and material in nacre and bone- Perspectives on de novo biomimetic materials. Prog. Mater. Sci. 54, 1059 (2009).  
+
+[11] Studart, A. R. Additive manufacturing of biologically- inspired materials. Chem. Soc. Rev. 45, 359 (2016).  
+
+[12] Meyers, M. A. et al. Biological materials: Structure and mechanical properties. Prog. Mater. Sci. 53, 1 (2008).  
+
+[13] Mayer, G. Rigid biological systems as models for synthetic composites. Science 310, 1144 (2005).  
+
+[14] Fang, T. et al. Revealing Extraordinary Intrinsic Tensile Plasticity in Gradient Nano- Grained Copper. Science 331, 1587 (2011).  
+
+[15] Cheng, Z. et al. Extra strengthening and work hardening in gradient nanotwinned metals. Science 362, 559 (2018).  
+
+[16] Liu, C. et al. Nano- dual- phase metallic glass film enhances strength and ductility of a gradient nanograined magnesium alloy. Adv. Mater. 7, 2001480 (2020).  
+
+[17] Zberg, B. et al. MgZnCa glasses without clinically observable hydrogen evolution for biodegradable implants. Nat Mater 8, 887 (2009).  
+
+[18] Lin, Y. et al. Mechanical properties and optimal grain size distribution profile of gradient grained nickel. Acta Mater. 153, 279 (2018).  
+
+[19] Meng, X. et al. Underwater Mechanically Robust Oil- Repellent Materials: Combining Conflicting Properties Using a Heterostructure. Adv. Mater. 30, 1706634 (2018).  
+
+[20] Yang, Y. et al. Electrically assisted 3D printing of nacre- inspired structures with self- sensing capability. Sci. Adv. 5, 9490 (2019).  
+
+[21] Li, J. et al. On strain hardening mechanism in gradient nanostructures. Int. J. Plast. 88, 89 (2017).
+
+<--- Page Split --->
+
+[22] Yang, D. K. et al. Work hardening in ultrafine- grained titanium: Multilayering and grading. Acta Mater. 61, 2840 (2013).  
+
+[23] Xu, W. et al. Strain- induced microstructure refinement in pure Al below 100 nm in size. Acta Mater. 152, 138- 147 (2018).  
+
+[24] Laine, S. J. et al. Microstructural characterisation of metallic shot peened and laser shock peened Ti- 6Al- 4V. Acta Mater. 123, 350 (2017).  
+
+[25] Ma, X. L. et al. Mechanical properties of copper/bronze laminates: Role of interfaces. Acta Mater. 116, 43 (2016).  
+
+[26] Greer, A. L. et al. Shear bands in metallic glasses. Mater. Sci. Eng. R. 74, 71 (2013).  
+
+[27] Zhang, Z. F. et al. Difference in compressive and tensile fracture mechanisms of Zr59Cu20Al10Ni8Ti3 bulk metallic glass. Acta Mater. 51, 1167 (2003).  
+
+[28] Zhang, Z. F. et al. Fracture mechanisms in bulk metallic glassy materials. Phys. Rev. Lett. 91, 045505 (2003).  
+
+[29] Hsueh, C. - H. et al. Shear fracture of bulk metallic glasses with controlled applied normal stresses. Scr. Mater. 59, 111 (2008).  
+
+[30] Baricco, M. et al. Glass formation and mechanical properties of (Cu50Zr50)100- xAlx (x=0, 4, 5, 7) bulk metallic glasses. J. Alloys Compd. 483, 125 (2009).  
+
+[31] Ketov, S. V. et al. Rejuvenation of metallic glasses by non- affine thermal strain. Nature 524, 200 (2015).  
+
+[32] Schuh, C. A. & Lund, A. C. Atomistic basis for the plastic yield criterion of metallic glass. Nat. Mater. 2, 449 (2003).  
+
+[33] Wu, Z. W. et al. Correlation between structural relaxation and connectivity of icosahedral clusters in CuZr metallic glass- forming liquids. Phys. Rev. B 88, 054202 (2013).  
+
+[34] Sopu, D. et al. Atomic- level processes of shear band nucleation in metallic glasses. Phys. Rev. Lett. 119, 195503 (2017).  
+
+[35] Sopu, D. et al. Atomic- scale origin of shear band multiplication in heterogeneous metallic glasses. Scr. Mater. 178, 57 (2020).  
+
+[36] Scudino, S. & Sopu, D. Strain distribution across an individual shear band in real and simulated metallic glasses. Nano Lett. 18, 1221 (2018).  
+
+[37] Mendelev, M. I. et al. Using atomistic computer simulations to analyze x- ray diffraction data
+
+<--- Page Split --->
+
+from metallic glasses. J. Appl. Phys. 102, 043501 (2007).  
+
+[38] Plimpton, S. Fast parallel algorithms for short- range molecular dynamics. J. Comput. Phys. 117, 1 (1995).  
+
+[39] Shimizu, F. et al. Theory of shear banding in metallic glasses and molecular dynamics calculations. Mater. Trans. 48, 2923 (2007).  
+
+[40] Stukowski, A. Visualization and analysis of atomistic simulation data with OVITO- the Open Visualization Tool. Model. Simulat. Mater. Sci. Eng. 18, 015012 (2010).
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1. Schematic description of gradient BMG. (a) Schematic representation of one dominant shear band in uniform MG. (b) Schematic diagram of proposed GMG. (c) Proposed shear band deflection mechanism in a GMG during the uniaxial compression process. </center>
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2. Generation and Characterization of the GMG. (a) Schematic of the GMG by means of CTC. (b) Density values as a function of holding time for the treated MGs. (c) Cryogenic thermal cycling procedure from a high temperature (323 K) to liquid nitrogen temperature (77 K), together with the waiting time, \(t\) , at both the maximum and the minimum temperatures. (d) Variation of average hardness value along the distance from the center. (e) The top shows the method for hardness measurements along various circles. The bottom shows the t150 sample for TEM. (f)-(h) TEM images with the corresponding selected-area-diffraction patterns (SAED) of the edge, middle, and center regions (labeled A, B, C) for the t150 sample. (i) Corresponding radial distribution functions calculated from the SAED of the edge, middle, and center regions (labeled A, B, C) in the t150 sample. </center>
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3. Enhanced mechanical properties, fracture surface change and shear band deflection. (a) Compressive stress-strain curves for the as-cast and treated MGs. (b) Variation of the plastic strain with the structural gradient. (c)-(g) Lateral morphologies of the fractured as-cast and gradient MG samples. (h)-(l) The corresponding 3D contours of the fracture surfaces. (m) Height variation profiles of the middle dash line (h) along the shear band plane for as-cast and gradient MG samples. (n)-(s) SEM surface morphologies at positions 1-3 in the as-cast (c) and 4-6 in the t150 sample (g). </center>
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4. Investigation of the deformation behavior of GMG by MD simulations. (a) Atomic free volumes as the functions of the position along the Y-direction in as-cast and gradient MGs. Inset shows an illustration of the design of GMG. (b) Representative stress-strain results for the as-cast and gradient MGs during compression along the X-direction. (c) The spatial distribution of atomic Mises strain (d) Rotation angle of the as-cast and gradient MGs at \(9\%\) compression strain. (e) The variations of the rotation angle along lines 1-3 in (d). (f) Schematic illustration of the STZ percolation mechanism in GMG. </center>
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1 </center>  
+
+Schematic description of gradient BMG. (a) Schematic representation of one dominant shear band in uniform MG. (b) Schematic diagram of proposed GMG. (c) Proposed shear band deflection mechanism in a GMG during the uniaxial compression process.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2 </center>  
+
+Generation and Characterization of the GMG. (a) Schematic of the GMG by means of CTC. (b) Density values as a function of holding time for the treated MGs. (c) Cryogenic thermal cycling procedure from a high temperature (323 K) to liquid nitrogen temperature (77 K), together with the waiting time, t, at both the maximum and the minimum temperatures. (d) Variation of average hardness value along the distance from the center. (e) The top shows the method for hardness measurements along various circles. The bottom shows the t150 sample for TEM. (f)-(h) TEM images with the corresponding selected- area- diffraction patterns (SAED) of the edge, middle, and center regions (labeled A, B, C) for the t150 sample. (i) Corresponding radial distribution functions calculated from the SAED of the edge, middle, and center regions (labeled A, B, C) in the t150 sample.
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3 </center>  
+
+Enhanced mechanical properties, fracture surface change and shear band deflection. (a) Compressive stress- strain curves for the as- cast and treated MGs. (b) Variation of the plastic strain with the structural gradient. (c)- (g) Lateral morphologies of the fractured as- cast and gradient MG samples. (h)- (l) The corresponding 3D contours of the fracture surfaces. (m) Height variation profiles of the middle dash line (h) along the shear band plane for as- cast and gradient MG samples. (n)- (s) SEM surface morphologies at positions 1- 3 in the as- cast (c) and 4- 6 in the t150 sample (g).
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4 </center>  
+
+Investigation of the deformation behavior of GMG by MD simulations. (a) Atomic free volumes as the functions of the position along the Y- direction in as- cast and gradient MGs. Inset shows an illustration of the design of GMG. (b) Representative stress- strain results for the as- cast and gradient MGs during compression along the X- direction. (c) The spatial distribution of atomic Mises strain (d) Rotation angle
+
+<--- Page Split --->
+
+of the as-cast and gradient MGs at \(9\%\) compression strain. (e) The variations of the rotation angle along lines 1-3 in (d). (f) Schematic illustration of the STZ percolation mechanism in GMG.
+
+<--- Page Split --->

preprint/preprint__7e519e008d15c7e10d8d8f8234fc8066280b0cc87c77ab82d8b32e00f8df0572/preprint__7e519e008d15c7e10d8d8f8234fc8066280b0cc87c77ab82d8b32e00f8df0572_det.mmd

@@ -0,0 +1,330 @@
+<|ref|>title<|/ref|><|det|>[[44, 106, 952, 175]]<|/det|>
+# Extra plasticity governed by shear band deflection in gradient metallic glasses  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 733, 238]]<|/det|>
+Yao Tang ( \(\boxed{\pm}\) tangyao@zju.edu.cn) International Center for New- Structured Materials (ICNSM), Zhejiang University  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 582, 285]]<|/det|>
+Haofei Zhou Zhejiang University https://orcid.org/0000- 0001- 9226- 9530  
+
+<|ref|>text<|/ref|><|det|>[[44, 290, 225, 331]]<|/det|>
+Xiaodong Wang Zhejiang University  
+
+<|ref|>text<|/ref|><|det|>[[44, 337, 225, 377]]<|/det|>
+Qingping Cao Zhejiang University  
+
+<|ref|>text<|/ref|><|det|>[[44, 383, 225, 424]]<|/det|>
+Dongxian Zhang Zhejiang University  
+
+<|ref|>text<|/ref|><|det|>[[44, 430, 225, 470]]<|/det|>
+Wei Yang Zhejiang University  
+
+<|ref|>text<|/ref|><|det|>[[44, 475, 225, 516]]<|/det|>
+Jian- Zhong Jiang Zhejiang University  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 558, 102, 575]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 593, 905, 637]]<|/det|>
+Keywords: Controlled Structural Gradients, Heat Treatment Engineering Protocol, Cryogenic Thermal Cycling  
+
+<|ref|>text<|/ref|><|det|>[[44, 654, 285, 674]]<|/det|>
+Posted Date: April 5th, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 692, 463, 712]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 366951/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 729, 910, 773]]<|/det|>
+License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 808, 910, 852]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on April 19th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29821- 4.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[168, 84, 828, 122]]<|/det|>
+# Extra plasticity governed by shear band deflection in gradient metallic glasses  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 223, 226, 239]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[147, 253, 852, 633]]<|/det|>
+Inspired by gradient materials in nature, advanced engineering components with controlled structural gradients have attracted significant research interest due to their exceptional combinations of properties. However, it remains challenging to generate structural gradients that penetrate through bulk materials, which is essential for achieving enhanced mechanical properties in metallic materials. Here, we propose a heat treatment engineering protocol to realize a controllable structural gradient in bulk metallic glasses (BMGs). By adjusting the holding time of cryogenic thermal cycling, a series of BMGs with gradient- distributed free volume contents from internal to external can be synthesized. Both mechanical testing and atomistic simulations demonstrate that the spatial gradient can endow BMGs with extra plasticity without sacrificing their ultrahigh strength. Such an enhanced mechanical property is governed by the gradient- induced deflection of shear deformation that fundamentally suppresses the unlimited shear localization on a straight plane that would be expected in BMGs without such a gradient.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 80, 852, 820]]<|/det|>
+Advances in modern science and technology continue to impose more stringent requirements for engineering materials, including exceptional strength and toughness. Unfortunately, these two properties are almost mutually exclusive in monolithic materials \(^{1,2}\) . Obtaining optimal mechanical performance is always a compromise, one which can be achieved by optimizing the microstructure through empirical design. Notably, the introduction of structural gradients can overcome the strength- ductility trade- off in metallic materials and give rise to high- performance functionalities \(^{3 - 8}\) . Concerning such gradients, nature provides a rich source of inspiration. Many natural materials have highly sophisticated structures with complex gradient designs that possess extremely impressive combinations of properties significantly surpassing those of their constituents \(^{9 - 13}\) . In view of the gradient structures of natural materials, exploring structural gradients to enhance the properties of engineering materials has generated strong interest. Typical examples are widely- exploited gradient metals with nano- grained (NG) \(^{14}\) or nano- twinned (NT) structures. \(^{15}\) In contrast to conventional homogeneous coarse- grained (CG) materials, the deformation mechanism of gradient nanostructured (GNS) materials is often heterogeneous and is regulated and constrained by the gradient structure. Also, structural gradients typically cause stress gradients and even activate new dislocation structures \(^{8}\) . Nevertheless, current GNS materials are limited to a few, pure face- centered- cubic metals and typical alloys. For example, Mg alloys can be strengthened by introducing a gradient nanograined structure while this strategy is unable to provide large ductility in Mg alloys. Recently, an Mg- based nano dual- phase metallic glass (NDP) coated on a gradient nano- grained Mg alloy showed enhanced ductility and yield strength compared to the base alloy \(^{16}\) . The success of this design strategy of combining heterogeneous metallic glass (MG) and gradient nanograined structure provides us a motivation to extend the principles of structural gradients to amorphous systems in designing ‘intrinsic’ gradient MGs (GMGs). Indeed, MGs with extraordinary physical and biomaterial properties have recently been developed \(^{17}\) , but severe brittleness holds one major weakness that precludes the wide application of MGs. The introduction of spatial gradients may offer a promising solution for tuning deformation behavior and enhancing the plasticity of MGs.  
+
+<|ref|>text<|/ref|><|det|>[[148, 821, 851, 915]]<|/det|>
+Over the past several years, various fabrication methods have been applied to develop structural gradients in engineering materials. The fabrication methods can be divided into two categories: bottom- up methods, including physical and chemical deposition \(^{18}\) , layer- by- layer assembly \(^{19}\) , and three- dimensional printing \(^{20}\) ; and top
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 83, 852, 352]]<|/det|>
+down methods including surface mechanical treatment methods<sup>21- 23</sup>, laser shock peening<sup>24</sup>, and roll bonding<sup>25</sup>. Despite their widespread use in engineering design, these methods suffer from marked constraints. Bottom- up methods are generally only feasible for making thin films or microscopic samples. Existing top- down methods, on the other hand, have limits for the range of bulk gradient materials. For instance, surface mechanical treatments always produce a limited volume fraction of gradients only near the surface, or they generate a negligible degree of structural gradients along the gradient direction. All of the aforementioned issues limit our ability to achieve a gradient throughout the bulk metallic glass (BMG) samples. It is essential to develop new strategies and practical methods to design and fabricate gradient BMGs to tailor their mechanical properties.  
+
+<|ref|>text<|/ref|><|det|>[[147, 358, 852, 620]]<|/det|>
+In this paper, we report a design strategy and a set of simple fabrication methods to produce the GMG in bulk form by introducing a controllable spatial gradient of the free volume content. Through experiments and MD simulations, we demonstrate that the excellent performance of GMG can be attributed to its "shear band deflection" capability that arises from its intrinsic gradient structure. A notable difference in the local free volume defects the angle of the shear band initiation and propagation. Using model heterogeneous materials, we discuss the atomic- scale origin of the observed variations in the SB dynamics and the angle with changing structural state. The design strategy offers a simple and yet versatile method to improve the mechanical properties of BMGs and, more importantly, to design new generations of high- performance structural materials.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 654, 215, 670]]<|/det|>
+## Results  
+
+<|ref|>text<|/ref|><|det|>[[147, 677, 852, 891]]<|/det|>
+The design strategy for gradient BMG. The design strategy for the GMG structure is proposed in Fig. 1. The plastic deformation of uniform BMGs is through shear localization into narrow bands (Fig. 1a). Such localization often leads to the running away of one dominant shear band, eventually leading to catastrophic failure and macroscopic brittle behavior<sup>26</sup>. The shear band plane in BMGs occurs along an angle at which the corresponding effective shear stress is maximized, which suggests the important influence of normal stress on the shear plane<sup>27,28</sup>. The normal stress effect on deformation in MG lies in the principle of atomistic friction, as embodied in the Mohr- Coulomb criterion:
+
+<--- Page Split --->
+<|ref|>equation<|/ref|><|det|>[[444, 87, 553, 105]]<|/det|>
+\[\tau_{y} = \tau_{0} - \alpha \sigma_{n}\]  
+
+<|ref|>text<|/ref|><|det|>[[148, 116, 850, 165]]<|/det|>
+where, \(\tau_{y}\) is the effective shear yield stress, \(\tau_{0}\) is a constant, and \(\alpha\) is an effective coefficient of friction that controls the strength of the normal stress effect<sup>29,30</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[147, 171, 853, 610]]<|/det|>
+We propose that the plasticity of BMGs can be enhanced through a gradient design of the microstructure, with the free volume concentration increasing from the outer to the inner part of the cylindrical BMG specimen (Fig. 1b). Changes in the free volume content and bonding conditions of structural units are expected to modify friction coefficient \(\alpha\) and therefore shear band angle \(\theta\) . Figure 1c shows schematic illustrations of the development of shear bands in such a gradient BMG specimen. The primary shear band initiates at the upper- left surface with a relatively lower content of free volumes, corresponding to a relatively larger friction coefficient \(\alpha\) . Taking normal stress into consideration, the effective shear yield stress is maximized at shear band angle \(\theta\) for the local hard region. As the shear band progresses toward the central soft region of the specimen, the increasing value of free volume concentration alters the normal stress effect on the shear band, inducing a gradual increase in the shear band angle. The shear band is thus deflected by the structural gradient of the BMG. Afterwards, as the shear band propagates from the center to the lower- right surface, free volume concentration declines, leading to a gradual decrease in the shear band angle and a reversed deflection pattern of the shear band. In short, the deflected shear band path avoids the, otherwise straight, transecting shear band across the whole sample, which provides a promising route for improving the plastic deformability of BMGs.  
+
+<|ref|>text<|/ref|><|det|>[[147, 639, 852, 904]]<|/det|>
+Characteristics of the GMG. To test this design strategy, we selected a relatively brittle \(\mathrm{Zr_{58}Cu_{22}Fe_{8}Al_{12}}\) MG to construct the gradient BMG structure. We used cryogenic thermal cycling between a certain temperature (323 K) and cryogenic temperature (77 K) to prepare the gradient samples. Figure 2a shows a detailed description of the cryogenic thermal cycling treatment apparatus used to introduce a gradient rejuvenation into the cylindrical MG samples. When the sample was first heated at a certain temperature and then cooled to a low temperature, the rejuvenation behavior came first, which is mainly related to the quasi- localized vibrations of atoms in the flow unit surrounded by the elastic matrix. The sample must be held for a sufficiently long time to enhance its rejuvenation effect. With a long holding time at high temperature, the atoms within flow units will move cooperatively and reversibly on a large scale, and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 83, 851, 176]]<|/det|>
+resulting in a fast relaxation in turn. Therefore, it can be expected that dynamic rejuvenation or relaxation behaviors vary at internal and external parts of the sample during their evolution with time. Controlling the holding time can induce gradient rejuvenation (free volume content) processes.  
+
+<|ref|>text<|/ref|><|det|>[[147, 183, 853, 594]]<|/det|>
+Figure 2b shows the results of density measurements of the as- cast sample and the treated samples. The treated samples refer to ten- cycled samples using the same thermal cycling process but with different holding times, as shown in Fig. 2c. The density of the treated samples is lower than that of the as- cast samples, which suggests a relatively large free volume content. The density of the treated sample decreases with the increased holding time but seemingly saturates when \(t\) is larger than about 60 s. Figure 2d shows the variation of hardness across the diameter on a cross- section of the \(2\mathrm{mm}\) cylindrical as- cast and treated samples (the top of Fig. 2e shows the method for hardness measurements along various circles, in which 8 indentations were performed to acquire an average harness value at each circle, together with 80 indentations from the center to the edge). For the t6 sample (treated with a 6 s holding time), the hardness exhibits little decrease from a distance of \(0.3\mathrm{mm}\) to the center. Notably, a gradient of the hardness value can be detected for the t25 sample. From the edge to the center, the hardness value of the t25 sample tends to decrease from 500 HV to 485 HV, respectively. In particular, a more obvious hardness- value gradient can be seen for the t70 and t150 samples. The gradient hardness values suggest that this method can provide a potentially low- cost manufacturing process for the scalable production of GMGs.  
+
+<|ref|>text<|/ref|><|det|>[[147, 601, 853, 913]]<|/det|>
+Transmission electron microscopy (TEM) analyses were performed to characterize the amorphous structure at different positions of the t150 GMG sample (bottom of Fig. 2e). Specifically, Figure 2f- h display the TEM images of edge, middle, and center regions (labeled A, B, and C) for the t150 sample. One can clearly see a grain- like microstructure with a dark- bright contrast in the sample at the edge. Of significant interest, for the sample in the middle, the sizes of the dark and bright regions are enlarged to 2–3 nm. The characteristic length of the inhomogeneous microstructure reaches 5 nm in the sample at the center. Considering that our TEM results show an increasing heterogeneity with a decreasing distance from the center, the hypothesis suggested by Ketov et al. provides a reasonable explanation for the gradient amorphous microstructures<sup>31</sup>. Ketov et al. attribute the rejuvenation effects to intrinsic non- uniformity of the glass structure, which gives a non- uniform coefficient of thermal expansion. The population and intensity of soft spots (dark), with lower elastic stiffness
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 83, 852, 398]]<|/det|>
+and higher CTE, increases with cycling. Hence, the brighter contrast in our TEM images at the center region may result from a lower density zone, while the edge region has a relatively high density in our treated samples. Following the corresponding selected area electron diffraction (SAED) patterns (insets of Fig. 2f- h), we confirm that the observed microstructures are all structurally amorphous. The radial distribution functions (RDFs, Fig. 2i), which were calculated from the SAED patterns of three parts in the t150 sample, have differences in their peak positions. It was found that the first peak position in RDFs is shifted to higher r values from the edge to the center of the treated sample, i.e., the average atomic bond distances increased from edge to center. All observed experimental results - i.e., at the center with a reduction in hardness, larger bright contrast in TEM images, and larger average atomic bond distance - reveal an enhancement of free volume in the center part. Therefore, gradient structure (or rejuvenation) indeed occurs from edge to center in the treated samples.  
+
+<|ref|>text<|/ref|><|det|>[[147, 428, 853, 791]]<|/det|>
+Mechanical behavior and deformation mechanisms of GMG. To demonstrate the enhanced plastic deformability of gradient BMGs, we compared the engineering compressive stress- strain curves of the gradient BMG samples with the as- cast sample (Fig. 3a). The as- cast sample undergoes negligible plastic strain before fracture, typical of the strong- and- brittle behavior reported in the literature. Of importance, plasticity strongly increased significantly without the expense to the strength in the gradient samples, reaching maximum when the holding time was 70 s and 150 s. For comparison, Fig. 3b summarizes variations of measured plastic strain as a function of the structural gradient. The structural gradient is a parameter that quantifies the structure difference of cryogenically treated samples as the change in hardness per unit thickness along the gradient direction. As illustrated in Fig. 3b, we observed an increase in plastic strain with the increase of the structural gradient. The samples with a structural gradient of 16 HV/mm exhibited the largest plastic strain, about four times that of the as- cast sample. The above observations strongly suggest that a substantial plasticity increment can be achieved solely by introducing the structural gradient.  
+
+<|ref|>text<|/ref|><|det|>[[148, 797, 852, 914]]<|/det|>
+To uncover the physical mechanisms underlying ductility enhancement in the GMGs, we explored the key structural parameters that are affected by the structural gradient. Figure 3c- g display the lateral morphologies of the as- cast and GMG samples obtained by SEM after the final failure. One dominating primary shear band plane, a typical failure mode observed in brittle MGs, exists in the as- cast sample (Fig. 3c). In
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 80, 853, 917]]<|/det|>
+contrast, the GMG samples demonstrate interesting fracture surface morphology. For the t6 sample (Fig. 3d), the fracture surface looks somewhat uneven. A small bulge on the fracture surface can be seen for the t25 sample. For the t70 and t150 samples (Fig. 3f and g), some humps can be clearly observed in the center region of the fracture surface. These fracture surfaces in Fig. 3c-g indicate that the shear band plane deflects obviously from the original shear planes when the shear bands propagate into the center of the GMG samples. To accurately characterize these fracture surfaces, three- dimensional (3- D) profiles of the fracture surfaces are also displayed in Fig. 3h- l. Ridges running parallel to the shear band plane direction in the middle (see line in Fig. 3h) are connected by a set of transverse ridges. It can be seen that the ridges in the fracture surface of the as- cast and t6 samples are almost flat. For the t25 sample, the branching of the ridges as well as their meandering in different directions as the shear plane front advances can be noted. An obvious hump can be observed in the height variation along the shear band plane from top to bottom of the fracture surfaces in the t70 and t150 samples. The shear band plane looks like a non- uniform surface with a bent shape, which means that the fracture angle changes at a particular stage during shear band propagation. Distributions of ridge heights, obtained from 2- D profiles of fracture surfaces along the dash line (Fig. 3h), are shown in Fig. 3m. The relative homogeneous as- cast sample only shows small height fluctuation. In contrast, the GMG samples exhibit marked height differences, i.e. lower on both sides and higher in the middle. From the calculated fracture angle at each point along this variation line, the fracture angle in the center is larger than that in the outer part, which suggests that shear band deflection indeed occurs during deformation. It is apparent that this is direct evidence that a novel non- uniform deformation mechanism in GMG occurred. The structural gradient exerts a direct influence on the shear banding, and the deflection of the shear band is thought to be closely related to the variation of the free volume content. In the edge region near the fracture origin of samples, a larger \(\alpha\) value might be expected, and consequently a larger effect of normal stress. In this case, the more difficult is the expected movement among structural units. The material also shows brittle behavior - hence the small angle of the shear band plane. On the other hand, a smaller \(\alpha\) might be associated with an increase in free volume content by gradient rejuvenation in the center part, leading to easier plastic flow and a larger angle of shear band plane. Our SEM images of the fracture surface morphology verify the above- mentioned analyses. Fig. 3n- s clearly show the morphologies of fracture as- cast and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 83, 852, 249]]<|/det|>
+t150 samples at different positions. The three positions of the as- cast sample exhibit typical viscous, river- like patterns along the shear direction with very narrow spacing. Distinct surface morphologies developed in the center regions of the two samples, which indicates a difference in the local free volume content. A clear vein pattern was observed in the center of typical t150 BMGs (position 5) in Fig. 3r. All these fracture morphologies reveal a novel deformation mechanism related to the controlled variation of free volume content in GMGs.  
+
+<|ref|>text<|/ref|><|det|>[[146, 275, 852, 917]]<|/det|>
+Atomistic mechanism of shear band deflection. Beyond these encouraging experimental results, we examined atomic- level details of the observed shear band deflection process in GMGs. Microscopically, the shear- banding process is controlled by shear transformation zone (STZ) percolation<sup>32</sup>. Owing to spatial and temporal confinement, detailed characterization of STZs in GMGs is impossible to probe experimentally. The molecular dynamic simulation method offers a powerful approach to explore the fundamental characteristics of shear band deflection and derive an atomistic description of the deflection mechanism. A typical gradient \(\mathrm{Cu}_{65}\mathrm{Zr}_{35}\) MG consisting of two regions with disparate amounts of free volume is created by randomly removing \(2\%\) of atoms in the right half of the simulated MG box - as shown in the inset in Fig. 4a, the right region (blue) of the gradient model is soft. Here, we selected Cu- Zr system as a prototype because of its high quality of potential, which has been well developed and frequently applied<sup>33</sup>, while no potential is available for \(\mathrm{Zr}_{58}\mathrm{Cu}_{22}\mathrm{Fe}_{8}\mathrm{Al}_{12}\) MG. Figure 4a depicts the variation of atomic volume along the Y- direction in the as- cast and gradient MG models. As- cast MG has almost no free volume fluctuation due to its relatively homogeneous structure, whereas GMG displays an obvious larger atomic volume value in the right part. Figure 4b depicts the calculated compressive stress- strain curves of the as- cast and gradient MGs. The calculated stress- strain curve of the relatively uniform as- cast MG with less free volume exhibits a yield strength and a distinct stress drop after about \(7\%\) strain, indicating substantial shear localization. In contrast, the stress drop is much less pronounced in the GMG. More interestingly, the GMG exhibits enhanced average flow stress in the \(10\% - 15\%\) strain region relative to the as- cast MG, which indicates enhanced plasticity in the GMG. This can be also deduced from the plastic deformation of as- cast and gradient MGs at \(9\%\) strain. As shown in Fig. 4c, the configurations of the shear bands observed in a GMG are very different from those in a relative uniform as- cast MG. For the as- cast MG, the shear
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 83, 852, 298]]<|/det|>
+band almost penetrated the entire sample along the maximum shear plane with a rough straight line. In contrast, the propagation appeared to be changed when it penetrated the gradient- transition region in the GMG. It is apparent that the shear band angle becomes significantly larger in the right soft region due to the presence of a large free volume and a small friction coefficient. These results, obtained by simulations, suggest that gradient free volume content plays a vital role in deflecting the propagation of shear bands in the region with changed Mises strain distribution, ultimately giving rise to the improved plasticity of the glass, which is roughly consistent with the experimental observations.  
+
+<|ref|>text<|/ref|><|det|>[[147, 301, 852, 916]]<|/det|>
+What is more, the shear band is not only related to the percolation of STZs, it is also related to the consecutive activation of STZs based on successive strong strain (triggering STZs) and rotation fields (vortex- like) \(^{34 - 36}\) . Thus, the deflection behavior of shear banding can be greatly affected by non- uniform stress/strain fields. Variation of the free volume content across the gradient- transition region not only perturbs the strain distribution but also changes the rotation fields, which can be obtained by analyzing the rotational part of the deformation gradient tensor. As shown in Figure 4d, for the ascast MG, two rotation fields around the STZ can be clearly identified. The direction of rotation is clockwise (white color), along with the shear front, and anti- clockwise (black color) if one moves perpendicular to it. More specifically, the clockwise rotation fields are strongly connected and concentrated in one shear band, whereas the anti- clockwise rotation fields are nearly uniformly distributed throughout the whole sample. These observations are consistent with the STZ- vortex mechanism \(^{34}\) proposed by Sopu et al. They reported that STZs can induce collective vortex- like motions in the shear front. The vortex- like motions in turn act as a medium, triggering the activation of successive STZs, and finally cause the rapid propagation of the shear band. For the GMG, the rotation fields around the STZ in the left part are similar to those observed in the ascast MG. However, the rotation fields around STZs clearly change in the right soft region. STZ percolation follows a new specific direction with a larger angle. Previously accumulated clockwise rotation fields are discontinuous within the single shear band. Meanwhile, new clockwise STZ- vortex sequences are activated with a deviation from the maximum shear band plane. The anti- clockwise rotation fields become weaker and decrease around the main shear band. To better visualize the change in the rotation fields, the rotation fields across the shear bands at different positions in the GMG are depicted in Fig. 4e, which shows the variation in rotation angle corresponding to lines
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 83, 852, 373]]<|/det|>
+1- 3 in Fig. 4d. Due to the accumulation of the clockwise STZ- vortex in the shear band, the symmetric profiles for the uniform as- cast MG (line 1) show a flat section with positive rotation angles. With the increased distance from the center of the shear band, the rotation angle significantly decreases to negative values, which corresponds to the anti- clockwise rotation fields. The anti- clockwise rotation fields obviously weaken in the gradient- transition region (line 2). The rotation angle displays only minor negative values away from the central shear band and varies between 6 degrees and 0 degrees. In the soft region (line 3), the rotation angle is almost the same in this line, and the anti- clockwise rotation fields almost vanish. The variation of the rotation fields across the shear band indicates that the STZ vortex- like motion mechanism characteristic in the GMG must be perturbed. This is due to the modification of the local strain field around the STZ caused by the gradient structure.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 405, 242, 421]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[147, 428, 852, 914]]<|/det|>
+Figure 4f schematically shows the STZ percolation mechanism under compressive strain in GMG. In the left hard region, since the rotation field shows a quadrupolar- like distribution around the STZs when subjected to compressive stress, the compressive strain is oriented along the y- axis, while the tensile strain is oriented along the x- axis. The activated STZ can perturb the surrounding STZ by generating strong, clockwise rotation strain fields. These clockwise rotation strain fields can compress the STZ located at the top right of the vortex while stretching the STZ located at the bottom left of the vortex. Meanwhile, this activated STZ can also be perturbed by the anti- clockwise rotation strain fields generated by surrounding STZs. In this state, the anti- clockwise rotation strain fields can also drive the movement of the vortex with a special angle. In this way the following STZ vortex will be activated and the shear band will percolate in this direction. When the STZ vortex moves to the soft region, the weakening of the atomic bonding will perturb the strain field around the STZ. The tensile strain fields will become larger, which will govern the vortex to be closer to the tensile strain direction. By the same reasoning, the influence from the anti- clockwise rotation strain fields will become weak. Hence, the angle of the STZ percolation path changes to a larger value than that in the hard region. The non- activated STZ will be aligned with the new local strain fields and the shear band will percolate in alignment with this new angle - ultimately, the shear band deflected as we observed experimentally. Hence, our model demonstrates that local strain fields can be effectively tuned by the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 83, 850, 151]]<|/det|>
+free volume gradient, and thus the STZ- vortex motion and shear banding behavior can be controlled. It is still desirable to optimize the gradient structure and strain fields to more effectively control the propagation of shear bands.  
+
+<|ref|>text<|/ref|><|det|>[[147, 158, 852, 397]]<|/det|>
+In summary, we have proposed a novel gradient BMG design strategy through cyclic heat treatment. The resulting substantial bulk scale structural gradient allows for extra plastic strain without sacrificing ultrahigh strength. Both experimental and computational evidence has demonstrated the importance of covering the whole structure with a tunable free volume gradient for the deflection of shear deformation. We have shown that the deflection of the shear band is accompanied by a gradient change in the vortex motion of the shear transformation zone, which is regulated by the variation of the non- uniform strain fields in the GMG. Our research highlights the potential of creating gradient structured engineering materials with high strength and plastic deformability.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 428, 232, 444]]<|/det|>
+## Methods:  
+
+<|ref|>text<|/ref|><|det|>[[147, 452, 852, 618]]<|/det|>
+Materials preparation. MGs with the atomic components of \(\mathrm{Zr_{58}Cu_{22}Fe_{8}Al_{12}}\) were prepared by arc melting of a mixture of pure elements (99.9% purity) in an argon atmosphere and injection casting into a copper moldwith a diameter of \(2\mathrm{mm}\) . For cryogenic thermal cycling, one cycle consisted of dipping the sample into liquid nitrogen for a certain period of time (6 s, \(25\mathrm{s}\) , \(70\mathrm{s}\) , and \(150\mathrm{s}\) ), followed by transferring it into hot water ( \(T = 323\mathrm{K}\) ) for the same time (6 s, \(25\mathrm{s}\) , \(70\mathrm{s}\) , and \(150\mathrm{s}\) ). All the samples were treated for 10 cycles.  
+
+<|ref|>text<|/ref|><|det|>[[147, 625, 852, 914]]<|/det|>
+Characterization and mechanical testing. Density measurements, based on the Archimedes method, were conducted using a high precision balance with an accuracy of \(\pm 0.01\mathrm{mg}\) . They were repeated at least 15 times to ensure data reliability. Vickers microhardness was measured on the cross- section using a Matsuzawa MMT- X indentation machine with a load of \(1.96\mathrm{N}\) and holding for \(15\mathrm{s}\) . The microstructure of the samples was examined by TEM using a Cs- corrected FEI Titan G2 80- 200 ChemiSTEM. Specimens for TEM were cut from the bulk MG specimens and thinned using a dual- beam FIB system (FEI Helios Nano- Lab 600i). Uniaxial compression tests were performed in an Instron 5982 mechanical testing machine at a strain rate of \(1\times 10^{- 4}\mathrm{s}^{- 1}\) at room temperature. The specimens, \(2\mathrm{mm}\) in diameter and \(4\mathrm{mm}\) in height with two ends, were carefully polished to ensure parallelism. The fracture features and surface morphology of all the samples after compression failure were investigated by
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 83, 850, 150]]<|/det|>
+scanning electron microscopy on a Zeiss Sigma500 field emission gun (FEG) SEM. The height maps of the fracture surface of all the samples were measured by the Dektak XT profile measurement.  
+
+<|ref|>text<|/ref|><|det|>[[146, 158, 852, 520]]<|/det|>
+Molecular dynamics simulations. Molecular dynamics simulations were performed using a Large- scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) \(^{37}\) with newly- developed Cu- Zr embedded atom method (EAM) potential \(^{38}\) . A cubic box containing 2213750 atoms was constructed with three- dimensional (3D) periodic boundary conditions (PBCs) as an initial model. The atoms were randomly arranged, and the system was melted by maintaining a high temperature of 2000 K for 8 ns with zero external pressure under the NPT ensemble. The melted model was then quenched at a constant cooling rate of \(10^{11} \mathrm{~K} / \mathrm{s}\) from 2000 K to 50 K and relaxed at 50 K for 2 ns to equilibrate the structure. The final size of the sample was about \(120 (\mathrm{x}) \times 60 (\mathrm{y}) \times 5 (\mathrm{z}) \mathrm{nm}^{3}\) . Afterwards, the gradient MG was prepared by randomly removing \(2\%\) of atoms in the right half of the model (along the Y direction). Once the gradient structure was generated, it was relaxed for 50 ps. The obtained gradient MG was further compressed along the X- direction at 50 K under a strain rate of \(1 \times 10^{8} \mathrm{~s}^{- 1}\) . The atomic- scale shear banding process was monitored by the atomic shear strain, \(\eta^{\mathrm{Mises39}}\) and rotation field by Ovito \(^{40}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 576, 250, 592]]<|/det|>
+## References:  
+
+<|ref|>text<|/ref|><|det|>[[145, 604, 852, 902]]<|/det|>
+[1] Ritchie, R. O. The conflicts between strength and toughness. Nat Mater 10, 817–822 (2011). [2] Launey, M. E. & Ritchie, R. O. On the fracture toughness of advanced materials. Adv. Mater. 21, 2103–2110 (2009). [3] Wu, X. et al. Extraordinary strain hardening by gradient structure. Proc. Natl. Acad. Sci. U.S.A. 111, 7197 (2014). [4] Wei, Y. J. et al. Evading the strength- ductility trade- off dilemma in steel through gradient hierarchical nanotwinsmade pioneering contributions to the mechanics of fracture, dislocations, and stress effects in solid materials. Nat. Commun. 5, 3580 (2014). [5] Chen, W. et al. Mechanically- induced grain coarsening in gradient nano- grained copper. Acta Mater. 125, 255 (2017). [6] Zhao, S. T. et al. Generating gradient germanium nanostructures by shock- induced
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[181, 90, 745, 107]]<|/det|>
+amorphization and crystallization. Proc. Natl. Acad. Sci. U.S.A. 114, 37 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[147, 118, 794, 135]]<|/det|>
+[7] Lu, K. Making strong nanomaterials ductile with gradients. Science 345, 1455 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[147, 146, 850, 191]]<|/det|>
+[8] Li, X. Y. et al. Mechanical properties and deformation mechanisms of gradient nanostructured metals and alloys. Nat. Rev. Mater. 5, 706 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[147, 201, 850, 247]]<|/det|>
+[9] Liu, Z. Q. et al. Functional gradients and heterogeneities in biological materials: Design principles, functions, and bioinspired applications. Prog. Mater. Sci. 88, 467 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[147, 257, 850, 303]]<|/det|>
+[10] Espinosa, H. D. et al. Merger of structure and material in nacre and bone- Perspectives on de novo biomimetic materials. Prog. Mater. Sci. 54, 1059 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[147, 312, 850, 357]]<|/det|>
+[11] Studart, A. R. Additive manufacturing of biologically- inspired materials. Chem. Soc. Rev. 45, 359 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[147, 368, 850, 413]]<|/det|>
+[12] Meyers, M. A. et al. Biological materials: Structure and mechanical properties. Prog. Mater. Sci. 53, 1 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[147, 423, 850, 468]]<|/det|>
+[13] Mayer, G. Rigid biological systems as models for synthetic composites. Science 310, 1144 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[147, 479, 850, 524]]<|/det|>
+[14] Fang, T. et al. Revealing Extraordinary Intrinsic Tensile Plasticity in Gradient Nano- Grained Copper. Science 331, 1587 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[147, 534, 850, 579]]<|/det|>
+[15] Cheng, Z. et al. Extra strengthening and work hardening in gradient nanotwinned metals. Science 362, 559 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[147, 590, 850, 635]]<|/det|>
+[16] Liu, C. et al. Nano- dual- phase metallic glass film enhances strength and ductility of a gradient nanograined magnesium alloy. Adv. Mater. 7, 2001480 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[147, 645, 850, 691]]<|/det|>
+[17] Zberg, B. et al. MgZnCa glasses without clinically observable hydrogen evolution for biodegradable implants. Nat Mater 8, 887 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[147, 701, 850, 746]]<|/det|>
+[18] Lin, Y. et al. Mechanical properties and optimal grain size distribution profile of gradient grained nickel. Acta Mater. 153, 279 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[147, 756, 850, 802]]<|/det|>
+[19] Meng, X. et al. Underwater Mechanically Robust Oil- Repellent Materials: Combining Conflicting Properties Using a Heterostructure. Adv. Mater. 30, 1706634 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[147, 812, 850, 857]]<|/det|>
+[20] Yang, Y. et al. Electrically assisted 3D printing of nacre- inspired structures with self- sensing capability. Sci. Adv. 5, 9490 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[147, 868, 850, 913]]<|/det|>
+[21] Li, J. et al. On strain hardening mechanism in gradient nanostructures. Int. J. Plast. 88, 89 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[145, 90, 850, 135]]<|/det|>
+[22] Yang, D. K. et al. Work hardening in ultrafine- grained titanium: Multilayering and grading. Acta Mater. 61, 2840 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[147, 146, 850, 190]]<|/det|>
+[23] Xu, W. et al. Strain- induced microstructure refinement in pure Al below 100 nm in size. Acta Mater. 152, 138- 147 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[147, 201, 850, 247]]<|/det|>
+[24] Laine, S. J. et al. Microstructural characterisation of metallic shot peened and laser shock peened Ti- 6Al- 4V. Acta Mater. 123, 350 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[147, 257, 850, 302]]<|/det|>
+[25] Ma, X. L. et al. Mechanical properties of copper/bronze laminates: Role of interfaces. Acta Mater. 116, 43 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[147, 312, 779, 330]]<|/det|>
+[26] Greer, A. L. et al. Shear bands in metallic glasses. Mater. Sci. Eng. R. 74, 71 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[147, 340, 851, 386]]<|/det|>
+[27] Zhang, Z. F. et al. Difference in compressive and tensile fracture mechanisms of Zr59Cu20Al10Ni8Ti3 bulk metallic glass. Acta Mater. 51, 1167 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[147, 396, 850, 441]]<|/det|>
+[28] Zhang, Z. F. et al. Fracture mechanisms in bulk metallic glassy materials. Phys. Rev. Lett. 91, 045505 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[147, 451, 850, 496]]<|/det|>
+[29] Hsueh, C. - H. et al. Shear fracture of bulk metallic glasses with controlled applied normal stresses. Scr. Mater. 59, 111 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[147, 507, 850, 552]]<|/det|>
+[30] Baricco, M. et al. Glass formation and mechanical properties of (Cu50Zr50)100- xAlx (x=0, 4, 5, 7) bulk metallic glasses. J. Alloys Compd. 483, 125 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[147, 562, 850, 607]]<|/det|>
+[31] Ketov, S. V. et al. Rejuvenation of metallic glasses by non- affine thermal strain. Nature 524, 200 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[147, 618, 848, 662]]<|/det|>
+[32] Schuh, C. A. & Lund, A. C. Atomistic basis for the plastic yield criterion of metallic glass. Nat. Mater. 2, 449 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[147, 673, 850, 718]]<|/det|>
+[33] Wu, Z. W. et al. Correlation between structural relaxation and connectivity of icosahedral clusters in CuZr metallic glass- forming liquids. Phys. Rev. B 88, 054202 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[147, 728, 848, 773]]<|/det|>
+[34] Sopu, D. et al. Atomic- level processes of shear band nucleation in metallic glasses. Phys. Rev. Lett. 119, 195503 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[147, 784, 850, 829]]<|/det|>
+[35] Sopu, D. et al. Atomic- scale origin of shear band multiplication in heterogeneous metallic glasses. Scr. Mater. 178, 57 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[147, 840, 850, 885]]<|/det|>
+[36] Scudino, S. & Sopu, D. Strain distribution across an individual shear band in real and simulated metallic glasses. Nano Lett. 18, 1221 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[147, 896, 850, 912]]<|/det|>
+[37] Mendelev, M. I. et al. Using atomistic computer simulations to analyze x- ray diffraction data
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[182, 90, 590, 106]]<|/det|>
+from metallic glasses. J. Appl. Phys. 102, 043501 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[147, 118, 849, 161]]<|/det|>
+[38] Plimpton, S. Fast parallel algorithms for short- range molecular dynamics. J. Comput. Phys. 117, 1 (1995).  
+
+<|ref|>text<|/ref|><|det|>[[147, 173, 850, 217]]<|/det|>
+[39] Shimizu, F. et al. Theory of shear banding in metallic glasses and molecular dynamics calculations. Mater. Trans. 48, 2923 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[147, 229, 850, 273]]<|/det|>
+[40] Stukowski, A. Visualization and analysis of atomistic simulation data with OVITO- the Open Visualization Tool. Model. Simulat. Mater. Sci. Eng. 18, 015012 (2010).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[171, 87, 830, 491]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 508, 852, 600]]<|/det|>
+<center>Figure 1. Schematic description of gradient BMG. (a) Schematic representation of one dominant shear band in uniform MG. (b) Schematic diagram of proposed GMG. (c) Proposed shear band deflection mechanism in a GMG during the uniaxial compression process. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[149, 108, 843, 551]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 566, 852, 831]]<|/det|>
+<center>Figure 2. Generation and Characterization of the GMG. (a) Schematic of the GMG by means of CTC. (b) Density values as a function of holding time for the treated MGs. (c) Cryogenic thermal cycling procedure from a high temperature (323 K) to liquid nitrogen temperature (77 K), together with the waiting time, \(t\) , at both the maximum and the minimum temperatures. (d) Variation of average hardness value along the distance from the center. (e) The top shows the method for hardness measurements along various circles. The bottom shows the t150 sample for TEM. (f)-(h) TEM images with the corresponding selected-area-diffraction patterns (SAED) of the edge, middle, and center regions (labeled A, B, C) for the t150 sample. (i) Corresponding radial distribution functions calculated from the SAED of the edge, middle, and center regions (labeled A, B, C) in the t150 sample. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[150, 125, 850, 608]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 632, 852, 824]]<|/det|>
+<center>Figure 3. Enhanced mechanical properties, fracture surface change and shear band deflection. (a) Compressive stress-strain curves for the as-cast and treated MGs. (b) Variation of the plastic strain with the structural gradient. (c)-(g) Lateral morphologies of the fractured as-cast and gradient MG samples. (h)-(l) The corresponding 3D contours of the fracture surfaces. (m) Height variation profiles of the middle dash line (h) along the shear band plane for as-cast and gradient MG samples. (n)-(s) SEM surface morphologies at positions 1-3 in the as-cast (c) and 4-6 in the t150 sample (g). </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[164, 80, 830, 700]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 712, 850, 901]]<|/det|>
+<center>Figure 4. Investigation of the deformation behavior of GMG by MD simulations. (a) Atomic free volumes as the functions of the position along the Y-direction in as-cast and gradient MGs. Inset shows an illustration of the design of GMG. (b) Representative stress-strain results for the as-cast and gradient MGs during compression along the X-direction. (c) The spatial distribution of atomic Mises strain (d) Rotation angle of the as-cast and gradient MGs at \(9\%\) compression strain. (e) The variations of the rotation angle along lines 1-3 in (d). (f) Schematic illustration of the STZ percolation mechanism in GMG. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[55, 100, 940, 700]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 728, 115, 747]]<|/det|>
+<center>Figure 1 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 769, 956, 836]]<|/det|>
+Schematic description of gradient BMG. (a) Schematic representation of one dominant shear band in uniform MG. (b) Schematic diagram of proposed GMG. (c) Proposed shear band deflection mechanism in a GMG during the uniaxial compression process.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[55, 60, 925, 680]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[43, 707, 117, 726]]<|/det|>
+<center>Figure 2 </center>  
+
+<|ref|>text<|/ref|><|det|>[[41, 746, 951, 951]]<|/det|>
+Generation and Characterization of the GMG. (a) Schematic of the GMG by means of CTC. (b) Density values as a function of holding time for the treated MGs. (c) Cryogenic thermal cycling procedure from a high temperature (323 K) to liquid nitrogen temperature (77 K), together with the waiting time, t, at both the maximum and the minimum temperatures. (d) Variation of average hardness value along the distance from the center. (e) The top shows the method for hardness measurements along various circles. The bottom shows the t150 sample for TEM. (f)-(h) TEM images with the corresponding selected- area- diffraction patterns (SAED) of the edge, middle, and center regions (labeled A, B, C) for the t150 sample. (i) Corresponding radial distribution functions calculated from the SAED of the edge, middle, and center regions (labeled A, B, C) in the t150 sample.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[62, 60, 920, 722]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 755, 117, 775]]<|/det|>
+<center>Figure 3 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 797, 944, 932]]<|/det|>
+Enhanced mechanical properties, fracture surface change and shear band deflection. (a) Compressive stress- strain curves for the as- cast and treated MGs. (b) Variation of the plastic strain with the structural gradient. (c)- (g) Lateral morphologies of the fractured as- cast and gradient MG samples. (h)- (l) The corresponding 3D contours of the fracture surfaces. (m) Height variation profiles of the middle dash line (h) along the shear band plane for as- cast and gradient MG samples. (n)- (s) SEM surface morphologies at positions 1- 3 in the as- cast (c) and 4- 6 in the t150 sample (g).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[66, 53, 768, 771]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 802, 116, 820]]<|/det|>
+<center>Figure 4 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 842, 950, 932]]<|/det|>
+Investigation of the deformation behavior of GMG by MD simulations. (a) Atomic free volumes as the functions of the position along the Y- direction in as- cast and gradient MGs. Inset shows an illustration of the design of GMG. (b) Representative stress- strain results for the as- cast and gradient MGs during compression along the X- direction. (c) The spatial distribution of atomic Mises strain (d) Rotation angle
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 944, 88]]<|/det|>
+of the as-cast and gradient MGs at \(9\%\) compression strain. (e) The variations of the rotation angle along lines 1-3 in (d). (f) Schematic illustration of the STZ percolation mechanism in GMG.
+
+<--- Page Split --->

preprint/preprint__7e55fccab658261127f61081ff8fb316e586f067d6d89676c5bf23eb2b0b3d2c/images_list.json

@@ -0,0 +1,220 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 | Univariate analysis of potential risk factors in COVID-19 mortality.",
+    "footnote": [],
+    "bbox": [
+      [
+        70,
+        45,
+        911,
+        870
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2 | Hematologic cancer is an independent risk factor for COVID-19 related mortality. (a) Kaplan Meier curve for COVID-19 survival of patients with solid (n=77) and hematologic (n=22) cancer. Cox regression-computed hazard ratio for mortality in hematologic vs solid cancer, adjusted for age, gender, smoking status, active cancer status, and ECOG performance status. (b) Ferritin, IL-6, and LDH in solid (n=62) and hematologic (n=15) cancer hospitalized for COVID-19. (All) Significance determined by Mann Whitney test: *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Median and 95% CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        247,
+        52,
+        770,
+        521
+      ]
+    ],
+    "page_idx": 37
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 | High dimensional analyses reveal immune phenotypes associated with mortality and distinct phenotypes between solid and hematologic cancers. (a) Demographic and mortality data for MESSI cohort at Penn. (b) Relative levels of SARS-CoV-2 IgG and IgM of solid (n=14) and hematologic (n=7) cancer patients and non-cancer patients (n=108). (c) (Left) Global UMAP projection of lymphocyte populations for all 45 patients pooled. (Right) Hierarchical clustering of Earth Mover's Distance (EMD) using Pearson correlation, calculated pairwise for lymphocyte populations. (d) UMAP projection of concatenated lymphocyte populations for each EMD cluster. (Yellow: High Density; Black: Low Density) (e) Heatmap showing expression patterns of various markers, stratified by EMD cluster. Heat scale calculated as column z-score of MFI. (f) Mortality, disease severity, and SARS-CoV-2 antibody data, stratified by EMD cluster (Cluster 5 n=5; Cluster 1,2,3,4 n=40). Mortality significance determined by Pearson Chi Square test. Severity assessed with NIH ordinal scale for COVID-19 clinical severity (1: Death; 8: Normal Activity)15. (g) UMAP projections of concatenated lymphocyte populations for solid cancer, hematologic cancer, and non-cancer patients. (h) CD8 and CD4 T cell and B cell frequencies in healthy donors (HD) (n=33), non-cancer (n=108), solid cancer (n=7), and heme cancer (n=4). (i) UMAP projection of non-naive CD8 T cell clusters identified by FlowSOM. (j) (Top) UMAP projections of non-naive CD8 T cells for non-cancer and cancer patients. (Bottom) UMAP projections indicating HLA-DR and CD38 protein expression on non-naive CD8 T cells for all patients pooled. (k) Frequency of activated FlowSOM clusters in HD (n=30), non-cancer (n=110), and cancer patients (n=8). (l) Representative flow plots and frequency of HLA-DR and CD38 co-expression in HD (n=30), non-cancer (n=110), solid cancer (n=7), and hematologic cancer (n=3) patients. (All) Significance determined by Mann Whitney test: *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Median and 95% CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        50,
+        40,
+        940,
+        750
+      ]
+    ],
+    "page_idx": 38
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 | CD8 T cell counts associated with survival in hematologic cancer patients with COVID-19. (a) Demographic and mortality data of MSKCC cohort. (b) (Left) Hierarchical clustering of Earth Mover's Distance (EMD) using Pearson correlation, calculated pairwise for lymphocyte populations. (Right) Global UMAP projection of lymphocyte populations pooled. (c) UMAP projection of concatenated lymphocyte populations for each EMD cluster. (Yellow: High Density; Black: Low Density) (d) Mortality (Cluster 5 n=7; Cluster 1,2,4 n=50), severity, and RT-PCR cycle threshold (Cluster 1 n=14; Cluster 2 n=5; Cluster 4 n=24; Cluster 5 n=6) (Lower Ct: Higher viral load) stratified by EMD cluster. Mortality significance determined by Pearson Chi Square test. (e) Relative levels of SARS-CoV-2 IgG and IgM of patients with recent cancer treatments (solid tx n=9; heme αCD20 n=7; heme other tx n=5). (f) Mortality, severity, and RT-PCR cycle threshold stratified by cancer treatment (remission n=9; solid obs n=6; solid tx n=19; heme obs n=5; heme chemo n=4; heme αCD20 n=10). Severity assessed with NIH ordinal scale for COVID-19 clinical severity. (g) Recent cancer treatment of patients in each EMD cluster. (h) Mortality of patients treated with B cell depleting therapy in EMD cluster 1 (red) and EMD cluster 4 (blue). (i) Absolute CD8 and CD4 T cell counts in patients treated with B cell depleting therapy (alive n=7; dead n=4). (j) Absolute CD8 and CD4 T cell counts and B cell counts in hematologic cancer patients (alive n=17; dead n=18). (k) Kaplan-Meier curve for survival in hematologic cancer patients stratified by CD8 T cell counts (threshold = 55.9; log-rank hazard ratio) (>=55.9 n=28; <55.9 n=13). CD8 count threshold determined by Classification and Regression Tree (CART) analysis. (All) Significance determined by Mann Whitney test: *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Median and 95% CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        50,
+        40,
+        950,
+        789
+      ]
+    ],
+    "page_idx": 39
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_1.jpg",
+    "caption": "Extended Data Fig. 1 | Inflammatory markers and blood cell counts in cancer patients with COVID-19. Clinical laboratory values for (a) inflammatory markers and (b) cell counts in solid (n=62) and hematologic (n=21) cancer patients. (All) Significance determined by Mann Whitney test: *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Median and 95% CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        112,
+        92,
+        857,
+        490
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_2.jpg",
+    "caption": "Extended Data Fig. 2 | SARS-CoV-2 antibody levels. (a) Relative levels of SARS-CoV-2 IgG and IgM in non-cancer (n=108) and cancer (n=21) patients. (b) Relative IgG levels in cancer patients. Each dot represents a cancer patient (Heme: Red; Solid: Yellow). (All) Significance determined by Mann Whitney test: *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Median and 95% CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        112,
+        88,
+        900,
+        310
+      ]
+    ],
+    "page_idx": 41
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_3.jpg",
+    "caption": "Extended Data Fig. 3 | Dimensionality reduction and EMD clustering of MESSI cohort. (a) UMAP projections of lymphocytes with indicated protein expression. (b) Frequencies of CD19+, CD3+, CD3+CD8+, and CD3+CD4+ cells of patients in each EMD cluster (Cluster 1 n=7; Cluster 2 n=16; Cluster 3 n=6; Cluster 4 n=10; Cluster 5 n=5). (All) Median and 95% CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        112,
+        95,
+        848,
+        460
+      ]
+    ],
+    "page_idx": 42
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_4.jpg",
+    "caption": "Extended Data Fig. 4 | Cellular phenotyping of COVID-19 patients with cancer. (a) Frequencies of circulating T follicular helper cells (cTfh), plasmalblasts, and CD138 expression on plasmalblasts (HD \\(n = 33\\) ; non-cancer \\(n = 108\\) ; solid cancer \\(n = 7\\) ; heme cancer \\(n = 3\\) ). (b) UMAP projection of non-naive CD8 T cells with indicated protein expression. (c) Heatmap showing expression patterns of various markers, stratified by FlowSOM clusters. Heat scale calculated as column z-score of MFI. (d) Frequencies of CD8 subsets: naive (CD45RA+CD27+CCR7+), central memory (CD45RA-CD27+CCR7+), transition memory (CD45RA-CD27+CCR7-), effector memory (CD45RA-CD27-CCR7-), and TEMRA (CD45RA+CD27-CCR7-) (HD \\(n = 33\\) ; non-cancer \\(n = 108\\) ; cancer \\(n = 9\\) ). (e) (Top) HLA-DR and CD38 coexpression in concatenated activated clusters (3, 4, and 5) and associated UMAP localization. (Bottom) Frequency of activated clusters (3, 4, and 5) in each patient (HD \\(n = 30\\) ; non-cancer \\(n = 110\\) ; solid-cancer \\(n = 8\\) ). (All) Significance determined by Mann Whitney test: \\(*p<0.05\\) , \\(**p<0.01\\) , \\(***p<0.001\\) , and \\(****p<0.0001\\) . Median and 95% CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        66,
+        50,
+        925,
+        760
+      ]
+    ],
+    "page_idx": 43
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_5.jpg",
+    "caption": "Extended Data Fig. 5 | Cellular, serologic, and clinical features in solid and hematologic cancer patients with COVID-19. (a) Absolute counts of CD4, CD8, and CD19 expression in remission (n=11), solid cancer (n=23), and hematologic cancer (n=41) patients. (b) Relative levels of SARS-CoV-2 IgG and IgM in solid (n=11) and hematologic cancer (n=14) patients. (c) Severity (NIH ordinal scale for COVID-19 clinical severity) and RT-PCR cycle threshold (remission n=9; solid n=25; heme n=28) (Lower Ct: Higher viral load). (d) NIH ordinal scale for COVID-19 clinical severity. (All) Significance determined by Mann Whitney test: *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Median and 95% CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        125,
+        90,
+        900,
+        626
+      ]
+    ],
+    "page_idx": 44
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_6.jpg",
+    "caption": "Extended Data Fig. 6 | Dimensionality reduction and EMD clustering of MSKCC cohort. (a) UMAP projections of lymphocytes with indicated protein expression. (b) Absolute counts of CD19+, CD3+, CD3+CD8+, and CD3+CD4+ cells of patients in each EMD cluster (Cluster 1 \\(n = 18\\) ; Cluster 2 \\(n = 6\\) ; Cluster 3 \\(n = 26\\) ; Cluster 4 \\(n = 7\\) ). (All) Median and 95% CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        112,
+        92,
+        864,
+        460
+      ]
+    ],
+    "page_idx": 45
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_9.jpg",
+    "caption": "Extended Data Fig. 9 | Association of mortality with cell counts and viral load. (a) RT-PCR cycle threshold of patients treated with aCD20 therapy (alive \\(n = 7\\) ; dead \\(n = 3\\) ). (b) Absolute counts of CD8+, CD4+, and CD19+ cells in solid cancer patients (alive \\(n = 16\\) ; dead \\(n = 7\\) ). (All) Significance determined by Mann Whitney test: \\(^{*}p< 0.05\\) , \\(^{**}p< 0.01\\) , \\(^{***}p< 0.001\\) , and \\(^{****}p< 0.0001\\) . Median and \\(95\\%\\) CI shown.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 46
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        66,
+        90,
+        803,
+        800
+      ]
+    ],
+    "page_idx": 47
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2 Hematologic cancer is an independent risk factor for COVID- 19 related mortality. (a) Kaplan Meier curve for COVID- 19 survival of patients with solid \\((n = 77)\\) and hematologic \\((n = 22)\\) cancer. Cox regression- computed hazard ratio for mortality in hematologic vs solid cancer, adjusted for age, gender, smoking status, active cancer status, and ECOG performance status. (b) Ferritin, IL- 6, and LDH in solid \\((n = 62)\\) and hematologic \\((n = 15)\\) cancer hospitalized for COVID- 19. (All) Significance determined by Mann Whitney test: \\(*p< 0.05\\) , \\(**p< 0.01\\) , \\(***p< 0.001\\) , and \\(****p< 0.0001\\) . Median and \\(95\\%\\) CI shown.",
+    "footnote": [],
+    "bbox": [
+      [
+        88,
+        60,
+        680,
+        597
+      ]
+    ],
+    "page_idx": 48
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        80,
+        45,
+        912,
+        730
+      ]
+    ],
+    "page_idx": 55
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        55,
+        45,
+        936,
+        787
+      ]
+    ],
+    "page_idx": 56
+  }
+]

preprint/preprint__7e6200f0e6beda7a0fa07dea90dcb4211a3ae35c56dedd427f32bb24a1ac1e71/images_list.json

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

preprint/preprint__7e6200f0e6beda7a0fa07dea90dcb4211a3ae35c56dedd427f32bb24a1ac1e71/preprint__7e6200f0e6beda7a0fa07dea90dcb4211a3ae35c56dedd427f32bb24a1ac1e71.mmd

@@ -0,0 +1,373 @@
+
+# Transglutaminase 3 crosslinks secreted MUC2 and stabilizes the colonic mucus layer  
+
+Jack Sharpen University of Manchester https://orcid.org/0000- 0002- 6268- 3244  
+
+Brendan Dolan University of Gothenburg  
+
+Elisabeth Nystrom University of Gothenburg  
+
+George Birchenough University of Gothenburg https://orcid.org/0000- 0003- 2283- 2353  
+
+Liisa Arike University of Gothenburg  
+
+Beatriz Martinez- Abad University of Gothenburg https://orcid.org/0000- 0002- 0521- 3473  
+
+Malin Johansson University of Gothenburg https://orcid.org/0000- 0002- 4237- 6677  
+
+Gunnar Hansson University of Gothenburg https://orcid.org/0000- 0002- 1900- 1869  
+
+Christian Recktenwald ( \(\square\) christian.recktenwald@medkem.gu.se) University of Gothenburg https://orcid.org/0000- 0003- 1710- 1863  
+
+## Article  
+
+Keywords: colonic mucus layer, TGM3, transglutaminase, MUC2 mucin  
+
+Posted Date: June 21st, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 555255/v1  
+
+License: \(\circledcirc\) 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 January 11th, 2022. See the published version at https://doi.org/10.1038/s41467- 021- 27743- 1.
+
+<--- Page Split --->
+
+# Transglutaminase 3 crosslinks secreted MUC2 and stabilizes the colonic mucus layer  
+
+3 Jack D. A. Sharpen, Brendan Dolan, Elisabeth E. L. Nyström, George M. H. Birchenough, Liisa Arike, Beatriz Martinez- Abad, Malin E. V. Johansson, Gunnar C. Hansson and Christian V. Recktenwald\* 7 8 From the Department of Medical Biochemistry, University of Gothenburg, SE- 405 30 Gothenburg, Sweden 10 \*Correspondence to: christian.recktenwald@medkem.gu.se  
+
+## Abstract  
+
+The colonic mucus layer is organized as a two- layered system providing a physical barrier against pathogens and simultaneously harboring the commensal flora. The factors contributing to the organization of this gel network are not well understood. In this study, the impact of transglutaminase activity on this architecture was analyzed. Here, we show that transglutaminase TGM3 is the major TGM isoform expressed and synthesized in the colon. Furthermore, intrinsic extracellular TGM activity in the secreted mucus was demonstrated in vitro and ex vivo. Absence of this acyl- transferase activity resulted in faster degradation of the major mucus component the MUC2 mucin and changed the biochemical properties of mucus. Finally, TGM3- deficient mice showed an early increased susceptibility to DSS- induced colitis. Thus, these observations suggest that natural isopeptide cross- linking by TGM3 is important for mucus homeostasis and protection of the colon from inflammation, a suggested pre- stage of colon carcinoma.
+
+<--- Page Split --->
+
+## Introduction  
+
+The epithelium in the intestinal tract is covered by mucus that provides protection from luminal challenges and bacterial infiltration \(^{1}\) . Despite the similar proteome composition, the organization of the mucus gel network differs considerably in the small and large intestine \(^{2}\) . Whereas small intestinal mucus is non- attached, the colonic mucus is a two- layered system with an attached, bacteria- free inner layer and an outer layer harboring the commensal flora \(^{1}\) , \(^{3,4}\) . The molecular mechanisms determining these structural differences are not well understood. The predominant component of mucus is the gel- forming MUC2 mucin that is synthesized by intestinal goblet cells. It has been shown that Muc2 \(^{2 / 3}\) mice develop spontaneous colitis, a pre- stage of colon carcinoma \(^{5,6}\) . Furthermore, the MUC2 levels in patients suffering from active ulcerative colitis (UC) are decreased when compared to healthy control patients \(^{7}\) .  
+
+The human MUC2 monomer consists of 5,130 amino acids organized in three complete and one partial von Willebrand D (VWD) domains in the N- terminal part followed by the first CysD domain and two Proline-, Threonine- and Serine- rich (PTS) sequences that are separated by the second CysD domain \(^{8,9}\) . The C- terminus harbors a fourth vWD domain, two vWC domains, and the cysteine- knot. During its transport through the endoplasmic reticulum and the Golgi- network MUC2 monomers first form C- terminal dimers and in the later stages of the secretory pathway N- terminal dimers or trimers \(^{10,11}\) . Furthermore, the PTS sequences become heavily O- glycosylated to form mucin domains. This posttranslational modification (PTM) shifts the mass of MUC2 from roughly 650 kDa to more than 2.5 MDa. During the later stages of the secretory pathway isopeptide bonds are introduced probably contributing to the insolubility of MUC2 in chaotropic salts, like guanidinium chloride \(^{12}\) . An enzyme family that is able to catalyze these natural protein cross- links are transglutaminases (TGM).
+
+<--- Page Split --->
+
+Transglutaminases (R:protein- glutamine \(\gamma\) - glutamyltransferases; E. C. 2.3.2.13) comprise a family of \(\mathrm{Ca^{2 + }}\) - dependent acyl- transferases that can catalyze the transamidation or deamidation of protein- bound glutamine residues that can lead to natural cross- links through the formation of an isopeptide bond between the side chains of glutamine and lysine. This PTM is known to limit protein degradation by conformational changes and modification of protease- labile Lys residues \(^{13,14}\) . There are nine mammalian TGMs where TGM2 is the most ubiquitously expressed isoform \(^{13,15}\) . This isoform is predominantly localized in the cell cytosol, but can also be found associated with the plasma membrane. Furthermore, it can be secreted by unknown mechanisms after P2X7 receptor activation \(^{16}\) . The enzymatic activity of TGM2 is normally silent but during mechanical injury it becomes activated and acts as a wound healing enzyme by stabilizing extracellular matrix (ECM) and cell- ECM interactions \(^{17,18}\) . Another process where TGMs are important is the morphogenesis of the skin. Here, TGM1, 3 and 5 are involved in the formation of the stratum corneum by cross- linking the envelope precursors such as inloricrin and involucrin \(^{19}\) .  
+
+Whether transamidation also has a role in the formation and stabilization of intestinal mucus is currently unknown. Mucus and mucins are stored highly concentrated in the granules of goblet cells and expand 1,000- fold in volume upon secretion. If TGM- catalyzed isopeptide cross- links contribute to mucus homoeostasis, this processing has to occur after secretion and expansion. Here, we suggest that extracellular TGM activity plays a role in organizing the mucus gel in the colon, especially by increasing its stability. To test this hypothesis the abundance of different TGM isozymes was evaluated and their enzymatic activity determined. We found that the formation of \(\mathrm{N^{E}}\) - ( \(\gamma\) - glutamyl)- lysine isopeptide cross- links in colonic mucus was based on extracellular TGM3- intrinsic activity. Furthermore, mice lacking this TGM isoform secrete a more protease- sensitive MUC2 molecule. In addition, \(Tgm3^{i / c}\) mice are less protected against dextran sodium sulfate (DSS) induced
+
+<--- Page Split --->
+
+colitis. Together, our observations indicate that TGM- catalyzed cross- links are important for the stabilization/homoeostasis of colonic mucus and its resistance against disease- inducing conditions.  
+
+## Results  
+
+Transglutaminase TGM 3 is a dominant cross- linking enzyme in the colonFirstly, we determined which transglutaminase isozymes were expressed and synthesized in the colonic epithelium. Mouse colon tissue of wild- type (WT) and Tgm knock- out mice were analyzed for protein abundance by using immunohistochemistry (IHC), mass spectrometry (MS) and gel electrophoresis followed by western blot. As we were interested on the impact of transglutaminases on mucus homeostasis a recently published single cell transcriptomic study<sup>20</sup> analyzing MUC2- producing goblet cells and non- goblet epithelial cells was mined for the expression profile and protein abundance of the various TGM family members. Analyzing mRNA levels in colonic goblet cells and the remaining epithelial cell populations revealed only transcripts for Tgm2 and Tgm3 genes (Fig. 1a). Next, the TGM2 and TGM3 protein abundance determined by mass spectrometry (MS) in these two cell fractions was extracted. This method revealed approximately 10- times lower levels of TGM3 in the goblet cells compared to the non- goblet epithelial cells whereas the abundance of TGM2 was two- three orders of magnitude lower than TGM3 in the respective cell population (Fig. 1b). To evaluate the tissue localization of TGM2 and TGM3, immunohistochemical analyses were performed in WT, Tgm2<sup>-/-</sup> and Tgm3<sup>-/-</sup> animals together with the UEA1 lectin staining for the highly glycosylated MUC2 mucin. None of the strains reacted with the anti- TGM2 antibody, confirming the low levels of this isoform (Fig. 1c). That this antibody was functional was tested on duodenal tissue sections where a signal for TGM2 was easily observed (Suppl. Fig. S1a). In line with the quantitative data from mRNA expression and protein abundance, both
+
+<--- Page Split --->
+
+WT and \(Tgm2^{- / - }\) animals showed a strong staining for the TGM3 isoenzyme in the epithelium and as expected no signal in \(Tgm3^{- / - }\) mice (Fig. 1d).  
+
+As TGM3 lacks a signal sequence, we determined if TGM3 could nonetheless be secreted into the mucus. To answer this, gel electrophoresis and western blot analyses for TGM2 and 3 in colonic mucus were performed. Recombinantly expressed TGM2 and cleaved TGM3 were also loaded as positive controls either non- activated or activated by \(\mathrm{Ca^{2 + }}\) - preincubation (Fig. 1e). The majority of TGM3 was represented by a band migrating around \(75\mathrm{kDa}\) and a weaker signal migrating at approximately \(50\mathrm{kDa}\) in both WT and \(Tgm2^{- / - }\) animals. These two bands represent the zymogenic and active form of the enzyme, respectively. Furthermore, several diffuse, but weak, TGM3- signals migrating between 150 and \(250\mathrm{kDa}\) were detected in the WT and \(Tgm2^{- / - }\) strains suggesting the self- multimerization of the enzyme and/or its incorporation into substrate proteins. As similar signals were detected in the activated positive control for TGM3, it is likely that self- multimerization occurs in mucus. In contrast, TGM 2 was not detected in the mucus samples of any mouse strain. Specificity of the used antibodies for the respective isoform was determined upon western blot analyses, the anti- TGM3 antibody showed a cross- reactivity \(< 8\%\) on TGM2 and similarly vice versa (Suppl. Fig. S1b) Together the results show that TGM3 is the predominant transglutaminase in the colonic epithelium and the only isozyme detected in the mucus. Furthermore, its expression in goblet cells suggests that its presence in mucus arises at least partly from active secretion and not only from shedded cells.  
+
+TGM3 activity is present in colonic mucus  
+
+Next, we asked if TGM3 is enzymatically active in the colonic mucus and could thereby contribute to its stability by the formation of additional cross- links. For that purpose, a qualitative assay using the incorporation of biotinylated isoform- specific substrate peptides
+
+<--- Page Split --->
+
+T26 (TGM2) and E51 (TGM3) in mucus was performed. The mucus was incubated with \(\mathrm{Ca^{2 + }}\) and the respective peptide probe followed by gel electrophoresis and western blot using streptavidin detection (Fig. 2a). Specific incorporation of the two peptides was observed in WT and \(Tgm2^{- / - }\) mucus, but not in mucus from \(Tgm3^{- / - }\) animals. Non- specific signals were observed in all samples, including control reactions where transglutaminase activity was inhibited by iodoacetamide (IAA). These bands are likely due to endogenously biotinylated proteins as for example pyruvate-carboxylase. Thus, the detected cross- linking activity in the mucus arises from TGM3- mediated catalysis. To analyze if endogenous mucus contains sufficient \(\mathrm{Ca^{2 + }}\) - ions for the activation of TGM3, the experiment was repeated without calcium addition. Similar results as with exogenous \(\mathrm{Ca^{2 + }}\) - addition were obtained, indicating the presence of intrinsic extracellular transglutaminase activity in colonic mucus (Fig. 2b). These results suggest that endogenous acyl- donor protein substrates are present in colonic mucus. However, the formation of a transglutaminase- catalyzed cross- linked mucus gel- network also requires the presence of acyl- acceptor proteins. Therefore, the \(\mathrm{Ca^{2 + }}\) - free experimental set up was modified by replacing the glutamine- donor with the primary amine 5- Biotinyl- pentylamine (5- BP) as acyl- donor. Similar to the results from the acyl- acceptor experiments, specific signals were detected when the acyl- donor compound was added to mucus of WT and \(Tgm2^{- / - }\) animals, but not in the \(Tgm3^{- / - }\) mucus or when IAA was added (Fig. 2c). Together, the results show that colonic mucus contains intrinsically, active TGM3 as well as both acyl- acceptor and - donor molecules allowing transamidating reactions to take place.  
+
+To quantify the intrinsic transamidating activity in colonic mucus, a colorimetric assay for the incorporation of a TGM- promiscuous peptide (A25) and the two isozyme- selective peptide substrates (peptides \(\mathrm{T26^{21}}\) and \(\mathrm{E51^{22}}\) ) into casein was performed (Fig. 2d). A natural cross- linking activity in WT mucus of \(\approx 8\pm 2\) U/mg for the promiscuous substrate was
+
+<--- Page Split --->
+
+determined. Substitution with the TGM3- specific substrate E51 led to a 1.5- fold increase \((\approx 12\pm 4\mathrm{U / mg})\) of the transamidating activity, whereas a residual activity of \(0.8\pm 0.3\mathrm{U / mg}\) was observed for the TGM2- specific substrate. However, no measurable activity could be obtained in the \(Tgm3^{- / - }\) mucus as the detected values were below the limit of detection for our assay (Fig. 2d, Suppl. Fig. S2). Blocking of the TGM- reaction with Z- DON led to an almost complete \((88\%)\) inhibition for the promiscuous peptide A25. In line with our other results (Fig. 1, Fig. 2 a- c), the natural cross- linking activity was related to TGM3 as the use of the TGM2- specific substrate T26 led to less than \(10\%\) transglutaminase activity compared to the TGM3- specific substrate in WT animals and was also below the limit of quantification of this assay. These experiments further demonstrated substantial intrinsic transamidating activity in colonic mucus of WT animals, but not in \(Tgm3^{- / - }\) , as addition of extra \(\mathrm{Ca^{2 + }}\) was not required.  
+
+The intrinsic mucus transamidating activity of TGM3 was further studied using an ex vivo approach where the distal colon from WT and \(Tgm3^{- / - }\) animals were mounted in a perfusion chamber and the fluorescently labelled glutamine- donor probe E51 was added and its incorporation monitored (Fig. 2e). Fig. 2f- h show the confocal microscopic analyzes of E51 incorporation in the respective tissue/mucus specimen in the \(\mathrm{x / z}\) plane (top panels) and snap shots of probe incorporation of the \(\mathrm{x / y}\) plane inside the mucus ( bottom panels). A homogeneous punctuated pattern of E51 fluorescence was observed throughout the whole mucus layers of WT animals (Fig. 2f and Suppl. Movie M1). However, when \(Tgm3^{- / - }\) mice were analyzed in the same way, the incorporation was dramatically reduced and limited to shedding epithelial cells (Fig. 2g). A similar lack of incorporation in WT animals was observed in the presence of the transglutaminase inhibitor Z- DON (Fig. 2h). These results demonstrate extracellular TGM3 activity ex vivo. Together these results show that the colonic mucus contains natural acyl- donor and acyl- acceptor molecules together with intrinsic TGM3- mediated transamidating activity.
+
+<--- Page Split --->
+
+Loss of TGM3 alters biochemical properties of mucus/MUC2The MUC2 monomer is a large glycoprotein with a mass of around 2.5 MDa (Fig. 3a). It is the most abundant constituent in colonic mucus and is thus a potential target for TGM3- mediated cross- linking, something that could influence its biochemical properties. Colonic mucus from WT, \(Tgm2^{- / - }\) and, \(Tgm3^{- / - }\) mice was isolated and disulfide bonds reduced followed by separation via composite agarose- PAGE (AgPAGE) and detected by in- gel immunostaining using anti- MUC2C3 antibody (Fig. 3b). WT and \(Tgm2^{- / - }\) showed two identical diffuse fast- moving bands assumed to be MUC2 monomeric bands and several additional slow- moving and heavily stained bands for higher oligomers. This was in contrast to the \(Tgm3^{- / - }\) - mucus, where MUC2 showed a faster migrating diffuse band and two to WT different bands migrating similar to the WT monomer.- These differences in the electrophoretic migration pattern suggest that \(Tgm3^{- / - }\) MUC2 is qualitatively different to that of WT and \(Tgm2^{- / - }\) and argues for TGM3- mediated isopeptide bond modification of MUC2.  
+
+As isopeptide bonds can prevent proteolytic cleavage and secreted mucus is normally exposed to numerous endogenous and bacterial proteolytic enzymes, we hypothesized that the different size of MUC2 formed in \(Tgm3^{- / - }\) mice was a result of protease- catalyzed degradation in vivo. To test this hypothesis, colonic mucus of WT and \(Tgm3^{- / - }\) mice was first isolated and solubilized by reduction with dithiothreitol. The resulting samples were treated with the serine protease LysC, followed by the separation of the reaction products via composite AgPAGE and Alcian Blue staining of the heavily glycosylated and protease- resistant MUC2 domains (PTS sequence). All three strains showed three identical intensely stained bands after LysC treatment (Fig. 3c). Interestingly, this band pattern was also observed in the non- treated \(Tgm3^{- / - }\) mice, but not in the WT or \(Tgm2^{- / - }\) animals. This could suggest that the faster migrating MUC2 bands in the non- treated \(Tgm3^{- / - }\) animals represent products that have been already degraded in vivo. To confirm this, the fastest MUC2
+
+<--- Page Split --->
+
+migrating bands from the non- treated WT (WT- M) and \(Tgm3^{- / - }\) ( \(Tgm3^{- / - }\) - M) samples were excised from the gels (Fig. 3b) followed by mass spectrometric analyses of their tryptic/AspN peptides. The peptide coverage of the MUC2 sequence of three biological replicates is summarized in a heat- map shown in Fig. 3d. The WT monomers showed peptides from all domains except the PTS as expected. Interestingly, the \(Tgm3^{- / - }\) MUC2 molecule showed almost exclusively peptides from the central CysD2 domain (Fig. 3a and d). The vWD4 domain was weakly covered in both animals explaining the anti- MUC2C3 staining. As the fastest migrating bands in the \(Tgm3^{- / - }\) mucus were stained by Alcian Blue and have masses larger than \(460\mathrm{kDa}\) , these bands must also include the two mucin domains surrounding CysD2. These PTS1 and PTS2 sequences are highly glycosylated, resistant to proteolytic enzymes, and not identifiable by mass spectrometry (Fig. 3a). Thus, the MUC2 mucin in the \(Tgm3^{- / - }\) mice is suggested to be already degraded in vivo due to it being more susceptible to degradation in the colon lumen.  
+
+The most likely explanation for the more degraded MUC2 in \(Tgm3^{- / - }\) mice is the loss of protective transglutaminase- catalyzed isopeptide bonds. To search for such bonds, we mined the mass spectrometry data sets for the presence or absence of such cross- links. An example is shown in the mass spectrum of a dipeptide for an intramolecular cross- link connecting Gln 2503 with Lys 2508 (Fig. 3e). This intramolecular cross- linked peptide was only detected in MUC2 from WT, but not in \(Tgm3^{- / - }\) animals. This isopeptide bridge is located between the vWC2 domain and the cysteine- knot (CK) domain (Fig. 3a). There are likely several additional cross- links and this isopeptide- bridged peptide is only one example, but its absence in \(Tgm3^{- / - }\) MUC2 supports this interpretation.  
+
+As non- reduced secreted MUC2 polymers in the intestine are known to be insoluble in guanidinium chloride due to isopeptide bonds formed intracellularly \(^{23}\) , we asked if TGM3- mediated isopeptide cross- links contributed to this property. To address this question,
+
+<--- Page Split --->
+
+insoluble mucus from WT and \(Tgm3^{- / - }\) mice was precipitated by centrifugation and the turbidity of soluble material in the supernatant recorded (Fig. 3f). The turbidity of the samples from \(Tgm3^{- / - }\) animals was increased by approximately \(30\%\) when compared to WT and \(Tgm2^{- / - }\) strains. This result further supports the idea that disintegration of the MUC2 mucin network was more prominent in the mice lacking the TGM3 enzyme.  
+
+Mucins have been shown to attach to hydrophobic surfaces \(^{24}\) . We hypothesized that natural isopeptide cross- links might contribute to this biophysical property and thus analyzed the hydrophobic character of colonic mucus by using a thermal fluorescent shift assay. Colonic mucus mixed with the hydrophobic dye SyproOrange was subjected to a linear temperature gradient and the fluorescence measured (Fig. 3g). At higher temperatures (>50°C) the \(Tgm3^{- / - }\) mucus showed an increased fluorescence in relation to WT, indicating an increased exposure of hydrophobic protein parts. Preincubation of \(Tgm3^{- / - }\) mucus with recombinant TGM3 partly normalized the mucus. It can be suggested that TGM3- mediated isopeptide bonds in WT mucus prevented the unfolding of its constituents.  
+
+Mucus processing and tissue secretory responses were assessed using ex vivo mucus measurement assays. Using this approach, we detected no differences in baseline mucus growth rate or carbachol- induced secretory responses between WT and \(Tgm3^{- / - }\) tissues (Suppl. Fig. S3a). A similar approach can be used to measure mucus barrier function by applying bacteria- sized (1μm diameter) beads to the mucus surface and determining the extent of bead penetration into the mucus via confocal microscopy. However, again no difference between WT and \(Tgm3^{- / - }\) tissues was detected using this approach (Suppl. Fig. S3b), which was surprising, as we had observed a more degraded MUC2 mucin in the \(Tgm3^{- / - }\) animals. Nonetheless, we hypothesized that lack of TGM3 would affect mucus barrier stability and thus treated colonic tissue from WT and \(Tgm3^{- / - }\) animals with pronase. In WT animals and before addition of pronase to \(Tgm3^{- / - }\) tissue, the fluorescent beads remained on
+
+<--- Page Split --->
+
+top of the mucus layer (Fig. 3h and Suppl. Movies M2 and M3). However, after pronase- treatment of \(Tgm3^{- / - }\) explants, a progressive decrease in mucus thickness was observed and the beads were more easily washed away and/or penetrated down to the epithelial surface indicating that \(Tgm3^{- / - }\) mucus was less protected against proteolytic attack (Fig. 3i and Suppl. Movie M3).  
+
+Tgm3- mice are more susceptible to early DSS- induced damageThe altered biochemical properties of mucus and its higher susceptibility to proteolytic degradation in the absence of TGM3 activity suggested that \(Tgm3^{- / - }\) mice could be more susceptible to dextran sodium sulfate (DSS) induced colitis. To test this, age- matched cohoused \(Tgm3^{- / - }\) and WT animals were challenged with DSS. The body weight of WT mice increased during the first four days whereas the \(Tgm3^{- / - }\) animals started to lose weight from day three and showed on trend decreased body weights compared to WT mice until day 6 (Fig. 4a). This was reflected by an earlier detection of occult blood in the feces of \(Tgm3^{- / - }\) mice one day after the start of the experiment (Fig. 4b). Consequently, the \(Tgm3^{- / - }\) animals showed a significant raised disease activity index score (DAI) from day two to day five after the start of the DSS- treatment (Fig. 4c). Higher DAI was maintained in the \(Tgm3^{- / - }\) compared to WT animals until day 6, when the colitis became also established in the WT animals. Finally, 50% of the \(Tgm3^{- / - }\) animals had to be sacrificed at day 7, compared to 10% of WT mice, due to suffering and loss of weight following the ethical permit (Fig. 4d). Furthermore, the colon length of \(Tgm3^{- / - }\) mice was reduced to 88% of the WT length after 7 days of DSS treatment (Fig. 4e and f). Histopathological analysis of the colonic tissue after eight days of DSS treatment revealed the loss of crypts and an extensive infiltration of immune cells in both strains. These effects were more pronounced in the distal colon (Fig. 4g). However, histological examination of Hematoxylin/Eosin- stained tissue by a blinded pathologist did not
+
+<--- Page Split --->
+
+detect significant differences between the two animal strains at the end of DSS- treatment (not shown). DSS has previously been shown to disrupt the mucus layer properties \(^{25}\) and mice lacking the MUC2 mucin are very susceptible already at day one of DSS treatment \(^{4}\) . The early on- set of DSS effects in the \(Tgm3^{- / - }\) supports the conclusion that the colonic mucus is defect in these animals. When colonic tissue was analyzed by immunohistochemistry for TGM2, this isozyme that was absent in non- treated WT and \(Tgm3^{- / - }\) as shown in Fig. 1a, was now detected in both the WT and \(Tgm3^{- / - }\) animals after 7 days of DSS- treatment (Fig. 4h). Taken together \(Tgm3^{- / - }\) animals were significantly more susceptible towards the colitis- inducing effects of DSS as a faster disease onset was observed resulting in a decreased probability of survival.  
+
+## Discussion  
+
+We have previously shown that the reduction- insensitive MUC2 oligomers formed in a cell line producing MUC2 were cross- linked by isopeptide bonds as catalyzed by a yet unidentified transglutaminase \(^{12}\) . However, this colorectal cell line does not secrete MUC2 and could not be used to learn if and how extracellular cross- linking could contribute to mucus homeostasis and colon barrier function. By using WT and knock- out mouse strains, we uncovered intrinsic transglutaminase activity in secreted colonic mucus mediated by TGM3. The observations provide evidence for the protective effect by natural cross- links. That TGM3 is the dominant transglutaminase of the colon is in accordance with a previous mucus proteome study \(^{2}\) . mRNAseq and MS studies detected minor amounts of TGM2, but based on the label- free mass spectrometric quantification TGM2 was \(< 1\%\) of that of TGM3 and could represent contaminating material from the ileum. In support of this, TGM2 was not detected by immunohistochemistry or gel electrophoresis/western blot. Previous work from Jeong and co- workers has claimed TGM2 as the major transglutaminase of the colon \(^{26}\) .
+
+<--- Page Split --->
+
+However, these authors used only immunohistochemistry to demonstrate the presence of TGM2 and no antibody staining against TGM3 was tested. Likely cross reactivity of the used antibody can explain this observation.  
+
+The strong TGM3 signals observed by gel electrophoresis/western blot analyses of mucus represented the zymogenic form of TGM3. In addition to this, several TGM3 bands with higher molecular masses were detected in the range between 150 and \(250\mathrm{kDa}\) . Together with control reactions performed with recombinantly activated TGM3, these signals strongly suggest that the enzyme can self- multimerize and/or incorporate itself into other molecules as previously observed for TGM2<sup>27</sup>.  
+
+Recent reports have shown that TGM2 is extracellularly inactive and can be activated after injury or stress<sup>17, 28</sup>. Here, we were able to demonstrate intrinsic, extracellular transglutaminase activity in both the WT and \(Tgm2^{- / - }\) mouse strains, but its absence in \(Tgm3^{- / - }\) animals. In addition, the obtained information showed the presence of natural acyl- donor and - acceptor molecules in colonic mucus thereby implying the possibility of in vivo isopeptide- based cross- linking of different mucus components. Furthermore, transglutaminase activity could be detected without calcium- addition, showing that extracellular transglutaminase activity is intrinsic to the large intestinal mucus. Since a \(>90\%\) reduction in TGM activity was observed when the TGM2- selective substrate T26 was used in a quantitative assay and no activity was found in the \(Tgm3^{- / - }\) animals, we can conclude that the transamidation activity of mucus is almost exclusively dependent on TGM3 in colon. This is in line with the shown absence of TGM2 protein. In an ex vivo assay, a punctuated incorporation of the specific TGM3 peptide substrate E51 in colonic mucus was observed confirming our in vitro observations. However, it was not possible to perform the ex vivo mucus incorporation approach under \(\mathrm{Ca^{2 + }}\) - free conditions since the normal cellular signaling of the tissue depends on an extracellular calcium pool. This is reflected by measurements of the luminal calcium
+
+<--- Page Split --->
+
+concentrations in the gut which vary between 5 and \(20 \mathrm{mM}\) depending on the feed state \(^{29,30}\) whereas the concentration in the used buffer was \(1.3 \mathrm{mM}\) representing the physiological luminal calcium concentration \(^{31}\) . Given that only \(20\%\) of the daily calcium intake is resorbed, mainly in the small intestine, the luminal \(\mathrm{Ca}^{2 + }\) - concentration in the colon should be sufficient to occupy the second and third \(\mathrm{Ca}^{2 + }\) - binding site of TGM3 necessary for its activation \(^{32,33}\) . Furthermore, TGM3 is expressed and synthesized in goblet cells, a secretory cell lineage whose secretory granules contain high calcium concentrations for the packing of the MUC2 mucin. Secretory granules can contain calcium concentrations of up to \(40 \mathrm{mM}\) \(^{34}\) . The pH in goblet cell granule is acidic and the \(\mathrm{Ca}^{2 + }\) - ions are bound to MUC2 and the other stored molecules. After secretion, the pH raises, and free \(\mathrm{Ca}^{2 + }\) - ions will become available. For the activation of TGM3, the \(\mathrm{Ca}^{2 + }\) - binding sites must be occupied and the zymogenic form of TGM3 needs to be cleaved in the loop harboring amino acids 462- 469. For this to take place, Cathepsin L or S have been suggested as activating proteases \(^{35}\) . Interestingly, Cathepsin S is a core mucus component and Cathepsin L is also expressed in colonic epithelial and goblet cells \(^{7}\) and Suppl. Fig. S4. This suggests that TGM3 can become fully activated in the colonic mucus and lumen, in line with the endogenously observed TGM3 activity in colon mucus. Overall, the availability of calcium and Cathepsin S and L together with an alkaline pH in the large intestinal lumen provide favorable conditions for TGM3 to catalyze transamidating reactions in colonic mucus.  
+
+The observed intrinsic, extracellular transglutaminase activity led us speculate about the putative functional impact of the natural cross- links in colonic mucus. The comparison of mucus from WT and \(Tgm3^{- / - }\) mice provided direct experimental evidence that the loss of TGM3 led to important biochemical alterations of the dominant mucus skeleton protein MUC2. We observed an extensive degradation of the polypeptide as the N- terminal part with the first three vWD and the first CysD domain was lost comprising approximately 1,300
+
+<--- Page Split --->
+
+amino acids as well as most of the C- terminus. Thus, leaving a central part of the MUC2 mucin consisting of the two highly glycosylated PTS sequences connected via the second CysD domain behind. The resistance of the mucin domains to proteolytic cleavage is due to their dense decoration by \(O\) - linked glycans resulting in steric hindrance to protease degradation. Another feature of MUC2 that was affected in \(Tgm3^{- / - }\) mice is its solubility. MUC2 polymers become insoluble during their transport through the later stages of the secretory pathway \(^{23}\) . In our turbidity assay, an effect on the MUC2 gel network was indicated by an increased optical density in the mucus supernatant from \(Tgm3^{- / - }\) mice. This suggests more soluble MUC2 in this mouse strain. In addition to its solubility, the hydrophobicity of MUC2 was also altered in the absence of TGM3. The assay showed an increased exposure of hydrophobic patches in partly purified MUC2 from \(Tgm3^{- / - }\) animals compared to WT animals upon heat- induced denaturation. This may reflect that the \(\mathrm{N}^6\) (- \(\gamma\) - glutamyl)- lysine bonds in WT- MUC2 stabilize the protein. This phenomenon was partly reverted by preincubating the \(Tgm3^{- / - }\) mucus with recombinant TGM3, supporting the impact of isopeptide bonds on this biophysical parameter. Our observations are consistent with previous studies showing that isopeptide bonds can stabilize bacterial pili proteins in this kind of assay \(^{36,37}\) . However, the insoluble nature of the MUC2 mucin makes it impossible to purify the MUC2 polypeptide in its native conformation, making recording of a specific melting temperature impossible.  
+
+In the \(Tgm3^{- / - }\) mice, the shortened and more degraded MUC2 still seems to be sufficient to provide enough protection for the colonic epithelium. This mouse strain behaves normally and shows no obvious signs of colon inflammation under normal conditions. It is likely that the highly \(O\) - glycosylated mucin domains of MUC2 are sufficient to trap microorganisms and prevent bacteria from reaching the epithelial cells. However, challenging the system by the addition of a mixture of serine proteases deciphered an altered phenotype in \(Tgm3^{- / - }\) animals
+
+<--- Page Split --->
+
+in an ex vivo bead penetration assay. In this approach, the mucus layer seemed to be more disintegrated and less organized allowing the bacteria- mimicking beads to reach the epithelium suggesting a compromised barrier function. We also obtained direct experimental support of this hypothesis by administering DSS to WT and \(Tgm3^{- / - }\) animals. DSS quickly disintegrates the inner colon mucus layer allowing bacteria to reach the epithelium and trigger inflammation typically observed after five days \(^{25}\) . DSS treatment showed that the \(Tgm3^{- / - }\) mice were more susceptible and showed defects already after only two days of treatment. The early onset suggest direct effects on mucus that are likely explained by the decreased inter- molecular cross- links in the mucus of \(Tgm3^{- / - }\) mice. Less cross- links will, as shown here, make the mucus more susceptible to proteolytic degradation and detachment leading to a faster mucus removal by intestinal peristalsis. Interestingly, two transcriptomic studies using either the 2,4,6- trinitrobenzene sulfonic acid or the adoptive T- cell transfer colitis model detected TGM3 downregulation after the establishment of the disease, thereby suggesting an impact of this enzyme for a healthy gut \(^{38, 39}\) . In contrast to non- treated mice, both the WT and \(Tgm3^{- / - }\) mice synthesized the TGM2 enzyme in their colonic mucosa after DSS treatment. This might reflect a role for TGM2 in wound healing as suggested previously \(^{18, 26}\) . As mucus can be regarded as our ‘inner skin’ it is not surprising that a weakened mucus barrier in the \(Tgm3^{- / - }\) mice could be regarded in analogy with the TGM3 function in the skin where earlier observations have shown that these animals have an impaired skin barrier \(^{40}\) .  
+
+This study identifies TGM3 as an important natural cross- linking enzyme acting on the expanded secreted mucus and by this contributing to the stabilization of the colonic mucus gel network. The MUC2 mucin and other mucus components are secreted into the harsh luminal environment where proteases from the host, the commensal bacteria, and eventually from pathogens reside. The TGM3- catalyzed formation of isopeptide bond cross- links strengthens the mucus barrier and thereby increase the mucus protection of the colonic
+
+<--- Page Split --->
+
+epithelium. However, further studies are required to more precisely understand the molecular details for the role of transglutaminases for the mucus structure. For example, as there exists an inverse gradient of TGM2 and 3 abundance from the small to the large intestine it would be interesting to determine the activity of TGM2 and decipher its role for small intestinal mucus. Our observations increase our understanding of the molecular mechanisms that contribute to the architecture of the colonic mucus layers and suggest potential treatment options for the human disease UC.  
+
+## Methods  
+
+Animals  
+
+C57/BL6N mice were from Taconic. \(Tgm2^{- / - }\) mice<sup>41</sup> were provided from Oslo University Hospital (Norway). \(Tgm3^{- / - }\) mice<sup>42</sup> were obtained from the University of Rome (Tor Vergata, Italy). All animal experiments were conducted according to the Swedish legislation (Jordbruksverket; Ethical permits no.: 2285/19 and 2292/19). Mice were maintained at \(22^{\circ}\mathrm{C}\) with light/dark cycles of 12 hours each. Animals received a standard rodent diet and water was supplied ad libitum.  
+
+Antibodies, enzymes, chemicals  
+
+If not otherwise specified chemicals were bought from Sigma. For the detection of TGM2 the monoclonal CUB7402 antibody (Thermo Fisher Scientific) was used for both immunohistochemistry (IHC) and Western Blot. TGM3 detection was performed using the polyclonal NBP1- 57678 antibody (Novus Biologicals) for both applications. Cross- reactivity of the two antibodies was analysed by western blot against recombinant TGM2 and 3 (Suppl. Fig. S1b). For IHC detection of TGM2 a goat- \(\alpha\) - mouse- IgG1 antibody coupled to AlexaFluor647 (Invitrogen) and a goat- \(\alpha\) - rabbit- IgG antibody coupled to AlexaFluor647
+
+<--- Page Split --->
+
+(Invitrogen) for TGM3 detection was used. For Western Blot detection of TGM2 a goat- \(\alpha\) - mouse- IgG1 antibody coupled to the IRdye 680LT (LI- COR and a goat- \(\alpha\) - rabbit- IgG antibody coupled to AlexaFluor 790 (Invitrogen) was used. Trypsin and AspN were from Promega. LysC was from WAKO (Japan). Recombinant TGM2 (T022) and TGM3 (T013) as well as the biotinylated glutamine donor substrates A25 (B001); T26 (B008); E51 (B009) and the biotinylated amine donor compound pentylamine (B002) were bought from Zedira (Germany). The FITC- labelled E51 probe was bought from CovalAb (France). Pronase was from Merck (Germany). The UEA1 lectin was from BioNordika.  
+
+## Immunohistochemistry  
+
+Paraffin- embedded tissue sections were deparaffinized with xylene and rehydrated in ethanol solutions ranging from \(100\%\) to \(30\%\) . Antigen- retrieval was performed by boiling the sections in \(10 \mathrm{mM}\) citrate buffer pH 6.0. The sections were blocked for one hour with \(5\%\) fetal bovine serum (FBS) in PBS. Afterwards the antibodies for TGM2 and 3 were added (1:200 diluted in PBS containing \(5\%\) FBS) and the sections incubated overnight at \(4^{\circ}\mathrm{C}\) in a humid chamber followed by three washing steps in PBS. Secondary antibodies coupled to the AlexaFluor647 dye ( \(\alpha\) - mouse- IgG for TGM2 \(\alpha\) - rabbit- IgG for TGM3, 1:1,000 diluted in PBS containing \(5\%\) FBS) were added together with the UEA1 lectin ( \(10 \mu \mathrm{g / ml}\) ) conjugated to the rhodamine dye for one hour. After three washing steps in PBS the nuclei were stained with the Sytox green stain for five minutes. After one additional washing step the sections were mounted using ProLong Gold- Antifade mountant (Thermo Fisher Scientific) and visualized by confocal microscopy (Zeiss Examiner 2.1; LSM 700).
+
+<--- Page Split --->
+
+Mucus supernatants (30 \(\mu \mathrm{g}\) ) were incubated either with \(10~\mu \mathrm{M}\) of the TGM2- or TGM3- specific glutamine- donor peptides \(\mathrm{T}26^{21}\) or \(\mathrm{E}51^{22}\) or the amine- donor compound 5- Biotinyl- pentalamine for one hour at \(37^{\circ}\mathrm{C}\) . Control reactions were performed in the presence of 25 mM IAA and the respective compounds. The reactions were stopped by the addition of SDS- loading buffer and heating to \(95^{\circ}\mathrm{C}\) for five minutes. Reaction products were separated by SDS- PAGE on \(4 - 15\%\) gradient gels followed by semidry transfer to PVDF membranes. After blocking with \(3\%\) BSA in TBS buffer the membrane was incubated with streptavidin coupled to AlexaFluor 680 (1:20,000, Invitrogen) and the incorporation of substrates revealed on an Odyssey Li- COR Clx workstation.  
+
+Quantitative determination of TGM activity was performed according to the method described by Trigwell and coworkers \(^{43}\) . Briefly, maxisorb 96- well plates (Thermo) were coated with \(250~\mu \mathrm{l}\) of a \(0.1\%\) casein solution in \(50~\mathrm{mM}\) sodium carbonate pH 9.8 for 12 hours. After emptying and washing \(250~\mu \mathrm{l}\) blocking solution ( \(0.1\%\) BSA in \(50~\mathrm{mM}\) sodium carbonate pH 9.8) was added and incubated for one hour at \(37^{\circ}\mathrm{C}\) . After washing, \(150~\mu \mathrm{l}\) reaction buffer ( \(100~\mathrm{mM}\) TrisHCl pH 8.5, \(6.7~\mathrm{mM}\) CaCl2, \(13.3~\mathrm{mM}\) DTT containing either 10 \(\mu \mathrm{M}\) biotinylated TGM- substrate peptide E51; T26, respectively or 5 \(\mu \mathrm{M}\) biotinylated TGM- substrate peptide A25) for the respective TGM standards was added to the wells. For the analysis of mucus samples, DTT and calcium were omitted in the reaction buffer. Measurements were carried out in triplicate per biological replicate. The reactions were started by the addition of either \(50~\mu \mathrm{l}\) TGM standards (0; 25; 50; 75; 100; 125 \(\mathrm{mU / well}\) ) or mucus samples and incubated for one hour at \(37^{\circ}\mathrm{C}\) on a rotational shaker set to \(100~\mathrm{rpm}\) . Afterwards, the reactions were stopped by emptying the wells and washing. The incorporation of the substrates in the casein matrix was probed by the addition of \(200~\mu \mathrm{l}\) Extravidin solution (Extravidin- peroxidase (1:10,000 in \(100~\mathrm{mM}\) TrisHCl pH 8.5 containing
+
+<--- Page Split --->
+
+1% BSA) for one hour and gentle shaking. Biotin- Extravidin binding was visualized by adding \(200\mu \mathrm{l}\) TMB developing solution \((3,3^{\prime},5,5^{\prime}\) - Tetramethylbenzidine, Sigma) and the reaction stopped by adding \(50\mu \mathrm{l}5\mathrm{M}\mathrm{H}_2\mathrm{SO}_4\) . The absorbance of reaction and standard wells was recorded at \(450\mathrm{nm}\) on a Victor2 Wallac work station (Perkin Elmer). The activities of the samples were subsequently normalized against the protein content of the sample using the BCA method.  
+
+Ex vivo analysis of transglutaminase activity  
+
+Ex vivo analysis of transglutaminase activityMice were anaesthetized using isoflourane and sacrificed by cervical dislocation. The colon was collected by dissection and flushed for the removal of intestinal content using Krebs buffer as previously described<sup>44</sup>. After removal of the muscle layer by microdissection the tissue was mounted in an in- house built horizontal chamber allowing basolateral perfusion with Krebs- Glucose buffer and apical Krebs- mannitol buffer (Fig. 2e). Two \(\mu \mathrm{M}\) FITC- labelled E51- probe in Krebs- mannitol buffer was added and the tissue incubated for 30 minutes at \(37^{\circ}\mathrm{C}\) . Afterwards, non- incorporated probe molecules were washed away with Krebs- mannitol buffer followed by analysis of incorporation of the TGM3- substrate peptide on an upright LSM700 confocal microscope (Carl Zeiss, Germany) equipped with a \(20\mathrm{x}\) immersion lens (Pan- Apochromat \(20\mathrm{x} / 1.0\) DIC \(75\mathrm{mm}\) ; Carl Zeiss, Germany). Images were acquired using Zen Black software (Carl Zeiss) and z- stacks were exported to TIFF format using the Imaris software. Inhibition of transglutaminase activity in WT mice was achieved by adding \(5\mu \mathrm{M}\) Z- DON (Zedira) together with the TGM3 substrate.  
+
+Ex vivo mucus integrity assay  
+
+Tissue was collected as described for the ex vivo analysis of transglutaminase activity.  
+
+Following mounting in the perfusion chamber, tissue was stained with Syto 9 (1:500 in
+
+<--- Page Split --->
+
+Kreb's- mannitol buffer; Thermo Fisher) and the mucus layer was visualized by the addition of \(1\mu \mathrm{m}\) fluorescent beads (Thermo Fisher). \(20\mathrm{mg / ml}\) of pronase was added to the apical Krebs- mannitol buffer and the integrity of the mucus layer was monitored on an upright LSM900 confocal microscope (Carl Zeiss) using a water Pan- Apochromat \(20\mathrm{x} / 1.0\mathrm{DIC}75\) mm lens (Carl Zeiss; Germany). Tissue explants were maintained at \(37^{\circ}\mathrm{C}\) throughout the experiments. Briefly, z- stacks were acquired every 5 minutes (total time \(1\mathrm{hr}\) .) using Zen Blue software (version 3.1; Carl Zeiss, Germany). In order to monitor mucus integrity beads and tissue surfaces were mapped to isosurfaces using Imaris software as described previously \(^{45}\) , data regarding the position of the fluorescent beads in relation to the tissue surface over time was then extracted and analyzed to generate normalized positional data over time (Prism version 9.1.0, Graphpad).  
+
+## Colitis induction by DSS  
+
+Age- and sex- matched WT C57/BL6 and \(Tgm3^{- / - }\) mice were cohoused for 4 to 5 weeks. Colitis was induced by adding \(3\%\) (w/v) dextran sodium sulfate (DSS) to the drinking water. Mice could drink ad libitum. The mice were sacrificed after eight days or if their body weight dropped by \(10\%\) from the initial weight. The probability of survival was defined when mice died or if they showed a body weight loss \(>10\%\) . The colon was dissected and its length measured from cecum to anus and subsequently normalized against the initial body weight of the respective animal. Afterwards, the colon was flushed with PBS for the removal of fecal content. The colons were fixed as Swiss rolls in \(4\%\) paraformaldehyde and stained for hematoxylin/eosin and Alcian Blue- PAS. The disease activity index (DAI) was calculated as the sum of the combined scores for stool consistency, hematochezia and weight loss according to the methods of Friedman and co- workers \(^{46}\) . The detection of occult blood was
+
+<--- Page Split --->
+
+performed using a Hemoccult kit (Beckman Coulter) according to the manufacturer’s instructions. Two litters of each mouse strain with five animals per litter were analyzed.  
+
+Composite agarose- PAGE  
+
+The separation of MUC2 was performed according to the protocol of Schulz and coworkers<sup>47</sup>. Briefly, mucus was scraped from mouse colon and emulsified in TBS. Mucus/Muc2 was precipitated by centrifugation at 16,000 x g and 4°C for 30 minutes. The mucus was solubilized by the addition of reducing gel- loading buffer (62.5 mM TrisHCl pH 6.8, 2% SDS, 50 mM DTT 10% (v/v) glycerol). 67 μg were separated via AgPAGE for 3.5 h at 30 mA. The gels were either stained with Alcian Blue or MUC2 was detected by in- gel immunodetection. For in- gel immunodetection, the gels were fixed in 50% (v/v) 2- Propanol/ 5% (v/v) acetic acid for 15 minutes and gentle shaking followed by 30 minutes washing in water. The primary antibody against MUC2 (Genentech; 1:500) was added for 12 hours at 4°C in PBS- T buffer containing 5% BSA. After three washing steps with PBS- T for 10 minutes each, the secondary antibody \(\alpha\) - rabbit- IgG- Licor790 (LiCOR, 1:5000) was added for one hour at ambient temperature. After three to five extensive additional washing steps, the immunostained gel was scanned with a LiCOR Clx instrument.  
+
+## Thermofluor assay  
+
+Mucus from the indicated mouse strains was scraped from their distal colons and emulsified in TBS buffer. Insoluble mucins were washed twice in TBS and recovered by centrifugation (16,000 x g; 4°C; 30 minutes). The protein concentration of the supernatant was determined and the mucus pellet emulsified to a concentration of 1 mg/ml in each sample. 45 μl of sample or TBS control were mixed with five μl of a 200- fold stock solution of SyproOrange (Molecular Probes) and subjected to an increasing temperature gradient of 0.5°C every 30
+
+<--- Page Split --->
+
+seconds from 25 to \(99^{\circ}\mathrm{C}\) in a CFX96 Real- time system (BioRad). The fluorescence was recorded every 30 seconds and the fluorescence intensity of the TBS control subtracted. To rescue the properties of mucus from WT mice 1 U recombinant TGM3 and \(4\mathrm{mM CaCl}_2\) were applied to the mucus from \(Tgm3^{- / - }\) mice and incubated for one hour at \(37^{\circ}\mathrm{C}\) . The reaction was terminated by the addition of \(5\mathrm{mM}\) IAA. The buffer controls for this part of the experiment were treated accordingly and the melting curve recorded as described above. Three biological replicates were analyzed in technical triplicates.  
+
+Analysis of MUC2 depolymerization by turbidity measurement Scraped mucus samples were adjusted to \(1\mathrm{mg / ml}\) and precipitated by centrifugation (1,000 x g, 30 minutes, \(4^{\circ}\mathrm{C}\) ). Afterwards, the turbidity of the supernatant was recorded at \(600\mathrm{nm}\) wavelength in a Spectramax photometer.  
+
+Single cell transcriptomic analysis  
+
+Goblet cells and non- goblet cells from the RedMUC2 reporter mouse strain were isolated by FACS as described recently<sup>20</sup>. The used bulk RNA- seq data (GSE144363) are deposited in GEO and belong to the superserie GSE144436. The quality of the data was assessed with FastQC (version 0.11.2) and filtered using Prinseq (version 0.20.3). The reads were aligned against the mouse reference genome mm10 with STAR (version 2.5.2b) and the number of mapped reads was calculated with HTseq (version 0.6.1p1). Data normalization, differential expression and statistical analysis were made with DESeq2 (version 1.14) in R.  
+
+In- gel digestion and mass spectrometric analyses  
+
+Protein bands of interest were excised from the gel and washed with \(50\%\) acetonitrile and dried in a vacuum centrifuge followed by reduction with DTT and alkylation with IAA.
+
+<--- Page Split --->
+
+Trypsin was added at a ratio of 1:50 and the samples incubated for 12 hours at \(37^{\circ}\mathrm{C}\) . Afterwards, AspN was added at a ratio of 1:50 and the samples incubated for additional 5 hours at \(37^{\circ}\mathrm{C}\) . The digestion was stopped by adding TFA to a concentration of \(0.5\%\) . Salt and buffer components were removed by in- house stage tips equipped with C18 resin \(^{48}\) and the peptides dissolved in \(0.1\%\) formic acid. The samples were analyzed on a Q- Exactive mass spectrometer as described earlier \(^{49}\) .  
+
+## MS Data analysis  
+
+MS raw files were transformed into \*.mgf files using the MS convert software. These files were analyzed using the MASCOT search engine (Matrix Science). Searches were performed against the UniProt database (version 06/2017 containing 554515 sequences) and an in- house database (http://www.medkem.gu.se/mucinbiology/databases/index.html) containing all human and mouse mucin sequences. Searches were performed with the following parameters: mass tolerance for the precursor ion of 5 ppm; tolerance for fragment ions 0.2 Da; full specificity for trypsin/AspN with a maximum of two missed cleavages; carbamidomethylation as static modification and oxidation of methionine as variable modification.  
+
+TGM- catalysed cross- linked peptides were searched using the StavroX software tool (version 3.6.6) \(^{50}\) against theoretical intra- and intermolecular isopeptide cross- linked (di)peptides of the murine MUC2 using the following parameters: mass tolerance for the precursor ion of 2 ppm; tolerance for fragment ions 20 ppm; full specificity for trypsin/AspN with a maximum of three missed cleavages; Gln and Lys as cross- linking sites; composition of the cross- link - NH3; carbamidomethylation as static modification and methionine oxidation as variable modification. Label- free mass spectrometric quantification of TGM isozymes was performed as recently described \(^{20}\) .
+
+<--- Page Split --->
+
+599 Data availability600 The proteomics data set for label- free quantification used has been published<sup>20</sup> and deposited to the601 ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) with the dataset602 identifier PXD011527. The bulk RNA- seq data (GSE144363) are deposited in GEO and belong603 to the superserie GSE144436<sup>20</sup>.604 Statistical analysis605 Statistical analyses were performed using the Prism software (version 9.0.1; GraphPad).606 Body weight and colon length were compared using the unpaired t- test with Welch’s607 correction. DAI scores were compared by multiple unpaired t- tests using the Holm- Sidák608 correction. Significance was accepted when p values were below 0.05. Data are expressed as609 mean ± standard deviation.610  
+
+## Acknowledgements  
+
+We acknowledge Ludvig Sollid, University of Oslo and Eleonara Candi, University of Rome for providing the \(Tgm2^{- / - }\) and \(Tgm3^{- / - }\) mice strains. This work was supported by the European Research Council ERC (694181), National Institute of Allergy and Infectious Diseases (U01AI095473, the content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH), The Knut and Alice Wallenberg Foundation (2017.0028), Swedish Research Council (2017- 00958), The Swedish Cancer Foundation (CAN 2017/360), IngaBritt and Arne Lundberg Foundation (2018- 0117), Sahlgren's University Hospital (ALFGBG- 440741, The ALF agreement 236501), Bill and Melinda Gates Foundation (OPP1202459), Wilhelm and Martina Lundgren's Foundation.
+
+<--- Page Split --->
+
+## 624 Author contributions  
+
+624 Author contributionsJDAS performed experiments and analyzed data; BD performed experiments and analyzed data; EELN performed experiments and analyzed data, LA performed experiments and analyzed data; GMHB performed experiments and analyzed data; BMA performed experiments and analyzed data; MEVJ data analysis; GCH conceptualized the study, analyzed data; CVR conceptualized the study, performed experiments and analyzed data. GCH and CVR wrote the paper. All authors reviewed the paper and accepted the final version.  
+
+## 633 Competing interests  
+
+634 The authors declare no competing interests.  
+
+## 636 References  
+
+637 1. Johansson, M.E. & Hansson, G.C. Immunological aspects of intestinal mucus and mucins. Nat Rev Immunol 16, 639- 649 (2016). 639 2. Rodriguez- Pineiro, A.M. et al. Studies of mucus in mouse stomach, small intestine, and colon. II. Gastrointestinal mucus proteome reveals Muc2 and Muc5ac accompanied by a set of core proteins. Am J Physiol Gastrointest Liver Physiol 305, G348- 356 (2013). 642 3. Atuma, C., Strugala, V., Allen, A. & Holm, L. The adherent gastrointestinal mucus gel layer: thickness and physical state in vivo. Am J Physiol Gastrointest Liver Physiol 280, G922- 929 (2001). 645 4. Johansson, M.E. et al. The inner of the two Muc2 mucin- dependent mucus layers in colon is devoid of bacteria. Proc Natl Acad Sci U S A 105, 15064- 15069 (2008). 647 5. Van der Sluis, M. et al. Muc2- deficient mice spontaneously develop colitis, indicating that MUC2 is critical for colonic protection. Gastroenterology 131, 117- 129 (2006). 649 6. Velcich, A. et al. Colorectal cancer in mice genetically deficient in the mucin Muc2. Science 295, 1726- 1729 (2002). 651 7. van der Post, S. et al. Structural weakening of the colonic mucus barrier is an early event in ulcerative colitis pathogenesis. Gut 68, 2142- 2151 (2019). 652 8. Svensson, F., Lang, T., Johansson, M.E.V. & Hansson, G.C. The central exons of the human MUC2 and MUC6 mucins are highly repetitive and variable in sequence between individuals. Sci Rep 8, 17503 (2018). 656 9. Hansson, G.C. Mucins and the Microbiome. Annu Rev Biochem 89, 769- 793 (2020). 657 10. Godl, K. et al. The N terminus of the MUC2 mucin forms trimers that are held together within a trypsin- resistant core fragment. The Journal of biological chemistry 277, 47248- 47256 (2002). 660 11. Javitt, G. et al. Assembly Mechanism of Mucin and von Willebrand Factor Polymers. Cell 183, 717- 729 e716 (2020).
+
+<--- Page Split --->
+
+662 12. Recktenwald, C.V. & Hansson, G.C. The Reduction-insensitive Bonds of the MUC2 Mucin Are Isopeptide Bonds. The Journal of biological chemistry 291, 13580- 13590 (2016). 663 13. Lorand, L. & Graham, R.M. Transglutaminases: crosslinking enzymes with pleiotropic functions. Nat Rev Mol Cell Biol 4, 140- 156 (2003). 665 14. Folk, J.E. & Finlayson, J.S. The epsilon-(gamma-glutamyl)lysine crosslink and the catalytic role of transglutaminases. Adv Protein Chem 31, 1- 133 (1977). 667 15. Eckert, R.L. et al. Transglutaminase regulation of cell function. Physiol Rev 94, 383- 417 (2014). 670 16. Adamczyk, M., Griffiths, R., Dewitt, S., Knauper, V. & Aeschlimann, D. P2X7 receptor activation regulates rapid unconventional export of transglutaminase- 2. J Cell Sci 128, 4615- 4628 (2015). 673 17. Siegel, M. et al. Extracellular transglutaminase 2 is catalytically inactive, but is transiently activated upon tissue injury. PLoS One 3, e1861 (2008). 675 18. Telci, D. & Griffin, M. Tissue transglutaminase (TG2)- - a wound response enzyme. Front Biosci 11, 867- 882 (2006). 677 19. Eckert, R.L., Sturniolo, M.T., Broome, A.M., Ruse, M. & Rorke, E.A. Transglutaminase function in epidermis. J Invest Dermatol 124, 481- 492 (2005). 679 20. Nystrom, E.E.L. et al. An intercrypt subpopulation of goblet cells is essential for colonic mucus barrier function. Science 372 (2021). 681 21. Sugimura, Y. et al. Screening for the preferred substrate sequence of transglutaminase using a phage- displayed peptide library: identification of peptide substrates for TGASE 2 and Factor XIII. The Journal of biological chemistry 281, 17699- 17706 (2006). 684 22. Yamane, A. et al. Identification of a preferred substrate peptide for transglutaminase 3 and detection of in situ activity in skin and hair follicles. FEBS J 277, 3564- 3574 (2010). 686 23. Axelsson, M.A., Asker, N. & Hansson, G.C. O- glycosylated MUC2 monomer and dimer from LS 174T cells are water- soluble, whereas larger MUC2 species formed early during biosynthesis are insoluble and contain nonreducible intermolecular bonds. The Journal of biological chemistry 273, 18864- 18870 (1998). 690 24. Kasdorf, B.T. et al. Mucin- Inspired Lubrication on Hydrophobic Surfaces. Biomacromolecules 18, 2454- 2462 (2017). 692 25. Johansson, M.E. et al. Bacteria penetrate the inner mucus layer before inflammation in the dextran sulfate colitis model. PLoS One 5, e12238 (2010). 694 26. Jeong, E.M. et al. Transglutaminase 2 is dispensable but required for the survival of mice in dextran sulfate sodium- induced colitis. Exp Mol Med 48, e267 (2016). 696 27. Stamnaes, J., Iversen, R., du Pre, M.F., Chen, X. & Sollid, L.M. Enhanced B- Cell Receptor Recognition of the Autoantigen Transglutaminase 2 by Efficient Catalytic Self- Multimerization. PLoS One 10, e0134922 (2015). 698 28. Gross, S.R., Balklava, Z. & Griffin, M. Importance of tissue transglutaminase in repair of extracellular matrices and cell death of dermal fibroblasts after exposure to a solarium ultraviolet A source. J Invest Dermatol 121, 412- 423 (2003). 702 29. Yildiz, H.M., Speciner, L., Ozdemir, C., Cohen, D.E. & Carrier, R.L. Food- associated stimuli enhance barrier properties of gastrointestinal mucus. Biomaterials 54, 1- 8 (2015). 704 30. Di Maio, S. & Carrier, R.L. Gastrointestinal contents in fasted state and post- lipid ingestion: in vivo measurements and in vitro models for studying oral drug delivery. J Control Release 151, 110- 122 (2011). 707 31. Wongdee, K., Rodrat, M., Teerapornpuntakit, J., Krishnamra, N. & Charoenphandhu, N. Factors inhibiting intestinal calcium absorption: hormones and luminal factors that prevent excessive calcium uptake. J Physiol Sci 69, 683- 696 (2019). 710 32. Ahvazi, B., Kim, H.C., Kee, S.H., Nemes, Z. & Steinert, P.M. Three- dimensional structure of the human transglutaminase 3 enzyme: binding of calcium ions changes structure for activation. EMBO J 21, 2055- 2067 (2002).
+
+<--- Page Split --->
+
+713 33. Ahvazi, B., Boeshans, K.M., Idler, W., Baxa, U. & Steinert, P.M. Roles of calcium ions in the activation and activity of the transglutaminase 3 enzyme. The Journal of biological chemistry 278, 23834- 23841 (2003). 716 34. Yoo, S.H. Secretory granules in inositol 1,4,5-trisphosphate-dependent Ca2+ signaling in the cytoplasm of neuroendocrine cells. FASEB J 24, 653- 664 (2010). 717 35. Cheng, T. et al. Cystatin M/E is a high affinity inhibitor of cathepsin V and cathepsin L by a reactive site that is distinct from the legumain-binding site. A novel clue for the role of cystatin M/E in epidermal cornification. The Journal of biological chemistry 281, 15893- 15899 (2006). 722 36. Walden, M., Crow, A., Nelson, M.D. & Banfield, M.J. Intramolecular isopeptide but not internal thioester bonds confer proteolytic and significant thermal stability to the S. pyogenes pilus adhesin Spy0125. Proteins 82, 517- 527 (2014). 725 37. Hagan, R.M. et al. NMR spectroscopic and theoretical analysis of a spontaneously formed Lys-Asp isopeptide bond. Angew Chem Int Ed Engl 49, 8421- 8425 (2010). 727 38. Wu, F. & Chakravarti, S. Differential expression of inflammatory and fibrogenic genes and their regulation by NF- kappaB inhibition in a mouse model of chronic colitis. J Immunol 179, 6988- 7000 (2007). 730 39. Lyons, J. et al. Integrated in vivo multiomics analysis identifies p21- activated kinase signaling as a driver of colitis. Sci Signal 11 (2018). 732 40. Bognar, P. et al. Reduced inflammatory threshold indicates skin barrier defect in transglutaminase 3 knockout mice. J Invest Dermatol 134, 105- 111 (2014). 734 41. De Laurenzi, V. & Melino, G. Gene disruption of tissue transglutaminase. Mol Cell Biol 21, 148- 155 (2001). 736 42. Frezza, V. et al. Transglutaminase 3 Protects against Photodamage. J Invest Dermatol 137, 1590- 1594 (2017). 738 43. Trigwell, S.M., Lynch, P.T., Griffin, M., Hargreaves, A.J. & Bonner, P.L. An improved colorimetric assay for the measurement of transglutaminase (type II) -(gamma- glutamyl) lysine cross- linking activity. Anal Biochem 330, 164- 166 (2004). 741 44. Gustafsson, J.K. et al. An ex vivo method for studying mucus formation, properties, and thickness in human colonic biopsies and mouse small and large intestinal explants. Am J Physiol Gastrointest Liver Physiol 302, G430- 438 (2012). 743 45. Birchenough, G.M., Nystrom, E.E., Johansson, M.E. & Hansson, G.C. A sentinel goblet cell guards the colonic crypt by triggering NLRp6- dependent Muc2 secretion. Science 352, 1535- 1542 (2016). 747 46. Friedman, D.J. et al. From the Cover: CD39 deletion exacerbates experimental murine colitis and human polymorphisms increase susceptibility to inflammatory bowel disease. Proc Natl Acad Sci U S A 106, 16788- 16793 (2009). 750 47. Schulz, B.L., Packer, N.H. & Karlsson, N.G. Small- scale analysis of O- linked oligosaccharides from glycoproteins and mucins separated by gel electrophoresis. Anal Chem 74, 6088- 6097 (2002). 753 48. Rappsilber, J., Ishihama, Y. & Mann, M. Stop and go extraction tips for matrix- assisted laser desorption/ionization, nanoelectrospray, and LC/MS sample pretreatment in proteomics. Anal Chem 75, 663- 670 (2003). 756 49. Fernandez- Blanco, J.A. et al. Attached stratified mucus separates bacteria from the epithelial cells in COPD lungs. JCI Insight 3 (2018). 758 50. Gotze, M. et al. StavroX- - a software for analyzing crosslinked products in protein interaction studies. J Am Soc Mass Spectrom 23, 76- 87 (2012).
+
+<--- Page Split --->
+
+## Figure Legends  
+
+## Figure 1: mRNA expression, protein abundance, and spatial localization of TGM isozymes in the large intestine.  
+
+(a) mRNA-seq expression data of the goblet cell and non-goblet cell fraction from a reporter mouse strain expressing fluorescently-labelled MUC2. Goblet cells were separated from other epithelial cell types using FACS-mediated cell sorting<sup>20</sup>. The graph shows the normalized expression levels of the transglutaminase family members \(Tgm1 - 7\) and \(F13a1\) in the goblet cell and non-goblet cell fraction. Four biological replicates were analyzed.  
+
+(b) Label-free relative quantification of TGM isozymes 2 and 3 in goblet cell and remaining epithelial cells after FACS-mediated cell sorting from RedMUC2<sup>98trTg</sup> mice<sup>20</sup>. After protein extraction, the abundance of TGM2 and 3 in the two fractions was measured by mass spectrometry and the data analyzed using the MaxQuant software. Four biological replicates were analyzed.  
+
+(c) Confocal microscopy of large intestinal tissue specimens from C57/BL6, \(Tgm3^{-/ - }\) and \(Tgm2^{-/ - }\) mice suggests no TGM2 biosynthesis in the colon. The sections were probed with a monoclonal antibody against TGM2 followed by detection with a secondary antibody coupled to Alexa Fluor 647 (red) and sections counterstaining with the UEA1 lectin coupled to rhodamine (green) for goblet cell and mucus visualization. Nuclei are shown in grey and were visualized using the Sytox green stain. The scale bar corresponds to 20 μm. Images are representative of three biological replicates.  
+
+(d) Analogously, confocal microscopy of colon specimen from C57/BL6, \(Tgm3^{-/ - }\) and \(Tgm2^{-/ - }\) mice analyzed for TGM3 (red) using a polyclonal anti-TGM3 antibody that was detected by a secondary antibody coupled to Alexa 647 indicating TGM3 biosynthesis in WT and \(Tgm2^{-/ - }\)
+
+<--- Page Split --->
+
+mice. UEA1 (green) and Hoechst (grey) were used for counterstaining. Images are representative of three biological replicates. The scale bar corresponds to \(20 \mu \mathrm{m}\) .  
+
+(e) Protein abundance analysis of TGM isoforms by Western blot in colonic mucus. The supernatant of precipitated mucus was analyzed for the presence of TGM2 and 3 using a monoclonal anti-TGM2 antibody and a polyclonal anti-TGM3 antibody. Goat anti-mouse IgG1-isoform antibody coupled to an IR680 dye and anti-rabbit IgG's coupled to an IR790 dye were used for visualization on a LI-COR Odyssey Clx workstation. Recombinant non-activated or calcium-activated TGM2 and 3 were loaded as positive controls. The red dashed line marks the IgG1 heavy chain (IgG1-HC) recognized by the secondary antibody against the TGM2 antibody and served as loading control. A representative analysis with three biological replicates per mouse strain is shown.  
+
+## Figure 2: Qualitative, quantitative and ex vivo analysis of extracellular  
+
+# transglutaminase activity.  
+
+(a) Qualitative determination of calcium-induced transglutaminase activity in colonic mucus samples. Samples from the indicated strains were spiked with biotinylated TGM2 (T26) and TGM3 (E51) selective acyl-acceptor peptide substrates and with calcium addition in the absence or presence of IAA followed by incubation for one hour at \(37^{\circ}\mathrm{C}\) . The reaction products were separated by SDS-PAGE and subsequently visualized by Western blot using streptavidin labelled with an IR680LT-dye on a LiCOR Odyssey Clx imager. Non-specific signals from endogenously biotinylated proteins were marked with a triangle. A representative example of three biological replicates per mouse strain is shown.  
+(b) Qualitative determination of intrinsic transglutaminase activity in colonic mucus samples. Samples from the indicated strains were supplied with biotinylated TGM2 (T26) and TGM3-(E51) specific acyl-acceptor peptide substrates without calcium addition in the absence or
+
+<--- Page Split --->
+
+presence of IAA and incubated for one hour at \(37^{\circ}\mathrm{C}\) . The reaction products were separated by SDS- PAGE followed by Western blot detection using streptavidin labelled with an IR680LT- dye. Non- specific signals from endogenously biotinylated proteins were marked with a triangle.A representative example of three biological replicates per mouse strain is shown. (c) Detection of putative acyl- acceptor proteins in mucus. Mucus samples from the different mouse strains were incubated in the presence of 5- Bioinyl- pentalamine (5- BP) for one hour at \(37^{\circ}\mathrm{C}\) . Control reactions were performed in the presence of IAA for the visualization of false- positive signals. The incorporation of 5- BP was detected by Western Blot using streptavidin labelled with an IR680LT- dye. Non- specific signals from endogenously biotinylated proteins were marked with a triangle. A representative example of three biological replicates per mouse strain is shown.  
+
+(d) Quantitative determination of transglutaminase activity in colonic mucus samples. TGM activity in mucus from the different mouse strains was determined by the incorporation of TGM2 (T26) and TGM3 (E51) specific peptide substrates or a promiscuous TGM acyl-acceptor peptide (A25) into casein as described under materials and methods. The respective cross-linking activity in the samples was calculated from the calibration curve of the recombinant activated TGM standards and subsequently normalized against the protein concentration of the samples. At least four biological replicates per substrate and mouse strain were analyzed.  
+
+(e- h) Ex vivo analysis of transglutaminase activity. Tissues were mounted in a perfusion chamber as illustrated (e) and transglutaminase activity probed with the glutamine- donor peptide E51 coupled to FITC (magenta) for 30 minutes at \(37^{\circ}\mathrm{C}\) . After washing away non- incorporated peptide the tissue specimen were analyzed by confocal microscopy. Mucus and nuclei were counterstained with the UEA1 lectin coupled to rhodamine (green) and the Hoechst stain (blue) respectively. The top panels show Z-stacks of the explant with (left) or
+
+<--- Page Split --->
+
+without (right) the UEA1 counterstain. The bottom panels show x/y projections of the indicated area from the respective Z- stack on top. Colonic specimen from WT mice (f), \(Tgm3^{- / - }\) mice (g) or WT mice in the presence of the pan- TGM inhibitor Z- DON (h) were probed for E51 incorporation. The scale bar corresponds to \(50 \mu \mathrm{m}\) . Three animals per mouse strain were analyzed.  
+
+## Figure 3: Loss of TGM3 causes biochemical alterations of mucus and MUC2.  
+
+(a) Schematic figure of the human and mouse MUC2 domain structure. The domains of the complete sequence excluding the signal sequence are shown. The abbreviations correspond to vWD, von-Willebrand D domain; CysD, Cystein-rich domain; PTS, Proline, Serine, Threonine-rich domain that after \(O\) -glycosylation becomes a mucin domain; vWC, von-Willebrand C domain; CK, Cysteine-knot domain.  
+
+(b) MUC2 mono- and oligomers from WT, \(Tgm2^{- / - }\) , and \(Tgm3^{- / - }\) colonic mucus were separated by composite AgPAGE and stained by in-gel immunodetection using a polyclonal anti-MUC2-C3 (Genentech) and secondary antibody coupled to the AlexaFlour790-dye on an Odyssey Clx workstation.  
+
+(c) Limited Proteolysis of MUC2 by the serine protease Lys-C. Mucus samples from the indicated mouse strains were incubated in the absence or presence of Lys-C for 90 minutes at \(25^{\circ}\mathrm{C}\) and the reaction products separated via composite AgPAGE followed by visualization of MUC2 with Alcian Blue.  
+
+(d) Heat map of the sequence coverage of MUC2 domains. The color coded sequence coverage of the different MUC2 domains from three biological replicates of non-treated MUC2 monomers from mucus samples of WT (WT-M) and \(Tgm3^{- / - }\) ( \(Tgm3^{- / - }\) -M) animals as indicated in Fig. 3c are shown. The various MUC2 domains are placed from the N-terminus (top) to the C-terminus (bottom) on the ordinate. Only peptides with an ion score \(>25\) were
+
+<--- Page Split --->
+
+taken into consideration. The two PTS domains were excluded as they are due to their high glycosylation heterogeneity not analyzable.  
+
+(e) Detection of transglutaminase reaction products. Example of an isopeptide dipeptide cross-link that was solely detected in MUC2 from WT animals. MS2 fragment spectrum of the parent ion [M+2H]2+ 775.44 is shown. B ions are labelled in red and y ions in blue. The prarent ion is labelled in green.  
+
+(f) Analysis of MUC2 polymerisation. Mucus samples from WT and the TGM knock-out strains and their protein concentrations adjusted to 1 mg/ml. After centrifugation (1,000 x g, 30 minutes, 4°C), precipitating the insoluble MUC2, the absorbance for soluble material was recorded at 600 nm. Three biological replicates per strain were analyzed.  
+
+(g) Hydrophobicity analysis of mucus from WT and \(Tgm3^{-/ - }\) mice. Mucus samples were adjusted to a protein concentration 1 mg/ml in TBS. SyproOrange was added and the melting curve of the samples analyzed by increasing the temperature by 0.5°C every 30 seconds from 25°C to 99°C in a thermocycler. The fluorescence change was recorded every 30 seconds and subsequently normalized by subtraction of the buffer control fluorescence for each data point. For rescuing the WT behavior, 1 U of activated recombinant TGM3 was added to mucus from \(Tgm3^{-/ - }\) animals and the samples incubated for 60 minutes at 37°C. Afterwards the cross-linking activity was inhibited with IAA before adding SyproOrange. The graph shows the arithmetic average of three biological replicates.  
+
+(h-i) Pronase-treatment of distal colon specimen from WT (h) and \(Tgm3^{-/ - }\) (i) mice. Colonic explants from WT and \(Tgm3^{-/ - }\) mice were mounted in a chamber and pronase in Krebs-buffer was added before examination under a confocal microscope as sketched in Fig. 2e. The mucus surface was visualized by placing fluorescently labelled beads with a diameter of one \(\mu \mathrm{m}\) on top of the mucus layer and the epithelium counterstained using the Syto 9 stain. The top panel shows the isosurfaces of the tissue and of the individual beads over time. The white
+
+<--- Page Split --->
+
+scale bar corresponds to \(50 \mu \mathrm{m}\) . The lower panel shows the distribution of the fluorescently- labelled beads in relation to the tissue surface as violin plot where the black bar marks the median of bead distance from the epithelium. Three animals per mouse strain were analysed. OM=outer (loose) mucus layer, IM=inner mucus layer.  
+
+## Figure 4: Dextran sodium sulfate treatment shows decreased mucus protection.  
+
+WT and \(\mathrm{Tgm3 - / - }\) mice were cohoused and supplied via drinking water with \(3\%\) (w/v) dextran sodium sulfate.  
+
+(a) Body weight change of DSS-treated mice over time. The body weight of the mice was recorded once per day throughout the whole experiment and the change in body weight respective to the starting body weight of both groups plotted against the time. The graph shows the comparison of one litter per strain consisting of five animals in each group.  
+
+(b) Detection of occult blood in feces. Fecal samples were collected from the DSS-treated animals and analyzed for hidden blood using a hemoccult kit as described in materials and methods. The mean ratio of hemoccult-positive samples from each group was plotted against time. The graph shows the comparison of one litter per strain consisting of five animals in each group.  
+
+(c) The disease activity index (DAI) was determined as sum of the changes in body weight, stool consistency, rectal bleeding for every animal for the indicated time points and the mean with standard deviation for the both groups plotted against the time. \(*p< 0.05\) . The graph shows the comparison of one litter per strain consisting of five animals in each group.  
+
+(d) Survival analysis of DSS-treated mice. The probability of survival was calculated using the GraphPad prism software. Mice were sacrificed when the initial body weight loss exceeded \(10\%\) . The graph shows the summary of the two litters per strain that were analysed independently and represents ten animals in each group.
+
+<--- Page Split --->
+
+(e-f) Colon length changes of DSS-treated WT and \(Tgm3^{- / - }\) mice. At day 8 or at the ethical endpoint, animals were sacrificed and the colon length of each animal measured. A representative colon of WT and \(Tgm3^{- / - }\) animals is shown in (e). (f) Normalized colon length of DSS-treated WT and \(Tgm3^{- / - }\) mice. The graph shows the summary of the two litters per strain that were analysed independently and represents ten animals in the WT and nine animals in the \(Tgm3^{- / - }\) group. \(p< 0.001\) .  
+
+(g) Histological analysis of DSS-treated WT and \(Tgm3^{- / - }\) mice. Representative Alcian Blue-Periodic Acid Schiff-stained sections from proximal (PC) and distal colon (DC) of WT and TGM3-deficient animals are shown. The black scale bar on the left corresponds to \(100 \mu \mathrm{m}\) . (h) Immunohistochemical analysis of WT and \(Tgm3^{- / - }\) mice for the presence of TGM2 after DSS treatment. Tissue specimen from WT and \(Tgm3^{- / - }\) animals were probed with a monoclonal anti-TGM2 antibody (red) and the UEA1 lectin (green). Nuclei were stained using the Hoechst stain (grey). The white scale bar on the right corresponds to \(30 \mu \mathrm{m}\) .
+
+<--- Page Split --->
+![PLACEHOLDER_36_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_37_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_38_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_39_0]
+
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+SupplementaryFiguresS14. pdf SupplementaryMovieM1. mp4 SupplementaryMovieM2. mp4 SupplementaryMovieM3. mp4
+
+<--- Page Split --->

preprint/preprint__7e6200f0e6beda7a0fa07dea90dcb4211a3ae35c56dedd427f32bb24a1ac1e71/preprint__7e6200f0e6beda7a0fa07dea90dcb4211a3ae35c56dedd427f32bb24a1ac1e71_det.mmd

@@ -0,0 +1,473 @@
+<|ref|>title<|/ref|><|det|>[[44, 107, 937, 175]]<|/det|>
+# Transglutaminase 3 crosslinks secreted MUC2 and stabilizes the colonic mucus layer  
+
+<|ref|>text<|/ref|><|det|>[[44, 196, 630, 234]]<|/det|>
+Jack Sharpen University of Manchester https://orcid.org/0000- 0002- 6268- 3244  
+
+<|ref|>text<|/ref|><|det|>[[44, 240, 274, 280]]<|/det|>
+Brendan Dolan University of Gothenburg  
+
+<|ref|>text<|/ref|><|det|>[[44, 288, 274, 328]]<|/det|>
+Elisabeth Nystrom University of Gothenburg  
+
+<|ref|>text<|/ref|><|det|>[[44, 335, 630, 374]]<|/det|>
+George Birchenough University of Gothenburg https://orcid.org/0000- 0003- 2283- 2353  
+
+<|ref|>text<|/ref|><|det|>[[44, 380, 274, 420]]<|/det|>
+Liisa Arike University of Gothenburg  
+
+<|ref|>text<|/ref|><|det|>[[44, 427, 630, 466]]<|/det|>
+Beatriz Martinez- Abad University of Gothenburg https://orcid.org/0000- 0002- 0521- 3473  
+
+<|ref|>text<|/ref|><|det|>[[44, 473, 630, 513]]<|/det|>
+Malin Johansson University of Gothenburg https://orcid.org/0000- 0002- 4237- 6677  
+
+<|ref|>text<|/ref|><|det|>[[44, 519, 630, 559]]<|/det|>
+Gunnar Hansson University of Gothenburg https://orcid.org/0000- 0002- 1900- 1869  
+
+<|ref|>text<|/ref|><|det|>[[44, 564, 630, 604]]<|/det|>
+Christian Recktenwald ( \(\square\) christian.recktenwald@medkem.gu.se) University of Gothenburg https://orcid.org/0000- 0003- 1710- 1863  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 650, 102, 667]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 687, 655, 707]]<|/det|>
+Keywords: colonic mucus layer, TGM3, transglutaminase, MUC2 mucin  
+
+<|ref|>text<|/ref|><|det|>[[44, 725, 299, 744]]<|/det|>
+Posted Date: June 21st, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 762, 462, 782]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 555255/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 800, 909, 843]]<|/det|>
+License: \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 878, 940, 921]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on January 11th, 2022. See the published version at https://doi.org/10.1038/s41467- 021- 27743- 1.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[71, 84, 844, 140]]<|/det|>
+# Transglutaminase 3 crosslinks secreted MUC2 and stabilizes the colonic mucus layer  
+
+<|ref|>text<|/ref|><|det|>[[70, 155, 870, 408]]<|/det|>
+3 Jack D. A. Sharpen, Brendan Dolan, Elisabeth E. L. Nyström, George M. H. Birchenough, Liisa Arike, Beatriz Martinez- Abad, Malin E. V. Johansson, Gunnar C. Hansson and Christian V. Recktenwald\* 7 8 From the Department of Medical Biochemistry, University of Gothenburg, SE- 405 30 Gothenburg, Sweden 10 \*Correspondence to: christian.recktenwald@medkem.gu.se  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 457, 197, 472]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[113, 487, 875, 867]]<|/det|>
+The colonic mucus layer is organized as a two- layered system providing a physical barrier against pathogens and simultaneously harboring the commensal flora. The factors contributing to the organization of this gel network are not well understood. In this study, the impact of transglutaminase activity on this architecture was analyzed. Here, we show that transglutaminase TGM3 is the major TGM isoform expressed and synthesized in the colon. Furthermore, intrinsic extracellular TGM activity in the secreted mucus was demonstrated in vitro and ex vivo. Absence of this acyl- transferase activity resulted in faster degradation of the major mucus component the MUC2 mucin and changed the biochemical properties of mucus. Finally, TGM3- deficient mice showed an early increased susceptibility to DSS- induced colitis. Thus, these observations suggest that natural isopeptide cross- linking by TGM3 is important for mucus homeostasis and protection of the colon from inflammation, a suggested pre- stage of colon carcinoma.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 85, 231, 101]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[115, 115, 877, 465]]<|/det|>
+The epithelium in the intestinal tract is covered by mucus that provides protection from luminal challenges and bacterial infiltration \(^{1}\) . Despite the similar proteome composition, the organization of the mucus gel network differs considerably in the small and large intestine \(^{2}\) . Whereas small intestinal mucus is non- attached, the colonic mucus is a two- layered system with an attached, bacteria- free inner layer and an outer layer harboring the commensal flora \(^{1}\) , \(^{3,4}\) . The molecular mechanisms determining these structural differences are not well understood. The predominant component of mucus is the gel- forming MUC2 mucin that is synthesized by intestinal goblet cells. It has been shown that Muc2 \(^{2 / 3}\) mice develop spontaneous colitis, a pre- stage of colon carcinoma \(^{5,6}\) . Furthermore, the MUC2 levels in patients suffering from active ulcerative colitis (UC) are decreased when compared to healthy control patients \(^{7}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 476, 877, 890]]<|/det|>
+The human MUC2 monomer consists of 5,130 amino acids organized in three complete and one partial von Willebrand D (VWD) domains in the N- terminal part followed by the first CysD domain and two Proline-, Threonine- and Serine- rich (PTS) sequences that are separated by the second CysD domain \(^{8,9}\) . The C- terminus harbors a fourth vWD domain, two vWC domains, and the cysteine- knot. During its transport through the endoplasmic reticulum and the Golgi- network MUC2 monomers first form C- terminal dimers and in the later stages of the secretory pathway N- terminal dimers or trimers \(^{10,11}\) . Furthermore, the PTS sequences become heavily O- glycosylated to form mucin domains. This posttranslational modification (PTM) shifts the mass of MUC2 from roughly 650 kDa to more than 2.5 MDa. During the later stages of the secretory pathway isopeptide bonds are introduced probably contributing to the insolubility of MUC2 in chaotropic salts, like guanidinium chloride \(^{12}\) . An enzyme family that is able to catalyze these natural protein cross- links are transglutaminases (TGM).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 84, 875, 530]]<|/det|>
+Transglutaminases (R:protein- glutamine \(\gamma\) - glutamyltransferases; E. C. 2.3.2.13) comprise a family of \(\mathrm{Ca^{2 + }}\) - dependent acyl- transferases that can catalyze the transamidation or deamidation of protein- bound glutamine residues that can lead to natural cross- links through the formation of an isopeptide bond between the side chains of glutamine and lysine. This PTM is known to limit protein degradation by conformational changes and modification of protease- labile Lys residues \(^{13,14}\) . There are nine mammalian TGMs where TGM2 is the most ubiquitously expressed isoform \(^{13,15}\) . This isoform is predominantly localized in the cell cytosol, but can also be found associated with the plasma membrane. Furthermore, it can be secreted by unknown mechanisms after P2X7 receptor activation \(^{16}\) . The enzymatic activity of TGM2 is normally silent but during mechanical injury it becomes activated and acts as a wound healing enzyme by stabilizing extracellular matrix (ECM) and cell- ECM interactions \(^{17,18}\) . Another process where TGMs are important is the morphogenesis of the skin. Here, TGM1, 3 and 5 are involved in the formation of the stratum corneum by cross- linking the envelope precursors such as inloricrin and involucrin \(^{19}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 541, 870, 890]]<|/det|>
+Whether transamidation also has a role in the formation and stabilization of intestinal mucus is currently unknown. Mucus and mucins are stored highly concentrated in the granules of goblet cells and expand 1,000- fold in volume upon secretion. If TGM- catalyzed isopeptide cross- links contribute to mucus homoeostasis, this processing has to occur after secretion and expansion. Here, we suggest that extracellular TGM activity plays a role in organizing the mucus gel in the colon, especially by increasing its stability. To test this hypothesis the abundance of different TGM isozymes was evaluated and their enzymatic activity determined. We found that the formation of \(\mathrm{N^{E}}\) - ( \(\gamma\) - glutamyl)- lysine isopeptide cross- links in colonic mucus was based on extracellular TGM3- intrinsic activity. Furthermore, mice lacking this TGM isoform secrete a more protease- sensitive MUC2 molecule. In addition, \(Tgm3^{i / c}\) mice are less protected against dextran sodium sulfate (DSS) induced
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 870, 166]]<|/det|>
+colitis. Together, our observations indicate that TGM- catalyzed cross- links are important for the stabilization/homoeostasis of colonic mucus and its resistance against disease- inducing conditions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 216, 185, 232]]<|/det|>
+## Results  
+
+<|ref|>text<|/ref|><|det|>[[115, 245, 876, 895]]<|/det|>
+Transglutaminase TGM 3 is a dominant cross- linking enzyme in the colonFirstly, we determined which transglutaminase isozymes were expressed and synthesized in the colonic epithelium. Mouse colon tissue of wild- type (WT) and Tgm knock- out mice were analyzed for protein abundance by using immunohistochemistry (IHC), mass spectrometry (MS) and gel electrophoresis followed by western blot. As we were interested on the impact of transglutaminases on mucus homeostasis a recently published single cell transcriptomic study<sup>20</sup> analyzing MUC2- producing goblet cells and non- goblet epithelial cells was mined for the expression profile and protein abundance of the various TGM family members. Analyzing mRNA levels in colonic goblet cells and the remaining epithelial cell populations revealed only transcripts for Tgm2 and Tgm3 genes (Fig. 1a). Next, the TGM2 and TGM3 protein abundance determined by mass spectrometry (MS) in these two cell fractions was extracted. This method revealed approximately 10- times lower levels of TGM3 in the goblet cells compared to the non- goblet epithelial cells whereas the abundance of TGM2 was two- three orders of magnitude lower than TGM3 in the respective cell population (Fig. 1b). To evaluate the tissue localization of TGM2 and TGM3, immunohistochemical analyses were performed in WT, Tgm2<sup>-/-</sup> and Tgm3<sup>-/-</sup> animals together with the UEA1 lectin staining for the highly glycosylated MUC2 mucin. None of the strains reacted with the anti- TGM2 antibody, confirming the low levels of this isoform (Fig. 1c). That this antibody was functional was tested on duodenal tissue sections where a signal for TGM2 was easily observed (Suppl. Fig. S1a). In line with the quantitative data from mRNA expression and protein abundance, both
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 875, 136]]<|/det|>
+WT and \(Tgm2^{- / - }\) animals showed a strong staining for the TGM3 isoenzyme in the epithelium and as expected no signal in \(Tgm3^{- / - }\) mice (Fig. 1d).  
+
+<|ref|>text<|/ref|><|det|>[[115, 147, 875, 728]]<|/det|>
+As TGM3 lacks a signal sequence, we determined if TGM3 could nonetheless be secreted into the mucus. To answer this, gel electrophoresis and western blot analyses for TGM2 and 3 in colonic mucus were performed. Recombinantly expressed TGM2 and cleaved TGM3 were also loaded as positive controls either non- activated or activated by \(\mathrm{Ca^{2 + }}\) - preincubation (Fig. 1e). The majority of TGM3 was represented by a band migrating around \(75\mathrm{kDa}\) and a weaker signal migrating at approximately \(50\mathrm{kDa}\) in both WT and \(Tgm2^{- / - }\) animals. These two bands represent the zymogenic and active form of the enzyme, respectively. Furthermore, several diffuse, but weak, TGM3- signals migrating between 150 and \(250\mathrm{kDa}\) were detected in the WT and \(Tgm2^{- / - }\) strains suggesting the self- multimerization of the enzyme and/or its incorporation into substrate proteins. As similar signals were detected in the activated positive control for TGM3, it is likely that self- multimerization occurs in mucus. In contrast, TGM 2 was not detected in the mucus samples of any mouse strain. Specificity of the used antibodies for the respective isoform was determined upon western blot analyses, the anti- TGM3 antibody showed a cross- reactivity \(< 8\%\) on TGM2 and similarly vice versa (Suppl. Fig. S1b) Together the results show that TGM3 is the predominant transglutaminase in the colonic epithelium and the only isozyme detected in the mucus. Furthermore, its expression in goblet cells suggests that its presence in mucus arises at least partly from active secretion and not only from shedded cells.  
+
+<|ref|>text<|/ref|><|det|>[[118, 772, 463, 789]]<|/det|>
+TGM3 activity is present in colonic mucus  
+
+<|ref|>text<|/ref|><|det|>[[115, 804, 850, 888]]<|/det|>
+Next, we asked if TGM3 is enzymatically active in the colonic mucus and could thereby contribute to its stability by the formation of additional cross- links. For that purpose, a qualitative assay using the incorporation of biotinylated isoform- specific substrate peptides
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 78, 875, 761]]<|/det|>
+T26 (TGM2) and E51 (TGM3) in mucus was performed. The mucus was incubated with \(\mathrm{Ca^{2 + }}\) and the respective peptide probe followed by gel electrophoresis and western blot using streptavidin detection (Fig. 2a). Specific incorporation of the two peptides was observed in WT and \(Tgm2^{- / - }\) mucus, but not in mucus from \(Tgm3^{- / - }\) animals. Non- specific signals were observed in all samples, including control reactions where transglutaminase activity was inhibited by iodoacetamide (IAA). These bands are likely due to endogenously biotinylated proteins as for example pyruvate-carboxylase. Thus, the detected cross- linking activity in the mucus arises from TGM3- mediated catalysis. To analyze if endogenous mucus contains sufficient \(\mathrm{Ca^{2 + }}\) - ions for the activation of TGM3, the experiment was repeated without calcium addition. Similar results as with exogenous \(\mathrm{Ca^{2 + }}\) - addition were obtained, indicating the presence of intrinsic extracellular transglutaminase activity in colonic mucus (Fig. 2b). These results suggest that endogenous acyl- donor protein substrates are present in colonic mucus. However, the formation of a transglutaminase- catalyzed cross- linked mucus gel- network also requires the presence of acyl- acceptor proteins. Therefore, the \(\mathrm{Ca^{2 + }}\) - free experimental set up was modified by replacing the glutamine- donor with the primary amine 5- Biotinyl- pentylamine (5- BP) as acyl- donor. Similar to the results from the acyl- acceptor experiments, specific signals were detected when the acyl- donor compound was added to mucus of WT and \(Tgm2^{- / - }\) animals, but not in the \(Tgm3^{- / - }\) mucus or when IAA was added (Fig. 2c). Together, the results show that colonic mucus contains intrinsically, active TGM3 as well as both acyl- acceptor and - donor molecules allowing transamidating reactions to take place.  
+
+<|ref|>text<|/ref|><|det|>[[115, 772, 870, 889]]<|/det|>
+To quantify the intrinsic transamidating activity in colonic mucus, a colorimetric assay for the incorporation of a TGM- promiscuous peptide (A25) and the two isozyme- selective peptide substrates (peptides \(\mathrm{T26^{21}}\) and \(\mathrm{E51^{22}}\) ) into casein was performed (Fig. 2d). A natural cross- linking activity in WT mucus of \(\approx 8\pm 2\) U/mg for the promiscuous substrate was
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 81, 880, 436]]<|/det|>
+determined. Substitution with the TGM3- specific substrate E51 led to a 1.5- fold increase \((\approx 12\pm 4\mathrm{U / mg})\) of the transamidating activity, whereas a residual activity of \(0.8\pm 0.3\mathrm{U / mg}\) was observed for the TGM2- specific substrate. However, no measurable activity could be obtained in the \(Tgm3^{- / - }\) mucus as the detected values were below the limit of detection for our assay (Fig. 2d, Suppl. Fig. S2). Blocking of the TGM- reaction with Z- DON led to an almost complete \((88\%)\) inhibition for the promiscuous peptide A25. In line with our other results (Fig. 1, Fig. 2 a- c), the natural cross- linking activity was related to TGM3 as the use of the TGM2- specific substrate T26 led to less than \(10\%\) transglutaminase activity compared to the TGM3- specific substrate in WT animals and was also below the limit of quantification of this assay. These experiments further demonstrated substantial intrinsic transamidating activity in colonic mucus of WT animals, but not in \(Tgm3^{- / - }\) , as addition of extra \(\mathrm{Ca^{2 + }}\) was not required.  
+
+<|ref|>text<|/ref|><|det|>[[113, 444, 870, 890]]<|/det|>
+The intrinsic mucus transamidating activity of TGM3 was further studied using an ex vivo approach where the distal colon from WT and \(Tgm3^{- / - }\) animals were mounted in a perfusion chamber and the fluorescently labelled glutamine- donor probe E51 was added and its incorporation monitored (Fig. 2e). Fig. 2f- h show the confocal microscopic analyzes of E51 incorporation in the respective tissue/mucus specimen in the \(\mathrm{x / z}\) plane (top panels) and snap shots of probe incorporation of the \(\mathrm{x / y}\) plane inside the mucus ( bottom panels). A homogeneous punctuated pattern of E51 fluorescence was observed throughout the whole mucus layers of WT animals (Fig. 2f and Suppl. Movie M1). However, when \(Tgm3^{- / - }\) mice were analyzed in the same way, the incorporation was dramatically reduced and limited to shedding epithelial cells (Fig. 2g). A similar lack of incorporation in WT animals was observed in the presence of the transglutaminase inhibitor Z- DON (Fig. 2h). These results demonstrate extracellular TGM3 activity ex vivo. Together these results show that the colonic mucus contains natural acyl- donor and acyl- acceptor molecules together with intrinsic TGM3- mediated transamidating activity.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 870, 500]]<|/det|>
+Loss of TGM3 alters biochemical properties of mucus/MUC2The MUC2 monomer is a large glycoprotein with a mass of around 2.5 MDa (Fig. 3a). It is the most abundant constituent in colonic mucus and is thus a potential target for TGM3- mediated cross- linking, something that could influence its biochemical properties. Colonic mucus from WT, \(Tgm2^{- / - }\) and, \(Tgm3^{- / - }\) mice was isolated and disulfide bonds reduced followed by separation via composite agarose- PAGE (AgPAGE) and detected by in- gel immunostaining using anti- MUC2C3 antibody (Fig. 3b). WT and \(Tgm2^{- / - }\) showed two identical diffuse fast- moving bands assumed to be MUC2 monomeric bands and several additional slow- moving and heavily stained bands for higher oligomers. This was in contrast to the \(Tgm3^{- / - }\) - mucus, where MUC2 showed a faster migrating diffuse band and two to WT different bands migrating similar to the WT monomer.- These differences in the electrophoretic migration pattern suggest that \(Tgm3^{- / - }\) MUC2 is qualitatively different to that of WT and \(Tgm2^{- / - }\) and argues for TGM3- mediated isopeptide bond modification of MUC2.  
+
+<|ref|>text<|/ref|><|det|>[[115, 509, 870, 890]]<|/det|>
+As isopeptide bonds can prevent proteolytic cleavage and secreted mucus is normally exposed to numerous endogenous and bacterial proteolytic enzymes, we hypothesized that the different size of MUC2 formed in \(Tgm3^{- / - }\) mice was a result of protease- catalyzed degradation in vivo. To test this hypothesis, colonic mucus of WT and \(Tgm3^{- / - }\) mice was first isolated and solubilized by reduction with dithiothreitol. The resulting samples were treated with the serine protease LysC, followed by the separation of the reaction products via composite AgPAGE and Alcian Blue staining of the heavily glycosylated and protease- resistant MUC2 domains (PTS sequence). All three strains showed three identical intensely stained bands after LysC treatment (Fig. 3c). Interestingly, this band pattern was also observed in the non- treated \(Tgm3^{- / - }\) mice, but not in the WT or \(Tgm2^{- / - }\) animals. This could suggest that the faster migrating MUC2 bands in the non- treated \(Tgm3^{- / - }\) animals represent products that have been already degraded in vivo. To confirm this, the fastest MUC2
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 878, 496]]<|/det|>
+migrating bands from the non- treated WT (WT- M) and \(Tgm3^{- / - }\) ( \(Tgm3^{- / - }\) - M) samples were excised from the gels (Fig. 3b) followed by mass spectrometric analyses of their tryptic/AspN peptides. The peptide coverage of the MUC2 sequence of three biological replicates is summarized in a heat- map shown in Fig. 3d. The WT monomers showed peptides from all domains except the PTS as expected. Interestingly, the \(Tgm3^{- / - }\) MUC2 molecule showed almost exclusively peptides from the central CysD2 domain (Fig. 3a and d). The vWD4 domain was weakly covered in both animals explaining the anti- MUC2C3 staining. As the fastest migrating bands in the \(Tgm3^{- / - }\) mucus were stained by Alcian Blue and have masses larger than \(460\mathrm{kDa}\) , these bands must also include the two mucin domains surrounding CysD2. These PTS1 and PTS2 sequences are highly glycosylated, resistant to proteolytic enzymes, and not identifiable by mass spectrometry (Fig. 3a). Thus, the MUC2 mucin in the \(Tgm3^{- / - }\) mice is suggested to be already degraded in vivo due to it being more susceptible to degradation in the colon lumen.  
+
+<|ref|>text<|/ref|><|det|>[[115, 510, 876, 792]]<|/det|>
+The most likely explanation for the more degraded MUC2 in \(Tgm3^{- / - }\) mice is the loss of protective transglutaminase- catalyzed isopeptide bonds. To search for such bonds, we mined the mass spectrometry data sets for the presence or absence of such cross- links. An example is shown in the mass spectrum of a dipeptide for an intramolecular cross- link connecting Gln 2503 with Lys 2508 (Fig. 3e). This intramolecular cross- linked peptide was only detected in MUC2 from WT, but not in \(Tgm3^{- / - }\) animals. This isopeptide bridge is located between the vWC2 domain and the cysteine- knot (CK) domain (Fig. 3a). There are likely several additional cross- links and this isopeptide- bridged peptide is only one example, but its absence in \(Tgm3^{- / - }\) MUC2 supports this interpretation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 805, 868, 889]]<|/det|>
+As non- reduced secreted MUC2 polymers in the intestine are known to be insoluble in guanidinium chloride due to isopeptide bonds formed intracellularly \(^{23}\) , we asked if TGM3- mediated isopeptide cross- links contributed to this property. To address this question,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 864, 233]]<|/det|>
+insoluble mucus from WT and \(Tgm3^{- / - }\) mice was precipitated by centrifugation and the turbidity of soluble material in the supernatant recorded (Fig. 3f). The turbidity of the samples from \(Tgm3^{- / - }\) animals was increased by approximately \(30\%\) when compared to WT and \(Tgm2^{- / - }\) strains. This result further supports the idea that disintegration of the MUC2 mucin network was more prominent in the mice lacking the TGM3 enzyme.  
+
+<|ref|>text<|/ref|><|det|>[[115, 246, 877, 530]]<|/det|>
+Mucins have been shown to attach to hydrophobic surfaces \(^{24}\) . We hypothesized that natural isopeptide cross- links might contribute to this biophysical property and thus analyzed the hydrophobic character of colonic mucus by using a thermal fluorescent shift assay. Colonic mucus mixed with the hydrophobic dye SyproOrange was subjected to a linear temperature gradient and the fluorescence measured (Fig. 3g). At higher temperatures (>50°C) the \(Tgm3^{- / - }\) mucus showed an increased fluorescence in relation to WT, indicating an increased exposure of hydrophobic protein parts. Preincubation of \(Tgm3^{- / - }\) mucus with recombinant TGM3 partly normalized the mucus. It can be suggested that TGM3- mediated isopeptide bonds in WT mucus prevented the unfolding of its constituents.  
+
+<|ref|>text<|/ref|><|det|>[[115, 542, 870, 890]]<|/det|>
+Mucus processing and tissue secretory responses were assessed using ex vivo mucus measurement assays. Using this approach, we detected no differences in baseline mucus growth rate or carbachol- induced secretory responses between WT and \(Tgm3^{- / - }\) tissues (Suppl. Fig. S3a). A similar approach can be used to measure mucus barrier function by applying bacteria- sized (1μm diameter) beads to the mucus surface and determining the extent of bead penetration into the mucus via confocal microscopy. However, again no difference between WT and \(Tgm3^{- / - }\) tissues was detected using this approach (Suppl. Fig. S3b), which was surprising, as we had observed a more degraded MUC2 mucin in the \(Tgm3^{- / - }\) animals. Nonetheless, we hypothesized that lack of TGM3 would affect mucus barrier stability and thus treated colonic tissue from WT and \(Tgm3^{- / - }\) animals with pronase. In WT animals and before addition of pronase to \(Tgm3^{- / - }\) tissue, the fluorescent beads remained on
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 872, 233]]<|/det|>
+top of the mucus layer (Fig. 3h and Suppl. Movies M2 and M3). However, after pronase- treatment of \(Tgm3^{- / - }\) explants, a progressive decrease in mucus thickness was observed and the beads were more easily washed away and/or penetrated down to the epithelial surface indicating that \(Tgm3^{- / - }\) mucus was less protected against proteolytic attack (Fig. 3i and Suppl. Movie M3).  
+
+<|ref|>text<|/ref|><|det|>[[115, 278, 880, 895]]<|/det|>
+Tgm3- mice are more susceptible to early DSS- induced damageThe altered biochemical properties of mucus and its higher susceptibility to proteolytic degradation in the absence of TGM3 activity suggested that \(Tgm3^{- / - }\) mice could be more susceptible to dextran sodium sulfate (DSS) induced colitis. To test this, age- matched cohoused \(Tgm3^{- / - }\) and WT animals were challenged with DSS. The body weight of WT mice increased during the first four days whereas the \(Tgm3^{- / - }\) animals started to lose weight from day three and showed on trend decreased body weights compared to WT mice until day 6 (Fig. 4a). This was reflected by an earlier detection of occult blood in the feces of \(Tgm3^{- / - }\) mice one day after the start of the experiment (Fig. 4b). Consequently, the \(Tgm3^{- / - }\) animals showed a significant raised disease activity index score (DAI) from day two to day five after the start of the DSS- treatment (Fig. 4c). Higher DAI was maintained in the \(Tgm3^{- / - }\) compared to WT animals until day 6, when the colitis became also established in the WT animals. Finally, 50% of the \(Tgm3^{- / - }\) animals had to be sacrificed at day 7, compared to 10% of WT mice, due to suffering and loss of weight following the ethical permit (Fig. 4d). Furthermore, the colon length of \(Tgm3^{- / - }\) mice was reduced to 88% of the WT length after 7 days of DSS treatment (Fig. 4e and f). Histopathological analysis of the colonic tissue after eight days of DSS treatment revealed the loss of crypts and an extensive infiltration of immune cells in both strains. These effects were more pronounced in the distal colon (Fig. 4g). However, histological examination of Hematoxylin/Eosin- stained tissue by a blinded pathologist did not
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 82, 875, 400]]<|/det|>
+detect significant differences between the two animal strains at the end of DSS- treatment (not shown). DSS has previously been shown to disrupt the mucus layer properties \(^{25}\) and mice lacking the MUC2 mucin are very susceptible already at day one of DSS treatment \(^{4}\) . The early on- set of DSS effects in the \(Tgm3^{- / - }\) supports the conclusion that the colonic mucus is defect in these animals. When colonic tissue was analyzed by immunohistochemistry for TGM2, this isozyme that was absent in non- treated WT and \(Tgm3^{- / - }\) as shown in Fig. 1a, was now detected in both the WT and \(Tgm3^{- / - }\) animals after 7 days of DSS- treatment (Fig. 4h). Taken together \(Tgm3^{- / - }\) animals were significantly more susceptible towards the colitis- inducing effects of DSS as a faster disease onset was observed resulting in a decreased probability of survival.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 446, 212, 462]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[115, 475, 860, 890]]<|/det|>
+We have previously shown that the reduction- insensitive MUC2 oligomers formed in a cell line producing MUC2 were cross- linked by isopeptide bonds as catalyzed by a yet unidentified transglutaminase \(^{12}\) . However, this colorectal cell line does not secrete MUC2 and could not be used to learn if and how extracellular cross- linking could contribute to mucus homeostasis and colon barrier function. By using WT and knock- out mouse strains, we uncovered intrinsic transglutaminase activity in secreted colonic mucus mediated by TGM3. The observations provide evidence for the protective effect by natural cross- links. That TGM3 is the dominant transglutaminase of the colon is in accordance with a previous mucus proteome study \(^{2}\) . mRNAseq and MS studies detected minor amounts of TGM2, but based on the label- free mass spectrometric quantification TGM2 was \(< 1\%\) of that of TGM3 and could represent contaminating material from the ileum. In support of this, TGM2 was not detected by immunohistochemistry or gel electrophoresis/western blot. Previous work from Jeong and co- workers has claimed TGM2 as the major transglutaminase of the colon \(^{26}\) .
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 875, 167]]<|/det|>
+However, these authors used only immunohistochemistry to demonstrate the presence of TGM2 and no antibody staining against TGM3 was tested. Likely cross reactivity of the used antibody can explain this observation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 181, 878, 365]]<|/det|>
+The strong TGM3 signals observed by gel electrophoresis/western blot analyses of mucus represented the zymogenic form of TGM3. In addition to this, several TGM3 bands with higher molecular masses were detected in the range between 150 and \(250\mathrm{kDa}\) . Together with control reactions performed with recombinantly activated TGM3, these signals strongly suggest that the enzyme can self- multimerize and/or incorporate itself into other molecules as previously observed for TGM2<sup>27</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[114, 377, 878, 892]]<|/det|>
+Recent reports have shown that TGM2 is extracellularly inactive and can be activated after injury or stress<sup>17, 28</sup>. Here, we were able to demonstrate intrinsic, extracellular transglutaminase activity in both the WT and \(Tgm2^{- / - }\) mouse strains, but its absence in \(Tgm3^{- / - }\) animals. In addition, the obtained information showed the presence of natural acyl- donor and - acceptor molecules in colonic mucus thereby implying the possibility of in vivo isopeptide- based cross- linking of different mucus components. Furthermore, transglutaminase activity could be detected without calcium- addition, showing that extracellular transglutaminase activity is intrinsic to the large intestinal mucus. Since a \(>90\%\) reduction in TGM activity was observed when the TGM2- selective substrate T26 was used in a quantitative assay and no activity was found in the \(Tgm3^{- / - }\) animals, we can conclude that the transamidation activity of mucus is almost exclusively dependent on TGM3 in colon. This is in line with the shown absence of TGM2 protein. In an ex vivo assay, a punctuated incorporation of the specific TGM3 peptide substrate E51 in colonic mucus was observed confirming our in vitro observations. However, it was not possible to perform the ex vivo mucus incorporation approach under \(\mathrm{Ca^{2 + }}\) - free conditions since the normal cellular signaling of the tissue depends on an extracellular calcium pool. This is reflected by measurements of the luminal calcium
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 80, 877, 696]]<|/det|>
+concentrations in the gut which vary between 5 and \(20 \mathrm{mM}\) depending on the feed state \(^{29,30}\) whereas the concentration in the used buffer was \(1.3 \mathrm{mM}\) representing the physiological luminal calcium concentration \(^{31}\) . Given that only \(20\%\) of the daily calcium intake is resorbed, mainly in the small intestine, the luminal \(\mathrm{Ca}^{2 + }\) - concentration in the colon should be sufficient to occupy the second and third \(\mathrm{Ca}^{2 + }\) - binding site of TGM3 necessary for its activation \(^{32,33}\) . Furthermore, TGM3 is expressed and synthesized in goblet cells, a secretory cell lineage whose secretory granules contain high calcium concentrations for the packing of the MUC2 mucin. Secretory granules can contain calcium concentrations of up to \(40 \mathrm{mM}\) \(^{34}\) . The pH in goblet cell granule is acidic and the \(\mathrm{Ca}^{2 + }\) - ions are bound to MUC2 and the other stored molecules. After secretion, the pH raises, and free \(\mathrm{Ca}^{2 + }\) - ions will become available. For the activation of TGM3, the \(\mathrm{Ca}^{2 + }\) - binding sites must be occupied and the zymogenic form of TGM3 needs to be cleaved in the loop harboring amino acids 462- 469. For this to take place, Cathepsin L or S have been suggested as activating proteases \(^{35}\) . Interestingly, Cathepsin S is a core mucus component and Cathepsin L is also expressed in colonic epithelial and goblet cells \(^{7}\) and Suppl. Fig. S4. This suggests that TGM3 can become fully activated in the colonic mucus and lumen, in line with the endogenously observed TGM3 activity in colon mucus. Overall, the availability of calcium and Cathepsin S and L together with an alkaline pH in the large intestinal lumen provide favorable conditions for TGM3 to catalyze transamidating reactions in colonic mucus.  
+
+<|ref|>text<|/ref|><|det|>[[115, 707, 880, 889]]<|/det|>
+The observed intrinsic, extracellular transglutaminase activity led us speculate about the putative functional impact of the natural cross- links in colonic mucus. The comparison of mucus from WT and \(Tgm3^{- / - }\) mice provided direct experimental evidence that the loss of TGM3 led to important biochemical alterations of the dominant mucus skeleton protein MUC2. We observed an extensive degradation of the polypeptide as the N- terminal part with the first three vWD and the first CysD domain was lost comprising approximately 1,300
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 80, 880, 696]]<|/det|>
+amino acids as well as most of the C- terminus. Thus, leaving a central part of the MUC2 mucin consisting of the two highly glycosylated PTS sequences connected via the second CysD domain behind. The resistance of the mucin domains to proteolytic cleavage is due to their dense decoration by \(O\) - linked glycans resulting in steric hindrance to protease degradation. Another feature of MUC2 that was affected in \(Tgm3^{- / - }\) mice is its solubility. MUC2 polymers become insoluble during their transport through the later stages of the secretory pathway \(^{23}\) . In our turbidity assay, an effect on the MUC2 gel network was indicated by an increased optical density in the mucus supernatant from \(Tgm3^{- / - }\) mice. This suggests more soluble MUC2 in this mouse strain. In addition to its solubility, the hydrophobicity of MUC2 was also altered in the absence of TGM3. The assay showed an increased exposure of hydrophobic patches in partly purified MUC2 from \(Tgm3^{- / - }\) animals compared to WT animals upon heat- induced denaturation. This may reflect that the \(\mathrm{N}^6\) (- \(\gamma\) - glutamyl)- lysine bonds in WT- MUC2 stabilize the protein. This phenomenon was partly reverted by preincubating the \(Tgm3^{- / - }\) mucus with recombinant TGM3, supporting the impact of isopeptide bonds on this biophysical parameter. Our observations are consistent with previous studies showing that isopeptide bonds can stabilize bacterial pili proteins in this kind of assay \(^{36,37}\) . However, the insoluble nature of the MUC2 mucin makes it impossible to purify the MUC2 polypeptide in its native conformation, making recording of a specific melting temperature impossible.  
+
+<|ref|>text<|/ref|><|det|>[[115, 707, 881, 890]]<|/det|>
+In the \(Tgm3^{- / - }\) mice, the shortened and more degraded MUC2 still seems to be sufficient to provide enough protection for the colonic epithelium. This mouse strain behaves normally and shows no obvious signs of colon inflammation under normal conditions. It is likely that the highly \(O\) - glycosylated mucin domains of MUC2 are sufficient to trap microorganisms and prevent bacteria from reaching the epithelial cells. However, challenging the system by the addition of a mixture of serine proteases deciphered an altered phenotype in \(Tgm3^{- / - }\) animals
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 78, 875, 700]]<|/det|>
+in an ex vivo bead penetration assay. In this approach, the mucus layer seemed to be more disintegrated and less organized allowing the bacteria- mimicking beads to reach the epithelium suggesting a compromised barrier function. We also obtained direct experimental support of this hypothesis by administering DSS to WT and \(Tgm3^{- / - }\) animals. DSS quickly disintegrates the inner colon mucus layer allowing bacteria to reach the epithelium and trigger inflammation typically observed after five days \(^{25}\) . DSS treatment showed that the \(Tgm3^{- / - }\) mice were more susceptible and showed defects already after only two days of treatment. The early onset suggest direct effects on mucus that are likely explained by the decreased inter- molecular cross- links in the mucus of \(Tgm3^{- / - }\) mice. Less cross- links will, as shown here, make the mucus more susceptible to proteolytic degradation and detachment leading to a faster mucus removal by intestinal peristalsis. Interestingly, two transcriptomic studies using either the 2,4,6- trinitrobenzene sulfonic acid or the adoptive T- cell transfer colitis model detected TGM3 downregulation after the establishment of the disease, thereby suggesting an impact of this enzyme for a healthy gut \(^{38, 39}\) . In contrast to non- treated mice, both the WT and \(Tgm3^{- / - }\) mice synthesized the TGM2 enzyme in their colonic mucosa after DSS treatment. This might reflect a role for TGM2 in wound healing as suggested previously \(^{18, 26}\) . As mucus can be regarded as our ‘inner skin’ it is not surprising that a weakened mucus barrier in the \(Tgm3^{- / - }\) mice could be regarded in analogy with the TGM3 function in the skin where earlier observations have shown that these animals have an impaired skin barrier \(^{40}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 707, 873, 890]]<|/det|>
+This study identifies TGM3 as an important natural cross- linking enzyme acting on the expanded secreted mucus and by this contributing to the stabilization of the colonic mucus gel network. The MUC2 mucin and other mucus components are secreted into the harsh luminal environment where proteases from the host, the commensal bacteria, and eventually from pathogens reside. The TGM3- catalyzed formation of isopeptide bond cross- links strengthens the mucus barrier and thereby increase the mucus protection of the colonic
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 83, 873, 300]]<|/det|>
+epithelium. However, further studies are required to more precisely understand the molecular details for the role of transglutaminases for the mucus structure. For example, as there exists an inverse gradient of TGM2 and 3 abundance from the small to the large intestine it would be interesting to determine the activity of TGM2 and decipher its role for small intestinal mucus. Our observations increase our understanding of the molecular mechanisms that contribute to the architecture of the colonic mucus layers and suggest potential treatment options for the human disease UC.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 347, 197, 363]]<|/det|>
+## Methods  
+
+<|ref|>text<|/ref|><|det|>[[118, 380, 190, 396]]<|/det|>
+Animals  
+
+<|ref|>text<|/ref|><|det|>[[115, 410, 874, 595]]<|/det|>
+C57/BL6N mice were from Taconic. \(Tgm2^{- / - }\) mice<sup>41</sup> were provided from Oslo University Hospital (Norway). \(Tgm3^{- / - }\) mice<sup>42</sup> were obtained from the University of Rome (Tor Vergata, Italy). All animal experiments were conducted according to the Swedish legislation (Jordbruksverket; Ethical permits no.: 2285/19 and 2292/19). Mice were maintained at \(22^{\circ}\mathrm{C}\) with light/dark cycles of 12 hours each. Animals received a standard rodent diet and water was supplied ad libitum.  
+
+<|ref|>text<|/ref|><|det|>[[118, 644, 377, 660]]<|/det|>
+Antibodies, enzymes, chemicals  
+
+<|ref|>text<|/ref|><|det|>[[115, 674, 878, 888]]<|/det|>
+If not otherwise specified chemicals were bought from Sigma. For the detection of TGM2 the monoclonal CUB7402 antibody (Thermo Fisher Scientific) was used for both immunohistochemistry (IHC) and Western Blot. TGM3 detection was performed using the polyclonal NBP1- 57678 antibody (Novus Biologicals) for both applications. Cross- reactivity of the two antibodies was analysed by western blot against recombinant TGM2 and 3 (Suppl. Fig. S1b). For IHC detection of TGM2 a goat- \(\alpha\) - mouse- IgG1 antibody coupled to AlexaFluor647 (Invitrogen) and a goat- \(\alpha\) - rabbit- IgG antibody coupled to AlexaFluor647
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 878, 330]]<|/det|>
+(Invitrogen) for TGM3 detection was used. For Western Blot detection of TGM2 a goat- \(\alpha\) - mouse- IgG1 antibody coupled to the IRdye 680LT (LI- COR and a goat- \(\alpha\) - rabbit- IgG antibody coupled to AlexaFluor 790 (Invitrogen) was used. Trypsin and AspN were from Promega. LysC was from WAKO (Japan). Recombinant TGM2 (T022) and TGM3 (T013) as well as the biotinylated glutamine donor substrates A25 (B001); T26 (B008); E51 (B009) and the biotinylated amine donor compound pentylamine (B002) were bought from Zedira (Germany). The FITC- labelled E51 probe was bought from CovalAb (France). Pronase was from Merck (Germany). The UEA1 lectin was from BioNordika.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 379, 305, 396]]<|/det|>
+## Immunohistochemistry  
+
+<|ref|>text<|/ref|><|det|>[[115, 409, 877, 790]]<|/det|>
+Paraffin- embedded tissue sections were deparaffinized with xylene and rehydrated in ethanol solutions ranging from \(100\%\) to \(30\%\) . Antigen- retrieval was performed by boiling the sections in \(10 \mathrm{mM}\) citrate buffer pH 6.0. The sections were blocked for one hour with \(5\%\) fetal bovine serum (FBS) in PBS. Afterwards the antibodies for TGM2 and 3 were added (1:200 diluted in PBS containing \(5\%\) FBS) and the sections incubated overnight at \(4^{\circ}\mathrm{C}\) in a humid chamber followed by three washing steps in PBS. Secondary antibodies coupled to the AlexaFluor647 dye ( \(\alpha\) - mouse- IgG for TGM2 \(\alpha\) - rabbit- IgG for TGM3, 1:1,000 diluted in PBS containing \(5\%\) FBS) were added together with the UEA1 lectin ( \(10 \mu \mathrm{g / ml}\) ) conjugated to the rhodamine dye for one hour. After three washing steps in PBS the nuclei were stained with the Sytox green stain for five minutes. After one additional washing step the sections were mounted using ProLong Gold- Antifade mountant (Thermo Fisher Scientific) and visualized by confocal microscopy (Zeiss Examiner 2.1; LSM 700).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 111, 879, 400]]<|/det|>
+Mucus supernatants (30 \(\mu \mathrm{g}\) ) were incubated either with \(10~\mu \mathrm{M}\) of the TGM2- or TGM3- specific glutamine- donor peptides \(\mathrm{T}26^{21}\) or \(\mathrm{E}51^{22}\) or the amine- donor compound 5- Biotinyl- pentalamine for one hour at \(37^{\circ}\mathrm{C}\) . Control reactions were performed in the presence of 25 mM IAA and the respective compounds. The reactions were stopped by the addition of SDS- loading buffer and heating to \(95^{\circ}\mathrm{C}\) for five minutes. Reaction products were separated by SDS- PAGE on \(4 - 15\%\) gradient gels followed by semidry transfer to PVDF membranes. After blocking with \(3\%\) BSA in TBS buffer the membrane was incubated with streptavidin coupled to AlexaFluor 680 (1:20,000, Invitrogen) and the incorporation of substrates revealed on an Odyssey Li- COR Clx workstation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 410, 875, 891]]<|/det|>
+Quantitative determination of TGM activity was performed according to the method described by Trigwell and coworkers \(^{43}\) . Briefly, maxisorb 96- well plates (Thermo) were coated with \(250~\mu \mathrm{l}\) of a \(0.1\%\) casein solution in \(50~\mathrm{mM}\) sodium carbonate pH 9.8 for 12 hours. After emptying and washing \(250~\mu \mathrm{l}\) blocking solution ( \(0.1\%\) BSA in \(50~\mathrm{mM}\) sodium carbonate pH 9.8) was added and incubated for one hour at \(37^{\circ}\mathrm{C}\) . After washing, \(150~\mu \mathrm{l}\) reaction buffer ( \(100~\mathrm{mM}\) TrisHCl pH 8.5, \(6.7~\mathrm{mM}\) CaCl2, \(13.3~\mathrm{mM}\) DTT containing either 10 \(\mu \mathrm{M}\) biotinylated TGM- substrate peptide E51; T26, respectively or 5 \(\mu \mathrm{M}\) biotinylated TGM- substrate peptide A25) for the respective TGM standards was added to the wells. For the analysis of mucus samples, DTT and calcium were omitted in the reaction buffer. Measurements were carried out in triplicate per biological replicate. The reactions were started by the addition of either \(50~\mu \mathrm{l}\) TGM standards (0; 25; 50; 75; 100; 125 \(\mathrm{mU / well}\) ) or mucus samples and incubated for one hour at \(37^{\circ}\mathrm{C}\) on a rotational shaker set to \(100~\mathrm{rpm}\) . Afterwards, the reactions were stopped by emptying the wells and washing. The incorporation of the substrates in the casein matrix was probed by the addition of \(200~\mu \mathrm{l}\) Extravidin solution (Extravidin- peroxidase (1:10,000 in \(100~\mathrm{mM}\) TrisHCl pH 8.5 containing
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 875, 266]]<|/det|>
+1% BSA) for one hour and gentle shaking. Biotin- Extravidin binding was visualized by adding \(200\mu \mathrm{l}\) TMB developing solution \((3,3^{\prime},5,5^{\prime}\) - Tetramethylbenzidine, Sigma) and the reaction stopped by adding \(50\mu \mathrm{l}5\mathrm{M}\mathrm{H}_2\mathrm{SO}_4\) . The absorbance of reaction and standard wells was recorded at \(450\mathrm{nm}\) on a Victor2 Wallac work station (Perkin Elmer). The activities of the samples were subsequently normalized against the protein content of the sample using the BCA method.  
+
+<|ref|>text<|/ref|><|det|>[[115, 313, 483, 330]]<|/det|>
+Ex vivo analysis of transglutaminase activity  
+
+<|ref|>text<|/ref|><|det|>[[115, 345, 864, 757]]<|/det|>
+Ex vivo analysis of transglutaminase activityMice were anaesthetized using isoflourane and sacrificed by cervical dislocation. The colon was collected by dissection and flushed for the removal of intestinal content using Krebs buffer as previously described<sup>44</sup>. After removal of the muscle layer by microdissection the tissue was mounted in an in- house built horizontal chamber allowing basolateral perfusion with Krebs- Glucose buffer and apical Krebs- mannitol buffer (Fig. 2e). Two \(\mu \mathrm{M}\) FITC- labelled E51- probe in Krebs- mannitol buffer was added and the tissue incubated for 30 minutes at \(37^{\circ}\mathrm{C}\) . Afterwards, non- incorporated probe molecules were washed away with Krebs- mannitol buffer followed by analysis of incorporation of the TGM3- substrate peptide on an upright LSM700 confocal microscope (Carl Zeiss, Germany) equipped with a \(20\mathrm{x}\) immersion lens (Pan- Apochromat \(20\mathrm{x} / 1.0\) DIC \(75\mathrm{mm}\) ; Carl Zeiss, Germany). Images were acquired using Zen Black software (Carl Zeiss) and z- stacks were exported to TIFF format using the Imaris software. Inhibition of transglutaminase activity in WT mice was achieved by adding \(5\mu \mathrm{M}\) Z- DON (Zedira) together with the TGM3 substrate.  
+
+<|ref|>text<|/ref|><|det|>[[117, 805, 363, 822]]<|/det|>
+Ex vivo mucus integrity assay  
+
+<|ref|>text<|/ref|><|det|>[[115, 837, 816, 855]]<|/det|>
+Tissue was collected as described for the ex vivo analysis of transglutaminase activity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 870, 825, 888]]<|/det|>
+Following mounting in the perfusion chamber, tissue was stained with Syto 9 (1:500 in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 82, 866, 432]]<|/det|>
+Kreb's- mannitol buffer; Thermo Fisher) and the mucus layer was visualized by the addition of \(1\mu \mathrm{m}\) fluorescent beads (Thermo Fisher). \(20\mathrm{mg / ml}\) of pronase was added to the apical Krebs- mannitol buffer and the integrity of the mucus layer was monitored on an upright LSM900 confocal microscope (Carl Zeiss) using a water Pan- Apochromat \(20\mathrm{x} / 1.0\mathrm{DIC}75\) mm lens (Carl Zeiss; Germany). Tissue explants were maintained at \(37^{\circ}\mathrm{C}\) throughout the experiments. Briefly, z- stacks were acquired every 5 minutes (total time \(1\mathrm{hr}\) .) using Zen Blue software (version 3.1; Carl Zeiss, Germany). In order to monitor mucus integrity beads and tissue surfaces were mapped to isosurfaces using Imaris software as described previously \(^{45}\) , data regarding the position of the fluorescent beads in relation to the tissue surface over time was then extracted and analyzed to generate normalized positional data over time (Prism version 9.1.0, Graphpad).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 477, 321, 495]]<|/det|>
+## Colitis induction by DSS  
+
+<|ref|>text<|/ref|><|det|>[[113, 508, 880, 859]]<|/det|>
+Age- and sex- matched WT C57/BL6 and \(Tgm3^{- / - }\) mice were cohoused for 4 to 5 weeks. Colitis was induced by adding \(3\%\) (w/v) dextran sodium sulfate (DSS) to the drinking water. Mice could drink ad libitum. The mice were sacrificed after eight days or if their body weight dropped by \(10\%\) from the initial weight. The probability of survival was defined when mice died or if they showed a body weight loss \(>10\%\) . The colon was dissected and its length measured from cecum to anus and subsequently normalized against the initial body weight of the respective animal. Afterwards, the colon was flushed with PBS for the removal of fecal content. The colons were fixed as Swiss rolls in \(4\%\) paraformaldehyde and stained for hematoxylin/eosin and Alcian Blue- PAS. The disease activity index (DAI) was calculated as the sum of the combined scores for stool consistency, hematochezia and weight loss according to the methods of Friedman and co- workers \(^{46}\) . The detection of occult blood was
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 825, 135]]<|/det|>
+performed using a Hemoccult kit (Beckman Coulter) according to the manufacturer’s instructions. Two litters of each mouse strain with five animals per litter were analyzed.  
+
+<|ref|>text<|/ref|><|det|>[[117, 183, 336, 199]]<|/det|>
+Composite agarose- PAGE  
+
+<|ref|>text<|/ref|><|det|>[[115, 213, 876, 626]]<|/det|>
+The separation of MUC2 was performed according to the protocol of Schulz and coworkers<sup>47</sup>. Briefly, mucus was scraped from mouse colon and emulsified in TBS. Mucus/Muc2 was precipitated by centrifugation at 16,000 x g and 4°C for 30 minutes. The mucus was solubilized by the addition of reducing gel- loading buffer (62.5 mM TrisHCl pH 6.8, 2% SDS, 50 mM DTT 10% (v/v) glycerol). 67 μg were separated via AgPAGE for 3.5 h at 30 mA. The gels were either stained with Alcian Blue or MUC2 was detected by in- gel immunodetection. For in- gel immunodetection, the gels were fixed in 50% (v/v) 2- Propanol/ 5% (v/v) acetic acid for 15 minutes and gentle shaking followed by 30 minutes washing in water. The primary antibody against MUC2 (Genentech; 1:500) was added for 12 hours at 4°C in PBS- T buffer containing 5% BSA. After three washing steps with PBS- T for 10 minutes each, the secondary antibody \(\alpha\) - rabbit- IgG- Licor790 (LiCOR, 1:5000) was added for one hour at ambient temperature. After three to five extensive additional washing steps, the immunostained gel was scanned with a LiCOR Clx instrument.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 675, 273, 692]]<|/det|>
+## Thermofluor assay  
+
+<|ref|>text<|/ref|><|det|>[[115, 706, 868, 888]]<|/det|>
+Mucus from the indicated mouse strains was scraped from their distal colons and emulsified in TBS buffer. Insoluble mucins were washed twice in TBS and recovered by centrifugation (16,000 x g; 4°C; 30 minutes). The protein concentration of the supernatant was determined and the mucus pellet emulsified to a concentration of 1 mg/ml in each sample. 45 μl of sample or TBS control were mixed with five μl of a 200- fold stock solution of SyproOrange (Molecular Probes) and subjected to an increasing temperature gradient of 0.5°C every 30
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 880, 300]]<|/det|>
+seconds from 25 to \(99^{\circ}\mathrm{C}\) in a CFX96 Real- time system (BioRad). The fluorescence was recorded every 30 seconds and the fluorescence intensity of the TBS control subtracted. To rescue the properties of mucus from WT mice 1 U recombinant TGM3 and \(4\mathrm{mM CaCl}_2\) were applied to the mucus from \(Tgm3^{- / - }\) mice and incubated for one hour at \(37^{\circ}\mathrm{C}\) . The reaction was terminated by the addition of \(5\mathrm{mM}\) IAA. The buffer controls for this part of the experiment were treated accordingly and the melting curve recorded as described above. Three biological replicates were analyzed in technical triplicates.  
+
+<|ref|>text<|/ref|><|det|>[[115, 345, 880, 465]]<|/det|>
+Analysis of MUC2 depolymerization by turbidity measurement Scraped mucus samples were adjusted to \(1\mathrm{mg / ml}\) and precipitated by centrifugation (1,000 x g, 30 minutes, \(4^{\circ}\mathrm{C}\) ). Afterwards, the turbidity of the supernatant was recorded at \(600\mathrm{nm}\) wavelength in a Spectramax photometer.  
+
+<|ref|>text<|/ref|><|det|>[[117, 510, 402, 528]]<|/det|>
+Single cell transcriptomic analysis  
+
+<|ref|>text<|/ref|><|det|>[[115, 540, 870, 758]]<|/det|>
+Goblet cells and non- goblet cells from the RedMUC2 reporter mouse strain were isolated by FACS as described recently<sup>20</sup>. The used bulk RNA- seq data (GSE144363) are deposited in GEO and belong to the superserie GSE144436. The quality of the data was assessed with FastQC (version 0.11.2) and filtered using Prinseq (version 0.20.3). The reads were aligned against the mouse reference genome mm10 with STAR (version 2.5.2b) and the number of mapped reads was calculated with HTseq (version 0.6.1p1). Data normalization, differential expression and statistical analysis were made with DESeq2 (version 1.14) in R.  
+
+<|ref|>text<|/ref|><|det|>[[117, 805, 520, 823]]<|/det|>
+In- gel digestion and mass spectrometric analyses  
+
+<|ref|>text<|/ref|><|det|>[[115, 836, 850, 887]]<|/det|>
+Protein bands of interest were excised from the gel and washed with \(50\%\) acetonitrile and dried in a vacuum centrifuge followed by reduction with DTT and alkylation with IAA.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 850, 268]]<|/det|>
+Trypsin was added at a ratio of 1:50 and the samples incubated for 12 hours at \(37^{\circ}\mathrm{C}\) . Afterwards, AspN was added at a ratio of 1:50 and the samples incubated for additional 5 hours at \(37^{\circ}\mathrm{C}\) . The digestion was stopped by adding TFA to a concentration of \(0.5\%\) . Salt and buffer components were removed by in- house stage tips equipped with C18 resin \(^{48}\) and the peptides dissolved in \(0.1\%\) formic acid. The samples were analyzed on a Q- Exactive mass spectrometer as described earlier \(^{49}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 314, 265, 330]]<|/det|>
+## MS Data analysis  
+
+<|ref|>text<|/ref|><|det|>[[115, 345, 880, 626]]<|/det|>
+MS raw files were transformed into \*.mgf files using the MS convert software. These files were analyzed using the MASCOT search engine (Matrix Science). Searches were performed against the UniProt database (version 06/2017 containing 554515 sequences) and an in- house database (http://www.medkem.gu.se/mucinbiology/databases/index.html) containing all human and mouse mucin sequences. Searches were performed with the following parameters: mass tolerance for the precursor ion of 5 ppm; tolerance for fragment ions 0.2 Da; full specificity for trypsin/AspN with a maximum of two missed cleavages; carbamidomethylation as static modification and oxidation of methionine as variable modification.  
+
+<|ref|>text<|/ref|><|det|>[[115, 640, 872, 890]]<|/det|>
+TGM- catalysed cross- linked peptides were searched using the StavroX software tool (version 3.6.6) \(^{50}\) against theoretical intra- and intermolecular isopeptide cross- linked (di)peptides of the murine MUC2 using the following parameters: mass tolerance for the precursor ion of 2 ppm; tolerance for fragment ions 20 ppm; full specificity for trypsin/AspN with a maximum of three missed cleavages; Gln and Lys as cross- linking sites; composition of the cross- link - NH3; carbamidomethylation as static modification and methionine oxidation as variable modification. Label- free mass spectrometric quantification of TGM isozymes was performed as recently described \(^{20}\) .
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[60, 84, 880, 460]]<|/det|>
+599 Data availability600 The proteomics data set for label- free quantification used has been published<sup>20</sup> and deposited to the601 ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) with the dataset602 identifier PXD011527. The bulk RNA- seq data (GSE144363) are deposited in GEO and belong603 to the superserie GSE144436<sup>20</sup>.604 Statistical analysis605 Statistical analyses were performed using the Prism software (version 9.0.1; GraphPad).606 Body weight and colon length were compared using the unpaired t- test with Welch’s607 correction. DAI scores were compared by multiple unpaired t- tests using the Holm- Sidák608 correction. Significance was accepted when p values were below 0.05. Data are expressed as609 mean ± standard deviation.610  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 509, 288, 526]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[117, 541, 878, 690]]<|/det|>
+We acknowledge Ludvig Sollid, University of Oslo and Eleonara Candi, University of Rome for providing the \(Tgm2^{- / - }\) and \(Tgm3^{- / - }\) mice strains. This work was supported by the European Research Council ERC (694181), National Institute of Allergy and Infectious Diseases (U01AI095473, the content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH), The Knut and Alice Wallenberg Foundation (2017.0028), Swedish Research Council (2017- 00958), The Swedish Cancer Foundation (CAN 2017/360), IngaBritt and Arne Lundberg Foundation (2018- 0117), Sahlgren's University Hospital (ALFGBG- 440741, The ALF agreement 236501), Bill and Melinda Gates Foundation (OPP1202459), Wilhelm and Martina Lundgren's Foundation.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[67, 85, 304, 101]]<|/det|>
+## 624 Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[67, 115, 856, 330]]<|/det|>
+624 Author contributionsJDAS performed experiments and analyzed data; BD performed experiments and analyzed data; EELN performed experiments and analyzed data, LA performed experiments and analyzed data; GMHB performed experiments and analyzed data; BMA performed experiments and analyzed data; MEVJ data analysis; GCH conceptualized the study, analyzed data; CVR conceptualized the study, performed experiments and analyzed data. GCH and CVR wrote the paper. All authors reviewed the paper and accepted the final version.  
+
+<|ref|>sub_title<|/ref|><|det|>[[67, 378, 293, 395]]<|/det|>
+## 633 Competing interests  
+
+<|ref|>text<|/ref|><|det|>[[67, 411, 472, 428]]<|/det|>
+634 The authors declare no competing interests.  
+
+<|ref|>sub_title<|/ref|><|det|>[[67, 478, 214, 494]]<|/det|>
+## 636 References  
+
+<|ref|>text<|/ref|><|det|>[[66, 508, 875, 910]]<|/det|>
+637 1. Johansson, M.E. & Hansson, G.C. Immunological aspects of intestinal mucus and mucins. Nat Rev Immunol 16, 639- 649 (2016). 639 2. Rodriguez- Pineiro, A.M. et al. Studies of mucus in mouse stomach, small intestine, and colon. II. Gastrointestinal mucus proteome reveals Muc2 and Muc5ac accompanied by a set of core proteins. Am J Physiol Gastrointest Liver Physiol 305, G348- 356 (2013). 642 3. Atuma, C., Strugala, V., Allen, A. & Holm, L. The adherent gastrointestinal mucus gel layer: thickness and physical state in vivo. Am J Physiol Gastrointest Liver Physiol 280, G922- 929 (2001). 645 4. Johansson, M.E. et al. The inner of the two Muc2 mucin- dependent mucus layers in colon is devoid of bacteria. Proc Natl Acad Sci U S A 105, 15064- 15069 (2008). 647 5. Van der Sluis, M. et al. Muc2- deficient mice spontaneously develop colitis, indicating that MUC2 is critical for colonic protection. Gastroenterology 131, 117- 129 (2006). 649 6. Velcich, A. et al. Colorectal cancer in mice genetically deficient in the mucin Muc2. Science 295, 1726- 1729 (2002). 651 7. van der Post, S. et al. Structural weakening of the colonic mucus barrier is an early event in ulcerative colitis pathogenesis. Gut 68, 2142- 2151 (2019). 652 8. Svensson, F., Lang, T., Johansson, M.E.V. & Hansson, G.C. The central exons of the human MUC2 and MUC6 mucins are highly repetitive and variable in sequence between individuals. Sci Rep 8, 17503 (2018). 656 9. Hansson, G.C. Mucins and the Microbiome. Annu Rev Biochem 89, 769- 793 (2020). 657 10. Godl, K. et al. The N terminus of the MUC2 mucin forms trimers that are held together within a trypsin- resistant core fragment. The Journal of biological chemistry 277, 47248- 47256 (2002). 660 11. Javitt, G. et al. Assembly Mechanism of Mucin and von Willebrand Factor Polymers. Cell 183, 717- 729 e716 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[55, 78, 880, 900]]<|/det|>
+662 12. Recktenwald, C.V. & Hansson, G.C. The Reduction-insensitive Bonds of the MUC2 Mucin Are Isopeptide Bonds. The Journal of biological chemistry 291, 13580- 13590 (2016). 663 13. Lorand, L. & Graham, R.M. Transglutaminases: crosslinking enzymes with pleiotropic functions. Nat Rev Mol Cell Biol 4, 140- 156 (2003). 665 14. Folk, J.E. & Finlayson, J.S. The epsilon-(gamma-glutamyl)lysine crosslink and the catalytic role of transglutaminases. Adv Protein Chem 31, 1- 133 (1977). 667 15. Eckert, R.L. et al. Transglutaminase regulation of cell function. Physiol Rev 94, 383- 417 (2014). 670 16. Adamczyk, M., Griffiths, R., Dewitt, S., Knauper, V. & Aeschlimann, D. P2X7 receptor activation regulates rapid unconventional export of transglutaminase- 2. J Cell Sci 128, 4615- 4628 (2015). 673 17. Siegel, M. et al. Extracellular transglutaminase 2 is catalytically inactive, but is transiently activated upon tissue injury. PLoS One 3, e1861 (2008). 675 18. Telci, D. & Griffin, M. Tissue transglutaminase (TG2)- - a wound response enzyme. Front Biosci 11, 867- 882 (2006). 677 19. Eckert, R.L., Sturniolo, M.T., Broome, A.M., Ruse, M. & Rorke, E.A. Transglutaminase function in epidermis. J Invest Dermatol 124, 481- 492 (2005). 679 20. Nystrom, E.E.L. et al. An intercrypt subpopulation of goblet cells is essential for colonic mucus barrier function. Science 372 (2021). 681 21. Sugimura, Y. et al. Screening for the preferred substrate sequence of transglutaminase using a phage- displayed peptide library: identification of peptide substrates for TGASE 2 and Factor XIII. The Journal of biological chemistry 281, 17699- 17706 (2006). 684 22. Yamane, A. et al. Identification of a preferred substrate peptide for transglutaminase 3 and detection of in situ activity in skin and hair follicles. FEBS J 277, 3564- 3574 (2010). 686 23. Axelsson, M.A., Asker, N. & Hansson, G.C. O- glycosylated MUC2 monomer and dimer from LS 174T cells are water- soluble, whereas larger MUC2 species formed early during biosynthesis are insoluble and contain nonreducible intermolecular bonds. The Journal of biological chemistry 273, 18864- 18870 (1998). 690 24. Kasdorf, B.T. et al. Mucin- Inspired Lubrication on Hydrophobic Surfaces. Biomacromolecules 18, 2454- 2462 (2017). 692 25. Johansson, M.E. et al. Bacteria penetrate the inner mucus layer before inflammation in the dextran sulfate colitis model. PLoS One 5, e12238 (2010). 694 26. Jeong, E.M. et al. Transglutaminase 2 is dispensable but required for the survival of mice in dextran sulfate sodium- induced colitis. Exp Mol Med 48, e267 (2016). 696 27. Stamnaes, J., Iversen, R., du Pre, M.F., Chen, X. & Sollid, L.M. Enhanced B- Cell Receptor Recognition of the Autoantigen Transglutaminase 2 by Efficient Catalytic Self- Multimerization. PLoS One 10, e0134922 (2015). 698 28. Gross, S.R., Balklava, Z. & Griffin, M. Importance of tissue transglutaminase in repair of extracellular matrices and cell death of dermal fibroblasts after exposure to a solarium ultraviolet A source. J Invest Dermatol 121, 412- 423 (2003). 702 29. Yildiz, H.M., Speciner, L., Ozdemir, C., Cohen, D.E. & Carrier, R.L. Food- associated stimuli enhance barrier properties of gastrointestinal mucus. Biomaterials 54, 1- 8 (2015). 704 30. Di Maio, S. & Carrier, R.L. Gastrointestinal contents in fasted state and post- lipid ingestion: in vivo measurements and in vitro models for studying oral drug delivery. J Control Release 151, 110- 122 (2011). 707 31. Wongdee, K., Rodrat, M., Teerapornpuntakit, J., Krishnamra, N. & Charoenphandhu, N. Factors inhibiting intestinal calcium absorption: hormones and luminal factors that prevent excessive calcium uptake. J Physiol Sci 69, 683- 696 (2019). 710 32. Ahvazi, B., Kim, H.C., Kee, S.H., Nemes, Z. & Steinert, P.M. Three- dimensional structure of the human transglutaminase 3 enzyme: binding of calcium ions changes structure for activation. EMBO J 21, 2055- 2067 (2002).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 80, 875, 840]]<|/det|>
+713 33. Ahvazi, B., Boeshans, K.M., Idler, W., Baxa, U. & Steinert, P.M. Roles of calcium ions in the activation and activity of the transglutaminase 3 enzyme. The Journal of biological chemistry 278, 23834- 23841 (2003). 716 34. Yoo, S.H. Secretory granules in inositol 1,4,5-trisphosphate-dependent Ca2+ signaling in the cytoplasm of neuroendocrine cells. FASEB J 24, 653- 664 (2010). 717 35. Cheng, T. et al. Cystatin M/E is a high affinity inhibitor of cathepsin V and cathepsin L by a reactive site that is distinct from the legumain-binding site. A novel clue for the role of cystatin M/E in epidermal cornification. The Journal of biological chemistry 281, 15893- 15899 (2006). 722 36. Walden, M., Crow, A., Nelson, M.D. & Banfield, M.J. Intramolecular isopeptide but not internal thioester bonds confer proteolytic and significant thermal stability to the S. pyogenes pilus adhesin Spy0125. Proteins 82, 517- 527 (2014). 725 37. Hagan, R.M. et al. NMR spectroscopic and theoretical analysis of a spontaneously formed Lys-Asp isopeptide bond. Angew Chem Int Ed Engl 49, 8421- 8425 (2010). 727 38. Wu, F. & Chakravarti, S. Differential expression of inflammatory and fibrogenic genes and their regulation by NF- kappaB inhibition in a mouse model of chronic colitis. J Immunol 179, 6988- 7000 (2007). 730 39. Lyons, J. et al. Integrated in vivo multiomics analysis identifies p21- activated kinase signaling as a driver of colitis. Sci Signal 11 (2018). 732 40. Bognar, P. et al. Reduced inflammatory threshold indicates skin barrier defect in transglutaminase 3 knockout mice. J Invest Dermatol 134, 105- 111 (2014). 734 41. De Laurenzi, V. & Melino, G. Gene disruption of tissue transglutaminase. Mol Cell Biol 21, 148- 155 (2001). 736 42. Frezza, V. et al. Transglutaminase 3 Protects against Photodamage. J Invest Dermatol 137, 1590- 1594 (2017). 738 43. Trigwell, S.M., Lynch, P.T., Griffin, M., Hargreaves, A.J. & Bonner, P.L. An improved colorimetric assay for the measurement of transglutaminase (type II) -(gamma- glutamyl) lysine cross- linking activity. Anal Biochem 330, 164- 166 (2004). 741 44. Gustafsson, J.K. et al. An ex vivo method for studying mucus formation, properties, and thickness in human colonic biopsies and mouse small and large intestinal explants. Am J Physiol Gastrointest Liver Physiol 302, G430- 438 (2012). 743 45. Birchenough, G.M., Nystrom, E.E., Johansson, M.E. & Hansson, G.C. A sentinel goblet cell guards the colonic crypt by triggering NLRp6- dependent Muc2 secretion. Science 352, 1535- 1542 (2016). 747 46. Friedman, D.J. et al. From the Cover: CD39 deletion exacerbates experimental murine colitis and human polymorphisms increase susceptibility to inflammatory bowel disease. Proc Natl Acad Sci U S A 106, 16788- 16793 (2009). 750 47. Schulz, B.L., Packer, N.H. & Karlsson, N.G. Small- scale analysis of O- linked oligosaccharides from glycoproteins and mucins separated by gel electrophoresis. Anal Chem 74, 6088- 6097 (2002). 753 48. Rappsilber, J., Ishihama, Y. & Mann, M. Stop and go extraction tips for matrix- assisted laser desorption/ionization, nanoelectrospray, and LC/MS sample pretreatment in proteomics. Anal Chem 75, 663- 670 (2003). 756 49. Fernandez- Blanco, J.A. et al. Attached stratified mucus separates bacteria from the epithelial cells in COPD lungs. JCI Insight 3 (2018). 758 50. Gotze, M. et al. StavroX- - a software for analyzing crosslinked products in protein interaction studies. J Am Soc Mass Spectrom 23, 76- 87 (2012).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 85, 255, 103]]<|/det|>
+## Figure Legends  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 149, 817, 199]]<|/det|>
+## Figure 1: mRNA expression, protein abundance, and spatial localization of TGM isozymes in the large intestine.  
+
+<|ref|>text<|/ref|><|det|>[[115, 214, 869, 350]]<|/det|>
+(a) mRNA-seq expression data of the goblet cell and non-goblet cell fraction from a reporter mouse strain expressing fluorescently-labelled MUC2. Goblet cells were separated from other epithelial cell types using FACS-mediated cell sorting<sup>20</sup>. The graph shows the normalized expression levels of the transglutaminase family members \(Tgm1 - 7\) and \(F13a1\) in the goblet cell and non-goblet cell fraction. Four biological replicates were analyzed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 378, 868, 529]]<|/det|>
+(b) Label-free relative quantification of TGM isozymes 2 and 3 in goblet cell and remaining epithelial cells after FACS-mediated cell sorting from RedMUC2<sup>98trTg</sup> mice<sup>20</sup>. After protein extraction, the abundance of TGM2 and 3 in the two fractions was measured by mass spectrometry and the data analyzed using the MaxQuant software. Four biological replicates were analyzed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 541, 866, 757]]<|/det|>
+(c) Confocal microscopy of large intestinal tissue specimens from C57/BL6, \(Tgm3^{-/ - }\) and \(Tgm2^{-/ - }\) mice suggests no TGM2 biosynthesis in the colon. The sections were probed with a monoclonal antibody against TGM2 followed by detection with a secondary antibody coupled to Alexa Fluor 647 (red) and sections counterstaining with the UEA1 lectin coupled to rhodamine (green) for goblet cell and mucus visualization. Nuclei are shown in grey and were visualized using the Sytox green stain. The scale bar corresponds to 20 μm. Images are representative of three biological replicates.  
+
+<|ref|>text<|/ref|><|det|>[[115, 771, 878, 857]]<|/det|>
+(d) Analogously, confocal microscopy of colon specimen from C57/BL6, \(Tgm3^{-/ - }\) and \(Tgm2^{-/ - }\) mice analyzed for TGM3 (red) using a polyclonal anti-TGM3 antibody that was detected by a secondary antibody coupled to Alexa 647 indicating TGM3 biosynthesis in WT and \(Tgm2^{-/ - }\)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 83, 850, 135]]<|/det|>
+mice. UEA1 (green) and Hoechst (grey) were used for counterstaining. Images are representative of three biological replicates. The scale bar corresponds to \(20 \mu \mathrm{m}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 860, 430]]<|/det|>
+(e) Protein abundance analysis of TGM isoforms by Western blot in colonic mucus. The supernatant of precipitated mucus was analyzed for the presence of TGM2 and 3 using a monoclonal anti-TGM2 antibody and a polyclonal anti-TGM3 antibody. Goat anti-mouse IgG1-isoform antibody coupled to an IR680 dye and anti-rabbit IgG's coupled to an IR790 dye were used for visualization on a LI-COR Odyssey Clx workstation. Recombinant non-activated or calcium-activated TGM2 and 3 were loaded as positive controls. The red dashed line marks the IgG1 heavy chain (IgG1-HC) recognized by the secondary antibody against the TGM2 antibody and served as loading control. A representative analysis with three biological replicates per mouse strain is shown.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 477, 725, 495]]<|/det|>
+## Figure 2: Qualitative, quantitative and ex vivo analysis of extracellular  
+
+<|ref|>title<|/ref|><|det|>[[117, 510, 344, 528]]<|/det|>
+# transglutaminase activity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 541, 870, 900]]<|/det|>
+(a) Qualitative determination of calcium-induced transglutaminase activity in colonic mucus samples. Samples from the indicated strains were spiked with biotinylated TGM2 (T26) and TGM3 (E51) selective acyl-acceptor peptide substrates and with calcium addition in the absence or presence of IAA followed by incubation for one hour at \(37^{\circ}\mathrm{C}\) . The reaction products were separated by SDS-PAGE and subsequently visualized by Western blot using streptavidin labelled with an IR680LT-dye on a LiCOR Odyssey Clx imager. Non-specific signals from endogenously biotinylated proteins were marked with a triangle. A representative example of three biological replicates per mouse strain is shown.  
+(b) Qualitative determination of intrinsic transglutaminase activity in colonic mucus samples. Samples from the indicated strains were supplied with biotinylated TGM2 (T26) and TGM3-(E51) specific acyl-acceptor peptide substrates without calcium addition in the absence or
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 83, 877, 430]]<|/det|>
+presence of IAA and incubated for one hour at \(37^{\circ}\mathrm{C}\) . The reaction products were separated by SDS- PAGE followed by Western blot detection using streptavidin labelled with an IR680LT- dye. Non- specific signals from endogenously biotinylated proteins were marked with a triangle.A representative example of three biological replicates per mouse strain is shown. (c) Detection of putative acyl- acceptor proteins in mucus. Mucus samples from the different mouse strains were incubated in the presence of 5- Bioinyl- pentalamine (5- BP) for one hour at \(37^{\circ}\mathrm{C}\) . Control reactions were performed in the presence of IAA for the visualization of false- positive signals. The incorporation of 5- BP was detected by Western Blot using streptavidin labelled with an IR680LT- dye. Non- specific signals from endogenously biotinylated proteins were marked with a triangle. A representative example of three biological replicates per mouse strain is shown.  
+
+<|ref|>text<|/ref|><|det|>[[115, 444, 870, 691]]<|/det|>
+(d) Quantitative determination of transglutaminase activity in colonic mucus samples. TGM activity in mucus from the different mouse strains was determined by the incorporation of TGM2 (T26) and TGM3 (E51) specific peptide substrates or a promiscuous TGM acyl-acceptor peptide (A25) into casein as described under materials and methods. The respective cross-linking activity in the samples was calculated from the calibration curve of the recombinant activated TGM standards and subsequently normalized against the protein concentration of the samples. At least four biological replicates per substrate and mouse strain were analyzed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 706, 870, 890]]<|/det|>
+(e- h) Ex vivo analysis of transglutaminase activity. Tissues were mounted in a perfusion chamber as illustrated (e) and transglutaminase activity probed with the glutamine- donor peptide E51 coupled to FITC (magenta) for 30 minutes at \(37^{\circ}\mathrm{C}\) . After washing away non- incorporated peptide the tissue specimen were analyzed by confocal microscopy. Mucus and nuclei were counterstained with the UEA1 lectin coupled to rhodamine (green) and the Hoechst stain (blue) respectively. The top panels show Z-stacks of the explant with (left) or
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 870, 233]]<|/det|>
+without (right) the UEA1 counterstain. The bottom panels show x/y projections of the indicated area from the respective Z- stack on top. Colonic specimen from WT mice (f), \(Tgm3^{- / - }\) mice (g) or WT mice in the presence of the pan- TGM inhibitor Z- DON (h) were probed for E51 incorporation. The scale bar corresponds to \(50 \mu \mathrm{m}\) . Three animals per mouse strain were analyzed.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 280, 782, 300]]<|/det|>
+## Figure 3: Loss of TGM3 causes biochemical alterations of mucus and MUC2.  
+
+<|ref|>text<|/ref|><|det|>[[115, 312, 881, 462]]<|/det|>
+(a) Schematic figure of the human and mouse MUC2 domain structure. The domains of the complete sequence excluding the signal sequence are shown. The abbreviations correspond to vWD, von-Willebrand D domain; CysD, Cystein-rich domain; PTS, Proline, Serine, Threonine-rich domain that after \(O\) -glycosylation becomes a mucin domain; vWC, von-Willebrand C domain; CK, Cysteine-knot domain.  
+
+<|ref|>text<|/ref|><|det|>[[115, 476, 877, 594]]<|/det|>
+(b) MUC2 mono- and oligomers from WT, \(Tgm2^{- / - }\) , and \(Tgm3^{- / - }\) colonic mucus were separated by composite AgPAGE and stained by in-gel immunodetection using a polyclonal anti-MUC2-C3 (Genentech) and secondary antibody coupled to the AlexaFlour790-dye on an Odyssey Clx workstation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 607, 877, 727]]<|/det|>
+(c) Limited Proteolysis of MUC2 by the serine protease Lys-C. Mucus samples from the indicated mouse strains were incubated in the absence or presence of Lys-C for 90 minutes at \(25^{\circ}\mathrm{C}\) and the reaction products separated via composite AgPAGE followed by visualization of MUC2 with Alcian Blue.  
+
+<|ref|>text<|/ref|><|det|>[[115, 739, 866, 890]]<|/det|>
+(d) Heat map of the sequence coverage of MUC2 domains. The color coded sequence coverage of the different MUC2 domains from three biological replicates of non-treated MUC2 monomers from mucus samples of WT (WT-M) and \(Tgm3^{- / - }\) ( \(Tgm3^{- / - }\) -M) animals as indicated in Fig. 3c are shown. The various MUC2 domains are placed from the N-terminus (top) to the C-terminus (bottom) on the ordinate. Only peptides with an ion score \(>25\) were
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 857, 135]]<|/det|>
+taken into consideration. The two PTS domains were excluded as they are due to their high glycosylation heterogeneity not analyzable.  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 860, 266]]<|/det|>
+(e) Detection of transglutaminase reaction products. Example of an isopeptide dipeptide cross-link that was solely detected in MUC2 from WT animals. MS2 fragment spectrum of the parent ion [M+2H]2+ 775.44 is shown. B ions are labelled in red and y ions in blue. The prarent ion is labelled in green.  
+
+<|ref|>text<|/ref|><|det|>[[115, 279, 872, 397]]<|/det|>
+(f) Analysis of MUC2 polymerisation. Mucus samples from WT and the TGM knock-out strains and their protein concentrations adjusted to 1 mg/ml. After centrifugation (1,000 x g, 30 minutes, 4°C), precipitating the insoluble MUC2, the absorbance for soluble material was recorded at 600 nm. Three biological replicates per strain were analyzed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 410, 879, 693]]<|/det|>
+(g) Hydrophobicity analysis of mucus from WT and \(Tgm3^{-/ - }\) mice. Mucus samples were adjusted to a protein concentration 1 mg/ml in TBS. SyproOrange was added and the melting curve of the samples analyzed by increasing the temperature by 0.5°C every 30 seconds from 25°C to 99°C in a thermocycler. The fluorescence change was recorded every 30 seconds and subsequently normalized by subtraction of the buffer control fluorescence for each data point. For rescuing the WT behavior, 1 U of activated recombinant TGM3 was added to mucus from \(Tgm3^{-/ - }\) animals and the samples incubated for 60 minutes at 37°C. Afterwards the cross-linking activity was inhibited with IAA before adding SyproOrange. The graph shows the arithmetic average of three biological replicates.  
+
+<|ref|>text<|/ref|><|det|>[[115, 706, 875, 890]]<|/det|>
+(h-i) Pronase-treatment of distal colon specimen from WT (h) and \(Tgm3^{-/ - }\) (i) mice. Colonic explants from WT and \(Tgm3^{-/ - }\) mice were mounted in a chamber and pronase in Krebs-buffer was added before examination under a confocal microscope as sketched in Fig. 2e. The mucus surface was visualized by placing fluorescently labelled beads with a diameter of one \(\mu \mathrm{m}\) on top of the mucus layer and the epithelium counterstained using the Syto 9 stain. The top panel shows the isosurfaces of the tissue and of the individual beads over time. The white
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 868, 201]]<|/det|>
+scale bar corresponds to \(50 \mu \mathrm{m}\) . The lower panel shows the distribution of the fluorescently- labelled beads in relation to the tissue surface as violin plot where the black bar marks the median of bead distance from the epithelium. Three animals per mouse strain were analysed. OM=outer (loose) mucus layer, IM=inner mucus layer.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 247, 800, 266]]<|/det|>
+## Figure 4: Dextran sodium sulfate treatment shows decreased mucus protection.  
+
+<|ref|>text<|/ref|><|det|>[[115, 280, 876, 330]]<|/det|>
+WT and \(\mathrm{Tgm3 - / - }\) mice were cohoused and supplied via drinking water with \(3\%\) (w/v) dextran sodium sulfate.  
+
+<|ref|>text<|/ref|><|det|>[[115, 345, 850, 460]]<|/det|>
+(a) Body weight change of DSS-treated mice over time. The body weight of the mice was recorded once per day throughout the whole experiment and the change in body weight respective to the starting body weight of both groups plotted against the time. The graph shows the comparison of one litter per strain consisting of five animals in each group.  
+
+<|ref|>text<|/ref|><|det|>[[115, 475, 870, 628]]<|/det|>
+(b) Detection of occult blood in feces. Fecal samples were collected from the DSS-treated animals and analyzed for hidden blood using a hemoccult kit as described in materials and methods. The mean ratio of hemoccult-positive samples from each group was plotted against time. The graph shows the comparison of one litter per strain consisting of five animals in each group.  
+
+<|ref|>text<|/ref|><|det|>[[115, 641, 874, 752]]<|/det|>
+(c) The disease activity index (DAI) was determined as sum of the changes in body weight, stool consistency, rectal bleeding for every animal for the indicated time points and the mean with standard deviation for the both groups plotted against the time. \(*p< 0.05\) . The graph shows the comparison of one litter per strain consisting of five animals in each group.  
+
+<|ref|>text<|/ref|><|det|>[[115, 770, 870, 889]]<|/det|>
+(d) Survival analysis of DSS-treated mice. The probability of survival was calculated using the GraphPad prism software. Mice were sacrificed when the initial body weight loss exceeded \(10\%\) . The graph shows the summary of the two litters per strain that were analysed independently and represents ten animals in each group.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 81, 868, 265]]<|/det|>
+(e-f) Colon length changes of DSS-treated WT and \(Tgm3^{- / - }\) mice. At day 8 or at the ethical endpoint, animals were sacrificed and the colon length of each animal measured. A representative colon of WT and \(Tgm3^{- / - }\) animals is shown in (e). (f) Normalized colon length of DSS-treated WT and \(Tgm3^{- / - }\) mice. The graph shows the summary of the two litters per strain that were analysed independently and represents ten animals in the WT and nine animals in the \(Tgm3^{- / - }\) group. \(p< 0.001\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 279, 864, 496]]<|/det|>
+(g) Histological analysis of DSS-treated WT and \(Tgm3^{- / - }\) mice. Representative Alcian Blue-Periodic Acid Schiff-stained sections from proximal (PC) and distal colon (DC) of WT and TGM3-deficient animals are shown. The black scale bar on the left corresponds to \(100 \mu \mathrm{m}\) . (h) Immunohistochemical analysis of WT and \(Tgm3^{- / - }\) mice for the presence of TGM2 after DSS treatment. Tissue specimen from WT and \(Tgm3^{- / - }\) animals were probed with a monoclonal anti-TGM2 antibody (red) and the UEA1 lectin (green). Nuclei were stained using the Hoechst stain (grey). The white scale bar on the right corresponds to \(30 \mu \mathrm{m}\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 997, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[43, 42, 311, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[43, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 350, 230]]<|/det|>
+SupplementaryFiguresS14. pdf SupplementaryMovieM1. mp4 SupplementaryMovieM2. mp4 SupplementaryMovieM3. mp4
+
+<--- Page Split --->

preprint/preprint__7e6f2e24c1c8d4e3aaa13e2875ea7c7452d975c29a599bbb90931019f462e57c/images_list.json

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

preprint/preprint__7e6f2e24c1c8d4e3aaa13e2875ea7c7452d975c29a599bbb90931019f462e57c/preprint__7e6f2e24c1c8d4e3aaa13e2875ea7c7452d975c29a599bbb90931019f462e57c.mmd

@@ -0,0 +1,436 @@
+
+# Effect of mRNA vaccination and previous infections on SARS-CoV-2 transmission across four variants: adjusted analysis of 111'432 declared contacts  
+
+Denis Mongin ( denis.mongin@hcuge.ch) University of Geneva  
+
+Nils Burgisser University Hospitals of Geneva  
+
+Gustavo Laurie Geneva Directorate of Health  
+
+Guillaume Schimmel Geneva Directorate of Health  
+
+Diem- Lan Vu Geneva Directorate of Health  
+
+Stephane Cullati University of Geneva  
+
+Delphine Courvoisier University Hospitals of Geneva  
+
+## Article  
+
+## Keywords:  
+
+Posted Date: February 7th, 2023  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 2510736/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 6th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 41109- 9.
+
+<--- Page Split --->
+
+Effect of mRNA vaccination and previous infections on SARS-CoV-2 transmission across four variants: adjusted analysis of 111'432 declared contacts  
+
+Denis Mongin<sup>1,\*</sup>, Nils Bürgisser<sup>2</sup>, Gustavo Laurie<sup>3</sup>, Guillaume Schilmmel<sup>3</sup>, Diem- Lan Vu<sup>1,3,4,5</sup>, Stephane Cullati<sup>6,7</sup>, Delphine Sophie Courvoisier<sup>1,6</sup>, and the Covid- SMC Study Group<sup>†</sup>  
+
+<sup>1</sup>Faculty of Medicine, University of Geneva, Geneva, Switzerland  <sup>2</sup>General internal medicine division, Department of Medicine, Geneva University Hospitals, Geneva, Switzerland  <sup>3</sup>Division of General cantonal physician, Geneva Directorate of Health, Geneva, Switzerland  <sup>4</sup>Division of Infectious Diseases, Geneva University Hospitals, Geneva, Switzerland.  <sup>5</sup>Laboratory of Virology, Division of Laboratory Medicine, Geneva University Hospitals, Geneva, Switzerland.  <sup>6</sup>Division Quality of care, Faculty of Medicine, University of Geneva, Geneva, Switzerland  
+
+†Membership of the Covid- SMC Study Group is provided in the appendix.  
+
+\*Correspondance to: Denis Mongin +41 223723642 denis.mongin@unige.ch Hôpital Beau séjour, service de rhumatologie. 26 avenue de Beau Séjour 1206 Genève Switzerland
+
+<--- Page Split --->
+
+## Abstract  
+
+The immunity conferred by SARS- CoV- 2 vaccines and infections reduces the transmission of the virus. But it is not clear how the effect of immunity is shared between a reduction of the contagiousness and an increased protection against infection. To answer this question, we used a register of \(>50^{\prime}000\) SARS- CoV- 2 positive index cases and their \(>110^{\prime}000\) declared contacts to estimate the association between secondary attack rate and immunity status (natural infection, vaccine, or both). Analyses were stratified per four SARS- CoV- 2 variants and adjusted for index cases and contacts sociodemographic characteristics and the propensity of the contacts to be tested.  
+
+The reduction of the propagation conferred by immunity was mainly a protection of the contacts against infection rather than a diminution of the contagiousness of the index cases. The largest immunity effect was conferred for both by natural infection, especially when recent for the contacts. Although of smaller amplitude, the effect of vaccination for index, i.e. the reduction of contagiousness, was less affected by SarS- CoV- 2 variant and time since vaccination than the effect for contacts, i.e. the increase of protection against infection. Indeed, vaccination offered protection against infection only if given less than 6 months before and only for the variants preceding omicron, while the vaccination kept moderately reducing the index infectivity during the omicron wave, even when given more than 6 months ago. Hybrid immunity (vaccination and infection) did not have increased effectiveness than recent infection. These findings support the idea that vaccination also protects others, and highlight the need for the implementation of non- personal intervention reducing Sars- CoV- 2 propagation, such as ventilation or air filtration.  
+
+## Introduction  
+
+Since its worldwide spread at the beginning of \(2020^{1}\) , the SARS- CoV- 2 virus has caused one of the most important health burden in recent history. It is estimated to have caused 18 million deaths as of end of 2021. SARS- CoV- 2 became a leading cause of death in some countries in these years and is responsible for an important burden of long lasting symptoms in the population. Its widespread circulation within human communities and possible animal reservoirs allows the SARS- CoV- 2 to mutate frequently and has resulted so far in more contagious, immunity- escaping variants 7,8 responsible for successive waves of infections worldwide.  
+
+The effect of immunity on the transmission of the successive SARS- CoV2 variants and its evolution in time are key factors for our understanding of the SARS- CoV2 propagation. Immunity can be acquired through vaccination or through natural infection. SARS- CoV2 mRNA vaccines have been shown to be effective in preventing re- infection shortly after injection. However, the immunity it confers wanes rapidly 10- 12 and a roll- out of booster vaccinations has been implemented in western countries to maintain an immunity against SARS- CoV2 13- 15. It was recently demonstrated that natural infection confers a stronger and longer lasting protection against reinfection than vaccination 7,16- 18, and that the combination of both type of immunity (hybrid immunity) may provide an even stronger protection 18,19.  
+
+Less is known, however, on the effect of immunity on the susceptibility to contaminate others, especially with regard to natural immunity. Depending on the variant of concern (VoC) considered, studies analysing the secondary attack rate show little to no effect of vaccine on being contagious, while recent in vitro studies, by measuring viral load and propagation, indirectly suggest that natural infection could better reduce the infectiousness than vaccine 21,22.
+
+<--- Page Split --->
+
+Secondary attack rate (SAR) is a good measure of SARS- Cov2 transmission, providing a full picture of both the protection against getting infected and the diminution of the contagiousness that the immunity may confer. Apart from the immunity of the population and the VoC considered, SAR is known to vary greatly by contact settings, ranging from \(20\%\) in households to \(6\%\) in social gatherings during the first year of the pandemic \(^{23 - 26}\) , but also by the symptoms of the index cases \(^{27,28}\) and the socio- demographic characteristics of the studied population \(^{26,28 - 31}\) . By definition, SAR also depends directly on the capacity to detect SARS- CoV- 2 infections among contacts, which includes the propensity of the contacts to get tested.  
+
+Using a register dataset of \(50'889\) index cases having declared 111'432 contacts in the State of Geneva \(^{32}\) , we propose to study the effect of the immune status on SARS- CoV2 transmission, considering vaccination and natural infection of index and contacts and main SARS- CoV- 2 variants while adjusting for demographic, social and health factors as well as the tendency to get tested for SARS- CoV- 2.  
+
+## Methods  
+
+## Setting and period  
+
+Data used for the present study consisted in a register dataset of links between an infected case (hereafter the index case) and a declared person with whom he/she had close contact during the 10 days preceding his/her test result (hereafter the contacts). These data stem from the ARGOS database \(^{32}\) , which is an ongoing operational COVID- 19 database created by the Geneva health state agency (Geneva Directorate of Health).  
+
+Geneva is a state of 511'921 inhabitants as of the last census in December 2021 \(^{33}\) , mainly urban, with a high population density, and which doubles its population on working days (excluding pandemic restrictions) as a result of national and international commuter traffic (mainly from neighbouring France). We used data from the \(26^{\text{th}}\) February of 2020 (first positive tested recorded in Geneva) to \(28^{\text{th}}\) February of 2022. Data was not collected after March \(1^{\text{st}}\) 2022 because contact declaration was stopped at this date in Switzerland.  
+
+This research received the agreement of the Cantonal Ethic Committee of Geneva (CCER protocol 2020- 01273). Participants had the possibility to refuse sharing their data for research through a form that was automatically sent. Those who did were removed from the analysis. Data are available upon request at https://edc.hcuge.ch/surveys/?s=TLT9EHE93C.  
+
+## Index cases and contacts  
+
+The register contains baseline, follow- up, and contact information of all SARS- CoV- 2 positive tested persons (index case) residing in the State of Geneva, Switzerland.  
+
+A contact was considered as infected by the index case if they had a positive COVID- 19 result within 10 days following their last contact with the index case. From February 2020 to end of April 2020, contact information was collected by interviewing the index case. From May 2020, index cases had the possibility to provide their contacts names and phone through an online form. Contacts were then approached using phone interviews. Additionally, an online form was implemented at the end of September 2020 to support the oral interviews, where the contacts had the ability to complete the required information themselves. From mid December 2021, the oral interviews could not be maintained, therefore contact information was only gathered from the online formula.  
+
+Contact information contained the type of contact setting between the index case and the declared contacts (see supplementary material), the date of the last contact between index and contact, the
+
+<--- Page Split --->
+
+birth date, gender, date of subsequent or anterior positive PCR or antigenic test results, as well as the living address and the vaccination dates. Information about the index cases included date of SARS- CoV2 test result, gender, date of birth, living address, presence or absence of any symptoms (see supplementary material), presence or absence of cough (cough, dry cough or wet cough), personal vulnerability (based on if the person reported difficulty to make ends meet, lived in a highly subsidized housing, or if they asked explicitly to avoid police control, see supplementary material), vaccination dates and date of previous infections.  
+
+## Secondary attack rate (outcome)  
+
+The secondary attack rate \(^{34}\) , first described by Dr Chapin at the beginning of the last century, refer to the probability of infection among close contacts of an index case in a particular setting (work, household, ...) \(^{35}\) and is one of the key estimate of the transmissibility of the virus. Its raw estimation consists in dividing the number of contaminated contacts by the total number of susceptible contacts declared by the index cases. Adjusted estimation of SAR can be performed using linear regression methods (see statistical analysis).  
+
+## Immunity status (main predictor)  
+
+Immunity status was calculated at the date of the last contact between the index and the contact and was categorized in the following categories:  
+
+- Vaccinated more than 6 months or less than 6 months. This category included the complete vaccine schemes for the different vaccines available in Geneva, including booster doses, for which the last date of vaccination was more or less than 6 months.- Infected at least one time, more than 6 months or less than 6 months. This category included persons not vaccinated but having at least one positive PCR test result, more or less than one year ago.- Not vaccinated not infected (NVNI). This category included persons not vaccinated and not infected previously to the date of last contact between index case and the contact.- Hybrid infections: persons with complete vaccine scheme and previous infection  
+
+Addresses were geo- coded using the exhaustive list of all addresses of the State of Geneva.  
+
+## Controls  
+
+Categorization of the socio- economic condition of the neighbourhood area (417 official neighbourhood areas in the State of Geneva) was, similarly to previous work \(^{36}\) , based on an index provided by the centre for the analysis of territorial inequalities (see supplementary material). The statistical office of Geneva provided the type of building and number of inhabitant for each address. The building type were categorised in three categories: building, single houses, or collective structure. This last category included nursing homes, jails, asylums and fire- stations.  
+
+Body mass index (BMI) was calculated from height and weight and was categorized in obese and non- obese categories. For age superior to 18 years, obese was considered for BMI above \(30 \text{kg} /\text{m}^2\) . For age below 18, we used the extended international body mass index cut- offs corresponding to the threshold of \(30 \text{kg} /\text{m}^2\) at 18 years old \(^{37}\) .  
+
+Tendency to test was estimated by counting the number of tests performed by each contact during the last 3 months preceding their contact with the index case. This number was categorized in three categories: 0, 1 and more than 2 (2+).
+
+<--- Page Split --->
+
+## SARS-CoV-2 variants  
+
+SARS- CoV- 2 variantsAs the ARGOS data did not contain information about the SARS- CoV- 2 variant type, we divided the study period into period of predominance of SARS- CoV- 2 variant of interest, based on the data provided by the Global Initiative on Sharing Avian Influenza Data \(^{38}\) in the Geneva region:  
+
+EU1 from 01- 06- 2020 to 01- 02- 2021 Alpha from 02- 02- 2021 to 01- 07- 2021 Delta from 02- 07- 2021 to 20- 12- 2021 Omicron from 21- 12- 2021 to 01- 03- 2022 (mainly BA.1)  
+
+## Statistical analysis  
+
+SAR was estimated using generalized estimating equations predicting a binary outcome indicating if the contact was contaminated by the index or not. The clusters considered were the index cases, and we assumed an exchangeable correlation structure. We used a Gaussian identity link \(^{39}\) , which allows to estimate the relative proportion increase provoked by each covariate relative to a reference proportion of infected contacts, that is the SAR. Missing data were handled using multiple imputation with chained equations (20 samples, 5 iterations) at the person, infection or contact level (see supplementary material). The analysis was then performed independently on each imputed dataset, and the results were pooled according to the Rubin's rules.  
+
+All analysis has been performed using R 4.0.0 \(^{40}\) , using the geepack library \(^{41}\) for the general estimating equation, mice \(^{42}\) for the multiple imputation with chained equation and ggplot2 for the figures and graphs. The code used for the analysis- has been made available at the following Gitlab repository: https://gitlab.com/dmongin/scientific_articles/- /tree/main/Effect_of_mRNA_vaccination.  
+
+## Results  
+
+During the period of interest (01- 06- 2020 to 01- 03- 2022), 65'077 infections were recorded among persons living in Geneva and who declared at least one contact person. Among them, 9'890 refused to share their data for research. 15'327 declared contacts also refused to share their data, removing an additional 4'298 infections. The resulting dataset consisted in 50'889 index cases and 111'432 declared contacts. The mean number of declared contact per infected person was 2.2 overall, with a net decrease during the Omicron period (1.6 mean contacts per index, see table 1).  
+
+Index cases were at \(73\%\) adults between 18 and 64 years, \(22\%\) children and \(4.6\%\) adults older than 65 years. The proportion of children for the index cases tripled between the EU1 wave (11%) and the Delta wave (38%). Overall, children were overrepresented and adults \(>65\) years underrepresented in our cohort when compared to the demographics of the Geneva state (18.5% of children, and 16.5% of adults above 65 years, see supplementary table S1).  
+
+The vast majority of the index cases had symptoms (94%), among whom more than half had cough (58%). The majority of the contacts reported by the index were persons sharing their home (62%), this percentage increasing up to 78% during the Omicron wave.  
+
+Concerning the immunity status, the proportion of vaccinated index cases increased from around \(2\%\) during the alpha wave, up to \(52\%\) during the Omicron wave, of which \(25\%\) had their last dose more than 6 months before the infection. Contacts were less vaccinated (37%) and a higher proportion of them were previously infected (10%, compared to 2.9% for the index cases).  
+
+## Overall results  
+
+Among the 111'432 declared contacts, 21'387 had a positive test result during the 10 days following the date of the last contact with the index case (raw SAR of 19.2%). This raw SAR increased almost
+
+<--- Page Split --->
+
+linearly of 3 percent point per day when increasing the delay from 0 to 8 days, to then plateauing after 10 days (see supplementary figure S1). For the rest of the study, a delay of 10 days was considered. The raw SAR changed across variant and was \(16.5\%\) during the EU1 wave, \(21.2\%\) during the alpha wave, \(16.8\%\) during the delta wave, and \(26.3\%\) during the omicron wave.  
+
+The reference category was defined as follow: two asymptomatic adult men, neither vaccinated nor with an antecedent infection (NVNI), between the age of 18 and 65, having a house contact in a building in a wealthy neighbourhood, of which the index case being not obese and not a vulnerable person, the contact person having performed one test in the past three months, and being both not vaccinated not infected. For this reference category, the multivariable analysis yielded a SAR of \(33.3\%\) (95%CI: [30.9,35.7]) for the EU1 variant, \(31.0\%\) (95%CI: [27.8,34.1]) for the alpha variant, \(32.8\%\) (95%CI: [32.0,37.0]) for the delta variant and \(41.0\%\) (95%CI: [37.4,44.6]) for the omicron variant.  
+
+The main variables influencing the SAR (see figure 1) were the immune status of both the index case and the contact, the presence of symptoms or the presence of cough for the index case, the type of relation between the index case and their contacts, the age of the contacts, and the number of test the contact had in the 3 months before the contact date. The age of the index case, as well as the index housing type had a limited effect on the SAR. The gender of both index and contacts, the obesity of the index, the index vulnerability or its neighbourhood socio- economic condition did not affect the SAR.  
+
+## Immune status  
+
+Being previously infected for the index case decreased the SAR, with no obvious difference if the infection was recent, older than 6 months or hybrid (see supplementary table S2 and supplementary figure S2). The reduction of SAR induced by an infection of the index case was of - 9.9 adjusted percent points (pp) (95%CI: [- 14.7, - 6.5]) during the EU1 variant wave, - 8.6pp during the alpha wave [- 13.2, - 4.1], - 10.9pp [- 13.0, - 7.1] during the delta wave and of 4.3pp [- 7.3, - 1.3] during the Omicron wave.  
+
+The effect of previous infection was stronger for the contacts, with a greater effect when the date of infection was less than 6 months before the index- contact date. Previous infection of less than 6 month or more than 6 months respectively lowered the SAR of - 17.3pp [- 19.3, - 15.4] and - 13.3pp [- 15.4, - 11.8] for EU1, - 26.5 pp [- 28.1, - 24.8] and - 20.6pp [- 23.3, - 18.0] for alpha, - 30.7pp [- 32.1, - 29.4] and - 14.9pp [- 16.8, - 12.9] during delta and - 32.0pp [- 33.9, - 30.0] and - 4.4pp [- 7.6, - 1.2] during omicron. Considering an interaction between immune status and testing tendency, the protection caused by previous infection was around - 6pp stronger for all variants if the contacts were tested at least once during the 3 months preceding their encounter with the index (see supplementary table S3 and supplementary figure S4).  
+
+Being vaccinated for the index consistently lowered the SAR across VoCs mainly when the last dose of vaccination was less than 6 months before the index- contact date (- 5.0pp [- 9.5, - 0.4] during alpha, 5.6pp [- 6.9, - 4.3] during delta and - 6.6pp [- 8.2, - 4.9] during omicron). A small but significant protective effect of vaccination older than 6 month was observed during omicron (- 2.8pp [- 4.6, - 0.9]).  
+
+The recent vaccination of the contacts had a strong protective effect for alpha (- 13.5pp [- 16.0, - 11.0]) and delta variants (- 9.5pp [- 10.5, - 8.4]). In this multivariable model without interaction, recent contact vaccination increased the SAR during the omicron wave. This increase vanished when considering an interaction between the immune status and the number of test performed the last 3 months (2.2pp [- 1.8, 6.3] and 2.9pp [- 2.0, 7.8] if the contact performed 1 or more than 2 tests, respectively). If vaccination occurred more than 6 months before the last meeting between index and contact, it did not have a significant effect during the delta variant and even had a net tendency
+
+<--- Page Split --->
+
+to increase the SAR with Omicron (increase of 13.3pp [11.2,15.4]). This increase remained similar even when considering an interaction between the immune status and the number of tests performed 3 months before.  
+
+Hybrid immunity had a higher protective effect than vaccination but lower than recent infection (- 20.1pp [- 25.8, - 14.3], - 21.7pp [- 22.9, - 20.6], and - 18.2p [- 20.4, - 16.1] for the alpha, delta and omicron wave respectively). The combined recent vaccination for both contact and index (interaction between both immune status) decreased the SAR by - 22pp [- 27, - 21] during alpha and - 17pp [- 14, - 18] during delta, but had no significant effect during omicron.  
+
+## Index Reported relationship with contact  
+
+When the contacts between index and contact took place outside the household, the SAR was substantially lower: it decreased by - 9.4pp [- 10.3, - 8.4], - 13.1pp [- 14.7, - 11.6], - 8.4pp [- 9.6, - 7.3], and - 7.2pp [- 8.8, - 5.5] for the EU1, alpha, delta and omicron variants when the index and its contacts were close relatives not living under the same roof. This diminution of SAR was more pronounced when contact and index were not closed relatives (e.g. professional or recreational relation): - 10.7pp [- 11.9, - 9.5], - 15.3pp [- 17.7, - 13.0], - 10.8pp [- 12.5, - 9.2] and - 11.5pp [- 13.9, - 9] for the EU1, alpha, delta and omicron variants respectively.  
+
+## Symptoms  
+
+Cough increased the transmissibility, but with an amplitude decreasing with the new variants. A coughing index increased the SAR by 5.3pp [4.4, 6.2] and 5.5pp [3.9, 7.1] for the EU1 and alpha variant, but this effect was reduced to 2.2pp [1.0, 3.3] for the delta variant and 1.8pp [0.4, 3.3] for omicron.  
+
+## Age  
+
+Contact children had a lower SAR, especially for early variant: the SAR was - 11.7pp [- 12.4, - 10.9] lower and - 8.1pp [- 9.5, - 6.8] lower for EU1 and alpha, when the decrease was only - 0.2pp [- 1, - 0.6] and - 4.8pp [- 6.1, - 3.5] for delta and omicron. Contact older than 65 tended to be more contaminated at the beginning of the pandemic (4.1pp [2.4, 5.8] supplementary points for EU1), but this effect became non- significant for alpha and delta, and even reversed for omicron, where contact older than 65 years had a SAR - 6.5pp [- 9.6, - 3.4] lower than the reference category.  
+
+Concerning the age of the index, index children seemed to be more contagious over time (- 4.9pp [- 6.7, - 3.1], - 2.6pp [- 3.8, - 1.4] and - 2.1pp [- 3.9, - 0.4] during the alpha, delta and omicron waves respectively). Being an infected adult older than 65 years increased contagiosity only during the alpha wave (3.7pp [0.1, 7.3]).  
+
+## Effect of testing during the past 90 days  
+
+The propensity of contacts to perform tests, measured by the number of test performed by the contacts the last 90 days preceding their last encounter with the index, had a large effect on SAR calculation. Those who did not perform any test during this period had a reduced SAR of - 16.3pp [- 17.4, - 15.1], - 11.1pp [- 12.7, - 9.5], - 13.3pp [- 14.6, - 12.0] and - 17.0pp [- 18.6, - 15.5] for the EU1, Alpha, delta and omicron respectively when compared to those who performed one test. Performing two tests or more did not clearly increase the SAR (an increase of 3.0pp [1.1, 4.8] only during the delta wave). The propensity of contacts to perform tests modified the effect of immunity on SAR when comparing univariable and multivariable adjustment. These results are detailed in supplementary material. After adjustment for an interaction between contact immunity of the number of test performed by contact during the past 90 days, testing more enhanced the effect of protection against infection conferred by recent infection and hybrid vaccination, but decreased this effect for
+
+<--- Page Split --->
+
+other immune status, especially during the omicron wave (see supplementary figure S4 and supplementary table S3).  
+
+## Gender differences  
+
+When stratifying for gender, multivariable models showed similar pattern of results. Nevertheless, previous infections of women index were more protective, and were associated with lower SAR for the EU1 and alpha variants. There was no difference for the previous infections of contacts, though they included both gender (see supplementary figure S3).  
+
+## Discussion  
+
+In this study of \(>50'000\) index cases and \(>110'000\) declared contacts, spanning four different SARS- CoV- 2 variants circulating over almost 2 years, we observe that the immunity conferred by vaccine or infection lowers both the transmission risk and the risk of being infected, and that the latter effect contributes more to the reduction of the virus propagation. The main immune factor lowering the secondary attack rate was natural infection, while vaccination had a more limited impact, even when recent enough. Although having a lower contribution to changes of secondary attack rate, the reduction of contagiousness conferred by vaccination appears to wane less in time and to be less sensitive to variant changes than the increase of the protection against infection. The other variables affecting the transmission of SARS- CoV- 2 were the age of the contact person, the presence of symptoms - especially cough - for the index, the setting of the encounter between index and contact (e.g., home, work) and the tendency of the contact to get tested.  
+
+Compared to non- vaccinated and never- infected person, vaccination was protective for both index and contacts, the effect for the index being smaller than for contacts, as reported previously \(^{43}\) . Because of the waning of the vaccine- induced immunity \(^{10}\) and the immune escape of successive variants when compared to the previous ones \(^{8,44}\) , the timing of the last vaccination and the VoC concerned were important, especially for the contact. Vaccination, even when performed less than 6 months before, did not add any protection to contacts during the omicron wave. On the other hand, the escaping capacities of omicron did not affect the reduction of contagiousness conferred by recent vaccination. This suggests that vaccine still lowers the viral load of persons infected by Omicron, in agreement with the fact that vaccine diminishes the occurrence of severe disease for this VoC \(^{13,45}\) .  
+
+Although vaccination more than 6 months ago still had some effect for index cases, it added no protection for the contacts during delta wave, and even had an opposite effect (i.e. an increase of the SAR) during the omicron wave. This counter intuitive effect might be due to a combination of the strong immune escape of this variant \(^{46}\) and the tendency of vaccinated people to comply less with COVID- 19 mitigation strategies \(^{47}\) , such as physical distancing and mask recommendations. Such result has been observed in a previous study \(^{48}\) .  
+
+Infected unvaccinated indexes had a reduced SAR across all variants. This reduction was higher than the one observed for vaccination for Delta, but not for Omicron, in agreement with recent measurement of viral load dynamics \(^{21}\) . Previous infection also showed a strong protective effect against being infected for the contacts, even more after adjusting for their tendency to test. This protection is reduced after 6 months. The reduction of transmission is rather small for early variants, in line with the recently observed slower immunity waning after infection when compared to vaccine \(^{7,16,18}\) , but is substantial for delta and omicron, due to their stronger potential for immune escape \(^{49}\) . This protection against infection was higher than vaccination for all variants (up to 7 times higher for Omicron, in agreement with recent estimate of Gazit and co- authors \(^{17}\) ). This higher and longer lasting protection of the infection when compared to vaccine induced immunity may find its root in a
+
+<--- Page Split --->
+
+more global immune response and maybe specific IgA response \(^{50}\) .  
+
+Hybrid immunity provided stronger protection and reduction of contagiousness than vaccines, as observed elsewhere \(^{51}\) , but no higher than recent infection, in agreement with a large Israeli study \(^{18}\) . The association of SAR and immune status, either due to recent vaccine or previous infection, changed notably between univariable and adjusted analyses. Though previous infection and recent vaccination were protective in univariable models, it became more so, for all VoCs, in the multivariable model. The main confounder of this association was the tendency of the contact to be tested, which modified the SAR of the non- immune population, our reference category. Indeed, the SAR was much higher among non- immune who tested compared to those who did not, whereas SAR was quite similar among previously infected or recently vaccinated people, irrespective of their tendency to get tested.  
+
+It is of note that the adjustment in our analysis corrects strongly the SAR value of each variant, resulting in a similar value for the EU1, alpha and delta variant, but a higher SAR for omicron, similar to what has been reported by large reviews \(^{20}\) .  
+
+The context of the encounter between index and contacts affected greatly the SAR, where more distant relations (work, leisure) led to lower SAR than housing relation, as noticed elsewhere \(^{26}\) . This confirms that the adaptations of measures for pandemic containment during the last waves, such as quarantine restricted to household, was appropriate.  
+
+Symptomatic indexes have consistently been shown to increase SAR since the beginning of the pandemic \(^{27,28}\) . However, the difference in SAR between symptomatic and asymptomatic is relatively small, suggesting that everyone should be careful to minimize their risk of transmitting the disease, even if not symptomatic. With respect to coughing, the impact of coughing, though significant, decreased for later VoCs. This could be due to a combination of a higher adherence to mask wearing within the population during these periods, and of changes in infection pathway. Indeed, since omicron infects mostly upper respiratory tract \(^{52}\) and produces a higher viral load \(^{53}\) , the higher quantity of virus expelled when naturally breathing or sneezing could explain the lower effect of coughing for this particular VoC.  
+
+As shown in previous studies \(^{28}\) , we also found that contact children had a lower SAR (both as index and contact) than adults. It has been postulated that difference in contact type, quantity of virus expelled, decreased receptor expression in the respiratory tract or age- related increase in innate immune response in children could explain this difference \(^{54 - 56}\) , but the tendency of children to be more asymptomatic \(^{57}\) could also play a role, as they tend to be less tested. However, this difference with adults decreased with delta and omicron variants compared with the other VoCs. This change is potentially due to both the preference of the new variants for this more immune and unvaccinated population \(^{58}\) and to a potential detection bias (children tended to be less tested at the beginning of the pandemic).  
+
+Adults older than 65 years had a slightly higher adjusted SAR during the early waves, as reported elsewhere \(^{54}\) . This effect disappeared later in time, probably due to multiple factors, such the implementation of physical distancing and protection, but also detection bias. The above- mentioned underrepresentation of this population in this study could also bias this result.  
+
+Interestingly, we found no association between living or personal socio- economic circumstances (SEC) and SAR. This result is in line with what was reported by a recent seroprevalence study in Geneva \(^{59}\) . It has to be noted that our study does not concern the first wave of SARS- CoV- 2 pandemic, and the association between COVID- 19 variables and socio- economic condition vary greatly among waves \(^{36}\) . Disparities across the social ladder of the society concerning COVID- 19 have been shown to
+
+<--- Page Split --->
+
+concern mainly the access to test \(^{36,60,61}\) , and the COVID- 19 mortality and morbidity \(^{62,63}\) . Even if dependence of SAR on some socio- economic variables have been shown in small samples in some countries \(^{64,65}\) , it may be dependent on a particular situation or time.  
+
+The associations between all variables and SAR were mostly similar between men and women, in agreement with seroprevalence studies in Switzerland \(^{58,66,67}\) and other SAR studies \(^{28}\) , which have shown that gender or sex affects access to healthcare, morbidity and mortality, but not the contagiousness of SARS- CoV2. Nevertheless, previous infections of women index were associated with lower SAR for the EU1 and alpha variants.  
+
+The strength of this study is based on the operational database gathering all SARS- CoV- 2 tests performed by a large population of indexes and their contacts, covering 2 years of pandemics and multiple variants of concern. Detailed information on cases and contacts were available, allowing adjustment for a wide range of covariates. In particular, the availability of vaccination status, for both index and contact, adds to the strength of the study. Finally, the canton of Geneva invested a lot of effort to reach vulnerable populations, including non- documented migrants, thus reducing potential selection bias.  
+
+This study also has limitations. As previously mentioned, people over 65 years old are underrepresented, while young people are overrepresented. Underrepresentation of old people may be due to the handling of contact tracing and isolation by their specific nursing home or healthcare facility. As a consequence, vaccinated people were also underrepresented in our cohort. This could potentially lead to selection bias, although we tried to adjust for most of the main factors potentially influencing the SAR.  
+
+The main limitation of this observational registry study is information and surveillance bias \(^{68}\) . Attack rate estimation depends on the tests being performed, since contacts will be considered positive only if they were tested. Indeed, we found that the tendency to get tested (number of tests in the 90 days before the date of contact) strongly influenced SAR, with people not testing in the preceding months having a much lower SAR. This propensity of the population to be tested varies over time and depends on the health policies implemented. This is especially the case for children, for whom the testing policies varied from almost no tests during the first waves, even when they were contacts (in part due to recommendations \(^{69}\) but also because they are often not symptomatic \(^{57}\) ) to compulsory autogenic testing in schools if more than two children were infected in a classroom by the end of 2021.  
+
+There are several potential effects of this bias. First, people from low socio- economic conditions may have avoided testing in order to escape quarantine \(^{36}\) . Although we adjusted for the propensity of the contact to test, we cannot exclude residual confounding partly explaining the absence of association between SEC and SAR. Similarly, previously infected people and vaccinated people tend to test less than non- vaccinated and non- infected persons, since they did not need a test result for their sanitary pass. The Swiss sanitary pass was introduced the \(26^{\text{th}}\) of June 2021, its rules were progressively toughened over time and, in December 2021, allowed only vaccinated or previously infected patients to use common social venues. Non- vaccinated and non- infected persons were thus more incited to test for SARS- CoV- 2 than infected or vaccinated persons. Although the adjustment for the propensity to test and for its interaction with the immune status confirmed and even strengthened the effect of immune status on SAR, we cannot completely rule out residual bias, inherent to any retrospective study.  
+
+Finally, our study did not assess variant by genotype results based on a PCR test, but was based on period of time of variant dominance. Due to an overlap between every variant change, this could
+
+<--- Page Split --->
+
+alter our result, but probably in a minimal way since variants became dominant quite quickly after they emerged.  
+
+We are now 3 years into this pandemic and the virus shows no sign of receding. Our study shows that mRNA vaccination alone, although effective for reducing severe outcomes or hospitalisations \(^{13,45}\) , had a limited effect but is not sufficient anymore to contains or moderate Sars- CoV- 2 propagation. Infections have important reduction effect on the virus transmission but they are accompanied, at a population level, by cumulative effects of Sars- CoV- 2 infections \(^{70}\) provoking potential immunity deficiency \(^{71}\) , long lasting symptoms \(^{72}\) , including cardiac \(^{73}\) and neurological \(^{74}\) damages. These health consequences concerning an increasing part of the population \(^{75}\) and the weakening of the health system due to overcrowding of hospitals and exhaustion of health personnel rule out the possibility of public health policies relying solely on natural infections. Public health policies would, on the contrary, focus on reducing the number of infections for all persons, vaccinated or not, with effective and socially acceptable non pharmaceutical interventions such as air purification \(^{76,77}\) , ventilation \(^{78,79}\) , or mask wearing \(^{80}\) . Indeed, since air purification and ventilation require no individual effort, they can be implemented everywhere, thus avoiding individual behavioural barriers. Finally, to be able to study the evolution of the SARS- CoV- 2 among the population, it is of prime importance to continue to monitor the infections in the community and the general population \(^{81}\) .
+
+<--- Page Split --->
+
+## References  
+
+1. Coronavirus Disease (COVID-19) Situation Reports. https://www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports.  
+2. Wang, H. et al. Estimating excess mortality due to the COVID-19 pandemic: a systematic analysis of COVID-19-related mortality, 2020–21. The Lancet 399, 1513–1536 (2022).  
+3. Shiels, M. S., Haque, A. T., Berrington de González, A. & Freedman, N. D. Leading Causes of Death in the US During the COVID-19 Pandemic, March 2020 to October 2021. JAMA Internal Medicine (2022) doi:10.1001/jamainternmed.2022.2476.  
+4. Chen, C. et al. Global Prevalence of Post COVID-19 Condition or Long COVID: A Meta-Analysis and Systematic Review. J Infect Dis jiac136 (2022) doi:10.1093/infdis/jiac136.  
+5. Smyth, D. S. et al. Tracking cryptic SARS-CoV-2 lineages detected in NYC wastewater. Nat Commun 13, 635 (2022).  
+6. Amoutzias, G. D. et al. The Remarkable Evolutionary Plasticity of Coronaviruses by Mutation and Recombination: Insights for the COVID-19 Pandemic and the Future Evolutionary Paths of SARS-CoV-2. Viruses 14, 78 (2022).  
+7. Hall, V. et al. Protection against SARS-CoV-2 after Covid-19 Vaccination and Previous Infection. N Engl J Med 386, 1207–1220 (2022).  
+8. Wang, P. et al. Antibody resistance of SARS-CoV-2 variants B.1.351 and B.1.1.7. Nature 593, 130–135 (2021).  
+9. Polack, F. P. et al. Safety and Efficacy of the BNT162b2 mRNA Covid-19 Vaccine. N Engl J Med 383, 2603–2615 (2020).  
+10. Feikin, D. R. et al. Duration of effectiveness of vaccines against SARS-CoV-2 infection and COVID-19 disease: results of a systematic review and meta-regression. The Lancet 399, 924–944 (2022).  
+11. Rosenberg, E. S. et al. Covid-19 Vaccine Effectiveness in New York State. N Engl J Med 386, 116–127 (2022).  
+12. Levin, E. G. et al. Waning Immune Humoral Response to BNT162b2 Covid-19 Vaccine over 6 Months. New England Journal of Medicine 385, e84 (2021).  
+13. Buchan, S. A. et al. Estimated Effectiveness of COVID-19 Vaccines Against Omicron or Delta Symptomatic Infection and Severe Outcomes. JAMA Network Open 5, e2232760 (2022).  
+14. Nordström, P., Ballin, M. & Nordström, A. Effectiveness of a fourth dose of mRNA COVID-19 vaccine against all-cause mortality in long-term care facility residents and in the oldest old: A nationwide, retrospective cohort study in Sweden. The Lancet Regional Health – Europe 21, (2022).  
+15. Patalon, T. et al. Waning effectiveness of the third dose of the BNT162b2 mRNA COVID-19 vaccine. Nat Commun 13, 3203 (2022).  
+16. Altarawneh, H. N. et al. Effects of Previous Infection and Vaccination on Symptomatic Omicron Infections. N Engl J Med 387, 21–34 (2022).  
+17. Gazit, S. et al. Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) Naturally Acquired Immunity versus Vaccine-induced Immunity, Reinfections versus Breakthrough Infections: A Retrospective Cohort Study. Clinical Infectious Diseases 75, e545–e551 (2022).  
+18. Goldberg, Y. et al. Protection and Waning of Natural and Hybrid Immunity to SARS-CoV-2. New England Journal of Medicine 386, 2201–2212 (2022).
+
+<--- Page Split --->
+
+19. Pilz, S., Theiler-Schwetz, V., Trummer, C., Krause, R. & Ioannidis, J. P. A. SARS-CoV-2 reinfections: Overview of efficacy and duration of natural and hybrid immunity. Environmental Research 209, 112911 (2022).  
+
+20. Madewell, Z. J., Yang, Y., Longini, I. M., Jr, Halloran, M. E. & Dean, N. E. Household Secondary Attack Rates of SARS-CoV-2 by Variant and Vaccination Status: An Updated Systematic Review and Meta-analysis. JAMA Network Open 5, e229317 (2022).  
+
+21. Woodbridge, Y., Amit, S., Huppert, A. & Kopelman, N. M. Viral load dynamics of SARS-CoV-2 Delta and Omicron variants following multiple vaccine doses and previous infection. Nat Commun 13, 6706 (2022).  
+
+22. Puhach, O. et al. Infectious viral load in unvaccinated and vaccinated individuals infected with ancestral, Delta or Omicron SARS-CoV-2. Nat Med 28, 1491-1500 (2022).  
+
+23. Karumanagoundar, K. et al. Secondary attack rate of COVID-19 among contacts and risk factors, Tamil Nadu, March-May 2020: a retrospective cohort study. BMJ Open 11, e051491 (2021).  
+
+24. Ng, O. T. et al. SARS-CoV-2 seroprevalence and transmission risk factors among high-risk close contacts: a retrospective cohort study. Lancet Infect Dis 21, 333-343 (2021).  
+
+25. Tian, T. & Huo, X. Secondary attack rates of COVID-19 in diverse contact settings, a meta-analysis. J Infect Dev Crites 14, 1361-1367 (2020).  
+
+26. Koh, W. C. et al. What do we know about SARS-CoV-2 transmission? A systematic review and meta-analysis of the secondary attack rate and associated risk factors. PLoS One 15, e0240205 (2020).  
+
+27. Madewell, Z. J., Yang, Y., Longini, I. M., Halloran, M. E. & Dean, N. E. Household Transmission of SARS-CoV-2: A Systematic Review and Meta-analysis. JAMA Netw Open 3, e2031756 (2020).  
+
+28. Madewell, Z. J., Yang, Y., Longini, I. M., Halloran, M. E. & Dean, N. E. Factors Associated With Household Transmission of SARS-CoV-2: An Updated Systematic Review and Meta-analysis. JAMA Netw Open e2122240-e2122240 (2021).  
+
+29. Ng, D. C.-E. et al. Risk factors associated with household transmission of SARS-CoV-2 in Negeri Sembilan, Malaysia. J Paediatr Child Health (2021) doi:10.1111/jpc.15821.  
+
+30. Elgersma, I. H., Telle, K., Elstrom, P., Skagseth, H. & Kacelnik, O. Transmission of severe acute respiratory coronavirus virus 2 (SARS-CoV-2) between hospital workers and members of their household: Nationwide, registry-based, cohort study from Norway. Infect Control Hosp Epidemiol 1-6 (2022).  
+
+31. Lyngse, F. P. et al. Nationwide study on SARS-CoV-2 transmission within households from lockdown to reopening, Denmark, 27 February 2020 to 1 August 2020. Euro surveill (2022).  
+
+32. Genecand, C. et al. Cohort profile: Actionable Register of Geneva Outpatients and inpatients with SARS-CoV-2 (ARGOS). BMJ Open 11, e048946 (2021).  
+
+33. Bilan et état de la population du canton de Genève en 2021. 3.  
+
+34. FOX, J. P. FAMILY-BASED EPIDEMIOLOGIC STUDIES: THE SECOND WADE HAMPTON FROST LECTURE12. American Journal of Epidemiology 99, 165-179 (1974).  
+
+35. Liu, Y., Eggo, R. M. & Kucharski, A. J. Secondary attack rate and superspreading events for SARS-CoV-2. Lancet 395, e47 (2020).  
+
+36. Mongin, D. et al. Neighbourhood socio-economic vulnerability and access to COVID-19 healthcare during the first two waves of the pandemic in Geneva, Switzerland: A gender perspective. eClinicalMedicine 46, (2022).  
+
+37. Cole, T. J. & Lobstein, T. Extended international (IOTF) body mass index cut-offs for thinness, overweight and obesity. Pediatric Obesity 7, 284-294 (2012).
+
+<--- Page Split --->
+
+38. Hodcroft, E. B. CoVariants: SARS-CoV-2 Mutations and Variants of Interest. (2021).  
+
+39. Cheung, Y. B. A modified least-squares regression approach to the estimation of risk difference. Am J Epidemiol 166, 1337-1344 (2007).  
+
+40. R Core Team. R: A Language and Environment for Statistical Computing. (R Foundation for Statistical Computing, 2019).  
+
+41. Hojsgaard, S., Halekoh, U. & Yan, J. The R Package geepack for Generalized Estimating Equations. Journal of Statistical Software 15, 1-11 (2005).  
+
+42. Buuren, S. van & Groothuis-Oudshoorn, K. mice: Multivariate Imputation by Chained Equations in R. Journal of Statistical Software 45, 1-67 (2011).  
+
+43. Ng, O. T. et al. Impact of Delta Variant and Vaccination on SARS-CoV-2 Secondary Attack Rate Among Household Close Contacts. The Lancet Regional Health - Western Pacific 17, (2021).  
+
+44. Thorne, L. G. et al. Evolution of enhanced innate immune evasion by SARS-CoV-2. Nature 602, 487-495 (2022).  
+
+45. Stowe, J., Andrews, N., Kirsebom, F., Ramsay, M. & Bernal, J. L. Effectiveness of COVID-19 vaccines against Omicron and Delta hospitalisation, a test negative case-control study. Nat Commun 13, 5736 (2022).  
+
+46. Willett, B. J. et al. SARS-CoV-2 Omicron is an immune escape variant with an altered cell entry pathway. Nat Microbiol 7, 1161-1179 (2022).  
+
+47. Andersson, O., Campos-Mercade, P., Meier, A. N. & Wengström, E. Anticipation of COVID-19 vaccines reduces willingness to socially distance. Journal of Health Economics 80, 102530 (2021).  
+
+48. Israel, A. et al. Elapsed time since BNT162b2 vaccine and risk of SARS-CoV-2 infection: test negative design study. BMJ 375, e067873 (2021).  
+
+49. Jalali, N. et al. Increased household transmission and immune escape of the SARS-CoV-2 Omicron compared to Delta variants. Nat Commun 13, 5706 (2022).  
+
+50. Pérez-Alós, L. et al. Modeling of waning immunity after SARS-CoV-2 vaccination and influencing factors. Nat Commun 13, 1614 (2022).  
+
+51. Abu-Raddad, L. J. et al. Association of Prior SARS-CoV-2 Infection With Risk of Breakthrough Infection Following mRNA Vaccination in Qatar. JAMA 326, 1930-1939 (2021).  
+
+52. Meng, B. et al. Altered TMPRSS2 usage by SARS-CoV-2 Omicron impacts infectivity and fusogenicity. Nature 603, 706-714 (2022).  
+
+53. Riediker, M. et al. Higher viral load and infectivity increase risk of aerosol transmission for Delta and Omicron variants of SARS-CoV-2. Swiss Medical Weekly (2022) doi:10.4414/smw.2022.w30133.  
+
+54. Stich, M. et al. Transmission of Severe Acute Respiratory Syndrome Coronavirus 2 in Households with Children, Southwest Germany, May-August 2020. Emerg Infect Dis 27, 3009-3019 (2021).  
+
+55. Reukers, D. F. M. et al. High Infection Secondary Attack Rates of Severe Acute Respiratory Syndrome Coronavirus 2 in Dutch Households Revealed by Dense Sampling. Clinical Infectious Diseases 74, 52-58 (2022).  
+
+56. Bunyavanich, S., Do, A. & Vicencio, A. Nasal Gene Expression of Angiotensin-Converting Enzyme 2 in Children and Adults. JAMA 323, 2427-2429 (2020).  
+
+57. Molteni, E. et al. Illness duration and symptom profile in symptomatic UK school-aged children tested for SARS-CoV-2. The Lancet Child & Adolescent Health 5, 708-718 (2021).  
+
+58. Zaballa, M.-E. et al. Seroprevalence of anti-SARS-CoV-2 antibodies and cross-variant neutralization capacity after the Omicron BA.2 wave in Geneva, Switzerland: a population-based study. Lancet Reg Health Eur 24, 100547 (2023).
+
+<--- Page Split --->
+
+59. Santa-Ramírez, H.-A. et al. Socio-economic determinants of SARS-CoV-2 infection: Results from a population-based cross-sectional serosurvey in Geneva, Switzerland. Frontiers in Public Health 10, (2022).  
+
+60. Riou, J. et al. Socioeconomic position and the COVID-19 care cascade from testing to mortality in Switzerland: a population-based analysis. The Lancet Public Health 0, (2021). 
+61. Vandentorren, S. et al. The effect of social deprivation on the dynamic of SARS-CoV-2 infection in France: a population-based analysis. The Lancet Public Health 0, (2022). 
+62. Williamson, E. J. et al. Factors associated with COVID-19-related death using OpenSAFELY. Nature 584, 430-436 (2020). 
+63. Bilal, U., Tabb, L. P., Barber, S. & Diez Roux, A. V. Spatial Inequities in COVID-19 Testing, Positivity, Confirmed Cases, and Mortality in 3 U.S. Cities. Ann Intern Med (2021) doi:10.7326/M20-3936. 
+64. Righolt, C. H., Zhang, G., Sever, E., Wilkinson, K. & Mahmud, S. M. Patterns and descriptors of COVID-19 testing and lab-confirmed COVID-19 incidence in Manitoba, Canada, March 2020-May 2021: A population-based study. The Lancet Regional Health - Americas 2, 100038 (2021). 
+65. Gentile, I. et al. COVID-19 Post-Exposure Evaluation (COPE) Study: Assessing the Role of Socio-Economic Factors in Household SARS-CoV-2 Transmission within Campania Region (Southern Italy). International Journal of Environmental Research and Public Health 19, 10262 (2022). 
+66. Amati, R. et al. Functional immunity against SARS-CoV-2 in the general population after a booster campaign and the Delta and Omicron waves, Switzerland, March 2022. Eurosurveillance 27, 2200561 (2022). 
+67. Stringhini, S. et al. Seroprevalence of anti-SARS-CoV-2 antibodies after the second pandemic peak. The Lancet Infectious Diseases 21, 600-601 (2021). 
+68. Tancredi, S., Anker, D., Rosella, L. & Chiolero, A. Elimination of covid-19: beware of surveillance bias. BMJ 374, n2126 (2021). 
+69. Testing strategy for children (0-12 years of age) and quarantining strategy of children and adolescents in Switzerland - Swiss National COVID-19 Science Task Force. https://sciencetaskforce.ch/en/policy-brief/testing-strategy-for-children-0-12-years-of-age-and-quarantining-strategy-of-children-and-adolescents-in-switzerland/. 
+70. Bowe, B., Xie, Y. & Al-Aly, Z. Acute and postacute sequelae associated with SARS-CoV-2 reinfection. Nat Med 28, 2398-2405 (2022). 
+71. Phetsuphanh, C. et al. Immunological dysfunction persists for 8 months following initial mild-to-moderate SARS-CoV-2 infection. Nat Immunol 23, 210-216 (2022). 
+72. Tran, V.-T., Porcher, R., Pane, I. & Ravaud, P. Course of post COVID-19 disease symptoms over time in the ComPaRe long COVID prospective e-cohort. Nat Commun 13, 1812 (2022). 
+73. Xie, Y., Xu, E., Bowe, B. & Al-Aly, Z. Long-term cardiovascular outcomes of COVID-19. Nat Med 28, 583-590 (2022). 
+74. Taquet, M. et al. Neurological and psychiatric risk trajectories after SARS-CoV-2 infection: an analysis of 2-year retrospective cohort studies including 1 284 437 patients. The Lancet Psychiatry 9, 815-827 (2022). 
+75. Global Burden of Disease Long COVID Collaborators. Estimated Global Proportions of Individuals With Persistent Fatigue, Cognitive, and Respiratory Symptom Clusters Following Symptomatic COVID-19 in 2020 and 2021. JAMA 328, 1604-1615 (2022).
+
+<--- Page Split --->
+
+76. Thuresson, S. et al. Airborne Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) in Hospitals: Effects of Aerosol-Generating Procedures, HEPA-Filtration Units, Patient Viral Load, and Physical Distance. Clin. Infect. Dis. 75, E89-E96 (2022).  
+
+77. Conway Morris, A. et al. The Removal of Airborne Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) and Other Microbial Bioaerosols by Air Filtration on Coronavirus Disease 2019 (COVID-19) Surge Units. Clinical Infectious Diseases 75, e97-e101 (2022).  
+
+78. Stadnytskyi, V., Anfinrud, P. & Bax, A. Breathing, speaking, coughing or sneezing: What drives transmission of SARS-CoV-2? J Intern Med 290, 1010-1027 (2021).  
+
+79. Zhang, R., Li, Y., Zhang, A. L., Wang, Y. & Molina, M. J. Identifying airborne transmission as the dominant route for the spread of COVID-19. Proceedings of the National Academy of Sciences 117, 14857-14863 (2020).  
+
+80. Cowger, T. L. et al. Lifting Universal Masking in Schools — Covid-19 Incidence among Students and Staff. New England Journal of Medicine 387, 1935-1946 (2022).  
+
+81. Chiolero, A., Tancredi, S. & Cullati, S. Covid-19: Surveillance systems for the new normal. BMJ 376, o710 (2022).
+
+<--- Page Split --->
+
+# Figures and tables
+
+Table1: socio-demographics characteristics of the index cases and declared contacts for the whole study period (Overall) and stratified per periods of variant predominant.
+
+<table><tr><td></td><td>Overall</td><td>EU1</td><td>alpha</td><td>delta</td><td>Omicron</td><td>Missing</td></tr><tr><td>Number of index cases</td><td>50889</td><td>18884</td><td>6692</td><td>11628</td><td>13685</td><td></td></tr><tr><td>Mean number of contact (SD) per index case</td><td>2.19 (1.90)</td><td>2.42 (2.30)</td><td>2.50 (1.83)</td><td>2.38 (1.87)</td><td>1.56 (0.93)</td><td></td></tr><tr><td>Total number of contacts</td><td>111432</td><td>45755</td><td>16755</td><td>27636</td><td>21286</td><td></td></tr><tr><td>Number of infected contacts within 10 days following<br>contact (%)</td><td>21387 (19.2)</td><td>7588 (16.6)</td><td>3546 (21.2)</td><td>4652 (16.8)</td><td>5601 (26.3)</td><td>0.0</td></tr><tr><td>Index Socio-economic-<br>neighbourhood</td><td></td><td></td><td></td><td></td><td></td><td>3.3</td></tr><tr><td>Healthy</td><td>39578 (36.4)</td><td>16066 (35.7)</td><td>5689 (34.3)</td><td>10715 (40.0)</td><td>7108 (34.7)</td><td></td></tr><tr><td>Slightly vulnerable</td><td>18127 (16.7)</td><td>7141 (15.9)</td><td>3025 (18.2)</td><td>4305 (16.1)</td><td>3656 (17.8)</td><td></td></tr><tr><td>Moderately vulnerable</td><td>19496 (17.9)</td><td>8200 (18.2)</td><td>3078 (18.6)</td><td>4513 (16.9)</td><td>3705 (18.1)</td><td></td></tr><tr><td>Highly vulnerable</td><td>31612 (29.1)</td><td>13582 (30.2)</td><td>4784 (28.9)</td><td>7223 (27.0)</td><td>6023 (29.4)</td><td></td></tr><tr><td>Index vulnerable person<br>(%)</td><td>8530(7.8)</td><td>3507(7.8)</td><td>1849 (11.1)</td><td>2149(8.0)</td><td>1025(5.0)</td><td>2.2</td></tr><tr><td>Index with symptoms (%)</td><td>98064 (94.1)</td><td>40877 (95.7)</td><td>15204 (91.4)</td><td>24538 (92.5)</td><td>17445 (95.5)</td><td>6.7</td></tr><tr><td>Index with cough (%)</td><td>60148 (57.7)</td><td>23657 (55.4)</td><td>9968 (59.9)</td><td>15124 (57.0)</td><td>11399 (62.4)</td><td>6.7</td></tr><tr><td>Index obesity (%)</td><td>9346 (10.5)</td><td>3892 (11.5)</td><td>1841 (11.6)</td><td>2052(8.3)</td><td>1561 (10.3)</td><td>20.1</td></tr><tr><td>Index women (%)</td><td>59604 (53.6)</td><td>24421 (53.4)</td><td>8973 (53.7)</td><td>14524 (52.8)</td><td>11686 (55.1)</td><td>0.2</td></tr><tr><td>Index age category (%)</td><td></td><td></td><td></td><td></td><td></td><td>0.0</td></tr><tr><td>18-65</td><td>81920 (73.5)</td><td>38236 (83.6)</td><td>12035 (71.8)</td><td>16101 (58.3)</td><td>15548 (73.1)</td><td></td></tr><tr><td>0-18</td><td>24358 (21.9)</td><td>4862 (10.6)</td><td>3980 (23.8)</td><td>10360 (37.5)</td><td>5156 (24.2)</td><td></td></tr><tr><td>65+</td><td>5135(4.6)</td><td>2656(5.8)</td><td>740(4.4)</td><td>1169(4.2)</td><td>570(2.7)</td><td></td></tr><tr><td>Index immune status (%)</td><td></td><td></td><td></td><td></td><td></td><td>0.0</td></tr><tr><td>Infected&lt;6 months</td><td>625(0.6)</td><td>262(0.6)</td><td>148(0.9)</td><td>44(0.2)</td><td>171(0.8)</td><td></td></tr><tr><td>Infected&gt;6 months</td><td>1286(1.2)</td><td>68(0.1)</td><td>58(0.3)</td><td>357(1.3)</td><td>803(3.8)</td><td></td></tr><tr><td>hybrid</td><td>1269(1.1)</td><td>0(0.0)</td><td>10(0.1)</td><td>148(0.5)</td><td>1111(5.2)</td><td></td></tr><tr><td>Non-vaccinated non<br>infected (NVNI)</td><td>89927(80.7)</td><td>45416(99.3)</td><td>16179(96.6)</td><td>19194(69.5)</td><td>9138(42.9)</td><td></td></tr><tr><td>Vaccinated&lt;6 months</td><td>11853(10.6)</td><td>9(0.0)</td><td>360(2.1)</td><td>6067(22.0)</td><td>5417(25.4)</td><td></td></tr><tr><td>Vaccinated&gt;6 months</td><td>6471(5.8)</td><td>0(0.0)</td><td>0(0.0)</td><td>1825(6.6)</td><td>4646(21.8)</td><td></td></tr><tr><td>Index housing type (%)</td><td></td><td></td><td></td><td></td><td></td><td>2.2</td></tr><tr><td>Building</td><td>88528(80.6)</td><td>36430(80.3)</td><td>13487(80.9)</td><td>21399(79.0)</td><td>17212(83.1)</td><td></td></tr><tr><td>Single house</td><td>18271(16.6)</td><td>7438(16.4)</td><td>2771(16.6)</td><td>5094(18.8)</td><td>2968(14.3)</td><td></td></tr><tr><td>Collective structure</td><td>3063(2.8)</td><td>1516(3.3)</td><td>420(2.5)</td><td>594(2.2)</td><td>533(2.6)</td><td></td></tr><tr><td>contact_type(%)</td><td></td><td></td><td></td><td></td><td></td><td>38.2</td></tr><tr><td>Same roof</td><td>43808(62.5)</td><td>16201(56.1)</td><td>10228(65.5)</td><td>13374(65.5)</td><td>4005(77.3)</td><td></td></tr><tr><td>Intimate or familial</td><td>19373(27.6)</td><td>8899(30.8)</td><td>4168(26.7)</td><td>5444(26.7)</td><td>862(16.6)</td><td></td></tr><tr><td>During the day</td><td>6908(9.9)</td><td>3777(13.1)</td><td>1211(7.8)</td><td>1606(7.9)</td><td>314(6.1)</td><td></td></tr><tr><td>Contact female (%)</td><td>56271(51.7)</td><td>22846(51.2)</td><td>8567(51.8)</td><td>14141(52.7)</td><td>10717(51.4)</td><td>2.3</td></tr><tr><td>Contact immune status (%)</td><td></td><td></td><td></td><td></td><td></td><td>9.5</td></tr></table>
+
+<--- Page Split --->
+
+
+<table><tr><td>Infected &lt; 6 months</td><td>5216(5.2)</td><td>1818(4.3)</td><td>871(5.6)</td><td>1394(6.1)</td><td>1133(5.5)</td><td></td></tr><tr><td>Infected &gt; 6 months</td><td>1727(1.7)</td><td>199(0.5)</td><td>185(1.2)</td><td>592(2.6)</td><td>751(3.7)</td><td></td></tr><tr><td>hybrid</td><td>3114(3.1)</td><td>0(0.0)</td><td>53(0.3)</td><td>1212(5.3)</td><td>1849(9.0)</td><td></td></tr><tr><td>Non-vaccinated non infected (NVNI)</td><td>76800(76.0)</td><td>40103(95.2)</td><td>13657(87.2)</td><td>12130(53.3)</td><td>10910(53.3)</td><td></td></tr><tr><td>Vaccinated &lt; 6 months</td><td>10271(10.2)</td><td>13(0.0)</td><td>891(5.7)</td><td>6360(28.0)</td><td>3007(14.7)</td><td></td></tr><tr><td>Vaccinated &gt; 6 months</td><td>3879(3.8)</td><td>0(0.0)</td><td>0(0.0)</td><td>1066(4.7)</td><td>2813(13.7)</td><td></td></tr><tr><td>Contact age category (%)</td><td></td><td></td><td></td><td></td><td></td><td>9.9</td></tr><tr><td>18-65</td><td>63497(63.2)</td><td>29161(69.3)</td><td>9266(59.2)</td><td>13334(58.7)</td><td>11736(58.6)</td><td></td></tr><tr><td>0-18</td><td>31733(31.6)</td><td>10421(24.8)</td><td>5518(35.3)</td><td>8128(35.8)</td><td>7666(38.3)</td><td></td></tr><tr><td>65+</td><td>5252(5.2)</td><td>2495(5.9)</td><td>858(5.5)</td><td>1269(5.6)</td><td>630(3.1)</td><td></td></tr><tr><td>Number of tests last 3 months (%)</td><td></td><td></td><td></td><td></td><td></td><td>0.0</td></tr><tr><td>0</td><td>80713(72.4)</td><td>36439(79.6)</td><td>11964(71.4)</td><td>19573(70.8)</td><td>12737(59.8)</td><td></td></tr><tr><td>1</td><td>21964(19.7)</td><td>7966(17.4)</td><td>3531(21.1)</td><td>5305(19.2)</td><td>5162(24.3)</td><td></td></tr><tr><td>2+</td><td>8755(7.9)</td><td>1350(3.0)</td><td>1260(7.5)</td><td>2758(10.0)</td><td>3387(15.9)</td><td></td></tr></table>
+
+<--- Page Split --->
+
+Figure 1: Adjusted Secondary Attack Rate (SAR) stratified per variant (EU1, alpha, delta and omicron), with the reference value indicated with a vertical line (reference of each covariate is indicated in bold in parenthesis). This multivariate analysis considers the effect of the index case gender, age, obesity, presence of symptoms, presence of cough, immunity status, neighbourhood socio-economic condition, vulnerability and type of living; the link between the index case and its contacts, and for the contact persons, their gender, age, number of tests performed the three months before the contact date with the index case, and their immunity status. The reference index case - contact relation of this multivariate analysis is the contact between two men of age below 65 living at the same place, the index being not vaccinated not infected (NVNI), not obese, living in a wealthy neighbourhood and being not a vulnerable person, living in a housing building, and the contact person being a NVNI adult men who performed one SARS- CoV- 2 test during the last 3 month preceding the contact.  
+
+![PLACEHOLDER_19_0]
+
+
+<--- Page Split --->
+
+Table 2: Coefficients of the multivariate generalized estimating equation [Confidence Interval], providing the additional effect on the reference secondary attack rate (first line), for the 4 periods of dominance of the variants EU1, alpha, delta, and omicron. The reference category for each categorical variable is indicated in bold in parenthesis. The left column indicates is the variable concern the index case, the contact, or their relation. p values are indicated with *. *: 0.01<p<0.05, **:0.001<p<0.01, ***: p<0.001 
+
+<table><tr><td colspan="2"></td><td>EU1</td><td>alpha</td><td>delta</td><td>omicron</td></tr><tr><td rowspan="3">Index</td><td>Reference</td><td>33.3*** [30.9,35.7]</td><td>31.0*** [27.8,34.1]</td><td>34.5*** [32.0,37.0]</td><td>41.0*** [37.4,44.6]</td></tr><tr><td>Immunity: previously infected (NVNI)</td><td>-9.9*** [-13.3,-6.6]</td><td>-8.6*** [-13.1,-4.1]</td><td>-10.9*** [-13.9,-7.9]</td><td>-4.3** [-7.3,-1.3]</td></tr><tr><td>Immunity: vaccinated &lt; 6 months (NVNI)</td><td></td><td>-5.0* [-9.5,-0.4]</td><td>-5.6*** [-6.9,-4.3]</td><td>-6.6*** [-8.2,-4.9]</td></tr><tr><td>Index</td><td>Immunity: vaccinated &gt; 6 months (NVNI)</td><td></td><td></td><td>-1.3 [-3.5,0.9]</td><td>-2.8** [-4.6,-0.9]</td></tr><tr><td>Index</td><td>women (men)</td><td>-0.3 [-1.2,0.5]</td><td>0.1 [-1.4,1.5]</td><td>0.1 [-1.0,1.1]</td><td>-1.7* [-3.0,-0.4]</td></tr><tr><td>Index</td><td>age 0-17 (18-64)</td><td>-0.7 [-2.1,0.6]</td><td>-4.9*** [-6.6,-3.1]</td><td>-2.6*** [-3.8,-1.3]</td><td>-2.1* [-3.9,-0.4]</td></tr><tr><td>Index</td><td>age 65+ (18-64)</td><td>1.2 [-0.6,3.0]</td><td>3.8* [0.1,7.4]</td><td>1.5 [-0.9,4.0]</td><td>2.0 [-2.0,6.1]</td></tr><tr><td>Index</td><td>Obese (not obese)</td><td>1.1 [-0.4,2.5]</td><td>2.1 [-0.3,4.5]</td><td>0.5 [-1.5,2.4]</td><td>-0.8 [-3.0,1.4]</td></tr><tr><td>Index</td><td>Symptoms (no symptoms)</td><td>2.0 [-0.1,4.0]</td><td>7.6*** [5.1,10.0]</td><td>5.2*** [3.4,7.0]</td><td>4.5** [1.5,7.6]</td></tr><tr><td>Index</td><td>Cough (no cough)</td><td>5.3*** [4.4,6.2]</td><td>5.5*** [3.9,7.1]</td><td>2.2*** [1.0,3.3]</td><td>1.8* [0.4,3.3]</td></tr><tr><td>Index</td><td>neighbourhood poverty: sligh (wealthy)</td><td>-0.3 [-1.6,1.0]</td><td>0.8 [-1.4,3.0]</td><td>-1.1 [-2.6,0.5]</td><td>0.2 [-1.7,2.2]</td></tr><tr><td>Index</td><td>neighbourhood poverty: moderate (wealthy)</td><td>-0.9 [-2.1,0.4]</td><td>-0.9 [-3.1,1.4]</td><td>-0.7 [-2.3,0.8]</td><td>-0.4 [-2.4,1.5]</td></tr><tr><td>Index</td><td>neighbourhood poverty: high (wealthy)</td><td>-0.4 [-1.5,0.7]</td><td>0.0 [-1.9,2.0]</td><td>-1.3 [-2.7,0.1]</td><td>-0.5 [-2.3,1.2]</td></tr><tr><td>Index</td><td>vulnerable (not vulnerable)</td><td>-0.7 [-2.3,0.8]</td><td>-1.2 [-3.6,1.3]</td><td>0.6 [-1.5,2.6]</td><td>-2.1 [-4.9,0.8]</td></tr><tr><td>Index</td><td>living: single house (building)</td><td>0.0 [-1.2,1.3]</td><td>-1.3 [-3.4,0.8]</td><td>0.0 [-1.4,1.5]</td><td>-0.9 [-2.9,1.0]</td></tr><tr><td>Index</td><td>living: collective structure (building)</td><td>0.7 [-1.8,3.2]</td><td>-1.9 [-6.3,2.6]</td><td>-5.1** [-8.3,-1.8]</td><td>-0.1 [-4.3,4.1]</td></tr><tr><td>Index - Contact</td><td>intimate/family (housing)</td><td>-9.4*** [-10.3,-8.4]</td><td>-13.1*** [-14.7,-11.6]</td><td>-8.4*** [-9.6,-7.3]</td><td>-8.0*** [-9.6,-6.3]</td></tr><tr><td>Index - Contact</td><td>pro/school/daily (housing)</td><td>-10.7*** [-11.9,-9.5]</td><td>-15.3*** [-17.7,-13.0]</td><td>-10.8*** [-12.5,-9.2]</td><td>-11.5*** [-13.9,-9.0]</td></tr><tr><td>Contact</td><td>Immunity: previously infected &lt; 6 months (NVNI)</td><td>-13.3*** [-15.4,-11.3]</td><td>-26.5*** [-28.1,-24.8]</td><td>-30.7*** [-32.1,-29.4]</td><td>-32.0*** [-33.9,-30.0]</td></tr><tr><td>Contact</td><td>Immunity: previously infected &gt; 6 months (NVNI)</td><td>-17.3*** [-19.3,-15.4]</td><td>-20.6*** [-23.3,-18.0]</td><td>-14.9*** [-16.8,-12.9]</td><td>-4.4** [-7.6,-1.2]</td></tr><tr><td>Contact</td><td>Immunity: vaccinated &lt; 6 months (NVNI)</td><td></td><td>-13.5*** [-16.0,-11.0]</td><td>-9.5*** [-10.5,-8.4]</td><td>7.0*** [5.0,8.9]</td></tr><tr><td>Contact</td><td>Immunity: vaccinated &gt; 6 months (NVNI)</td><td></td><td></td><td>0.8 [-1.6,3.3]</td><td>13.3*** [11.2,15.4]</td></tr></table>
+
+<--- Page Split --->
+
+<table><tr><td colspan="2"></td><td>EU1</td><td>alpha</td><td>delta</td><td>omicron</td></tr><tr><td>Contact</td><td>Immunity: hybrid (NVNI)</td><td></td><td>-20.1*** [-25.8,-14.3]</td><td>-21.7*** [-22.9,-20.6]</td><td>-18.2*** [-20.4,-16.1]</td></tr><tr><td>Contact</td><td>women (men)</td><td>0.2 [-0.4,0.8]</td><td>-1.0 [-2.1,0.1]</td><td>-0.2 [-1.0,0.6]</td><td>-0.4 [-1.5,0.7]</td></tr><tr><td>Contact</td><td>age: 0-17 (18-64)</td><td>-11.7*** [-12.5,-11.0]</td><td>-8.2*** [-9.5,-6.8]</td><td>-0.2 [-1.3,0.8]</td><td>-4.8*** [-6.2,-3.4]</td></tr><tr><td>Contact</td><td>65+ (18-64)</td><td>4.1*** [2.4,5.8]</td><td>1.2 [-1.7,4.1]</td><td>-1.4 [-3.2,0.3]</td><td>-6.5*** [-9.6,-3.4]</td></tr><tr><td>Contact</td><td>N test 30days: 0 (1)</td><td>-16.3*** [-17.4,-15.1]</td><td>-11.1*** [-12.7,-9.5]</td><td>-13.3*** [-14.6,-12.0]</td><td>-17.0*** [-18.6,-15.5]</td></tr><tr><td>Contact</td><td>N test 30days: 2+ (1)</td><td>1.3 [-1.2,3.9]</td><td>1.7 [-1.0,4.4]</td><td>3.0** [1.1,4.8]</td><td>-0.1 [-2.0,1.8]</td></tr></table>
+
+<--- Page Split --->
+
+## Statements & Declarations  
+
+## Funding  
+
+This research was supported by the research project SELFISH, financed by the Swiss National Science Foundation, grant number 51NF40- 160590 (LIVES Center international research project call).  
+
+## Competing Interests  
+
+The authors have no competing interest to declare  
+
+## Acknowledgements  
+
+We thank the Geneva Directorate of Health for collecting and providing the data.  
+
+## Ethics approval  
+
+Research received the agreement of the Cantonal Ethic Committee of Geneva (CCER protocol 2020- 01273). Individuals who refused to share their data were removed from the analysis.
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+adjustedsecondaryattackratesuppmatV2. pdf
+
+<--- Page Split --->

preprint/preprint__7e6f2e24c1c8d4e3aaa13e2875ea7c7452d975c29a599bbb90931019f462e57c/preprint__7e6f2e24c1c8d4e3aaa13e2875ea7c7452d975c29a599bbb90931019f462e57c_det.mmd

@@ -0,0 +1,588 @@
+<|ref|>title<|/ref|><|det|>[[44, 107, 949, 208]]<|/det|>
+# Effect of mRNA vaccination and previous infections on SARS-CoV-2 transmission across four variants: adjusted analysis of 111'432 declared contacts  
+
+<|ref|>text<|/ref|><|det|>[[44, 229, 437, 270]]<|/det|>
+Denis Mongin ( denis.mongin@hcuge.ch) University of Geneva  
+
+<|ref|>text<|/ref|><|det|>[[44, 277, 325, 317]]<|/det|>
+Nils Burgisser University Hospitals of Geneva  
+
+<|ref|>text<|/ref|><|det|>[[44, 325, 310, 365]]<|/det|>
+Gustavo Laurie Geneva Directorate of Health  
+
+<|ref|>text<|/ref|><|det|>[[44, 371, 310, 410]]<|/det|>
+Guillaume Schimmel Geneva Directorate of Health  
+
+<|ref|>text<|/ref|><|det|>[[44, 417, 310, 456]]<|/det|>
+Diem- Lan Vu Geneva Directorate of Health  
+
+<|ref|>text<|/ref|><|det|>[[44, 463, 234, 503]]<|/det|>
+Stephane Cullati University of Geneva  
+
+<|ref|>text<|/ref|><|det|>[[44, 510, 325, 550]]<|/det|>
+Delphine Courvoisier University Hospitals of Geneva  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 590, 101, 607]]<|/det|>
+## Article  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 628, 135, 646]]<|/det|>
+## Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 666, 325, 686]]<|/det|>
+Posted Date: February 7th, 2023  
+
+<|ref|>text<|/ref|><|det|>[[44, 705, 474, 724]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 2510736/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 742, 910, 784]]<|/det|>
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 803, 530, 822]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 859, 950, 901]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on September 6th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 41109- 9.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[133, 87, 866, 241]]<|/det|>
+Effect of mRNA vaccination and previous infections on SARS-CoV-2 transmission across four variants: adjusted analysis of 111'432 declared contacts  
+
+<|ref|>text<|/ref|><|det|>[[115, 270, 870, 306]]<|/det|>
+Denis Mongin<sup>1,\*</sup>, Nils Bürgisser<sup>2</sup>, Gustavo Laurie<sup>3</sup>, Guillaume Schilmmel<sup>3</sup>, Diem- Lan Vu<sup>1,3,4,5</sup>, Stephane Cullati<sup>6,7</sup>, Delphine Sophie Courvoisier<sup>1,6</sup>, and the Covid- SMC Study Group<sup>†</sup>  
+
+<|ref|>text<|/ref|><|det|>[[115, 315, 868, 550]]<|/det|>
+<sup>1</sup>Faculty of Medicine, University of Geneva, Geneva, Switzerland  <sup>2</sup>General internal medicine division, Department of Medicine, Geneva University Hospitals, Geneva, Switzerland  <sup>3</sup>Division of General cantonal physician, Geneva Directorate of Health, Geneva, Switzerland  <sup>4</sup>Division of Infectious Diseases, Geneva University Hospitals, Geneva, Switzerland.  <sup>5</sup>Laboratory of Virology, Division of Laboratory Medicine, Geneva University Hospitals, Geneva, Switzerland.  <sup>6</sup>Division Quality of care, Faculty of Medicine, University of Geneva, Geneva, Switzerland  
+
+<|ref|>text<|/ref|><|det|>[[115, 581, 668, 599]]<|/det|>
+†Membership of the Covid- SMC Study Group is provided in the appendix.  
+
+<|ref|>text<|/ref|><|det|>[[115, 609, 460, 755]]<|/det|>
+\*Correspondance to: Denis Mongin +41 223723642 denis.mongin@unige.ch Hôpital Beau séjour, service de rhumatologie. 26 avenue de Beau Séjour 1206 Genève Switzerland
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 86, 209, 103]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[115, 134, 880, 255]]<|/det|>
+The immunity conferred by SARS- CoV- 2 vaccines and infections reduces the transmission of the virus. But it is not clear how the effect of immunity is shared between a reduction of the contagiousness and an increased protection against infection. To answer this question, we used a register of \(>50^{\prime}000\) SARS- CoV- 2 positive index cases and their \(>110^{\prime}000\) declared contacts to estimate the association between secondary attack rate and immunity status (natural infection, vaccine, or both). Analyses were stratified per four SARS- CoV- 2 variants and adjusted for index cases and contacts sociodemographic characteristics and the propensity of the contacts to be tested.  
+
+<|ref|>text<|/ref|><|det|>[[115, 255, 877, 460]]<|/det|>
+The reduction of the propagation conferred by immunity was mainly a protection of the contacts against infection rather than a diminution of the contagiousness of the index cases. The largest immunity effect was conferred for both by natural infection, especially when recent for the contacts. Although of smaller amplitude, the effect of vaccination for index, i.e. the reduction of contagiousness, was less affected by SarS- CoV- 2 variant and time since vaccination than the effect for contacts, i.e. the increase of protection against infection. Indeed, vaccination offered protection against infection only if given less than 6 months before and only for the variants preceding omicron, while the vaccination kept moderately reducing the index infectivity during the omicron wave, even when given more than 6 months ago. Hybrid immunity (vaccination and infection) did not have increased effectiveness than recent infection. These findings support the idea that vaccination also protects others, and highlight the need for the implementation of non- personal intervention reducing Sars- CoV- 2 propagation, such as ventilation or air filtration.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 479, 252, 497]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[115, 526, 883, 647]]<|/det|>
+Since its worldwide spread at the beginning of \(2020^{1}\) , the SARS- CoV- 2 virus has caused one of the most important health burden in recent history. It is estimated to have caused 18 million deaths as of end of 2021. SARS- CoV- 2 became a leading cause of death in some countries in these years and is responsible for an important burden of long lasting symptoms in the population. Its widespread circulation within human communities and possible animal reservoirs allows the SARS- CoV- 2 to mutate frequently and has resulted so far in more contagious, immunity- escaping variants 7,8 responsible for successive waves of infections worldwide.  
+
+<|ref|>text<|/ref|><|det|>[[115, 656, 877, 811]]<|/det|>
+The effect of immunity on the transmission of the successive SARS- CoV2 variants and its evolution in time are key factors for our understanding of the SARS- CoV2 propagation. Immunity can be acquired through vaccination or through natural infection. SARS- CoV2 mRNA vaccines have been shown to be effective in preventing re- infection shortly after injection. However, the immunity it confers wanes rapidly 10- 12 and a roll- out of booster vaccinations has been implemented in western countries to maintain an immunity against SARS- CoV2 13- 15. It was recently demonstrated that natural infection confers a stronger and longer lasting protection against reinfection than vaccination 7,16- 18, and that the combination of both type of immunity (hybrid immunity) may provide an even stronger protection 18,19.  
+
+<|ref|>text<|/ref|><|det|>[[115, 812, 877, 896]]<|/det|>
+Less is known, however, on the effect of immunity on the susceptibility to contaminate others, especially with regard to natural immunity. Depending on the variant of concern (VoC) considered, studies analysing the secondary attack rate show little to no effect of vaccine on being contagious, while recent in vitro studies, by measuring viral load and propagation, indirectly suggest that natural infection could better reduce the infectiousness than vaccine 21,22.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 82, 875, 220]]<|/det|>
+Secondary attack rate (SAR) is a good measure of SARS- Cov2 transmission, providing a full picture of both the protection against getting infected and the diminution of the contagiousness that the immunity may confer. Apart from the immunity of the population and the VoC considered, SAR is known to vary greatly by contact settings, ranging from \(20\%\) in households to \(6\%\) in social gatherings during the first year of the pandemic \(^{23 - 26}\) , but also by the symptoms of the index cases \(^{27,28}\) and the socio- demographic characteristics of the studied population \(^{26,28 - 31}\) . By definition, SAR also depends directly on the capacity to detect SARS- CoV- 2 infections among contacts, which includes the propensity of the contacts to get tested.  
+
+<|ref|>text<|/ref|><|det|>[[116, 230, 861, 315]]<|/det|>
+Using a register dataset of \(50'889\) index cases having declared 111'432 contacts in the State of Geneva \(^{32}\) , we propose to study the effect of the immune status on SARS- CoV2 transmission, considering vaccination and natural infection of index and contacts and main SARS- CoV- 2 variants while adjusting for demographic, social and health factors as well as the tendency to get tested for SARS- CoV- 2.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 333, 216, 352]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 360, 281, 376]]<|/det|>
+## Setting and period  
+
+<|ref|>text<|/ref|><|det|>[[116, 378, 872, 465]]<|/det|>
+Data used for the present study consisted in a register dataset of links between an infected case (hereafter the index case) and a declared person with whom he/she had close contact during the 10 days preceding his/her test result (hereafter the contacts). These data stem from the ARGOS database \(^{32}\) , which is an ongoing operational COVID- 19 database created by the Geneva health state agency (Geneva Directorate of Health).  
+
+<|ref|>text<|/ref|><|det|>[[116, 473, 875, 576]]<|/det|>
+Geneva is a state of 511'921 inhabitants as of the last census in December 2021 \(^{33}\) , mainly urban, with a high population density, and which doubles its population on working days (excluding pandemic restrictions) as a result of national and international commuter traffic (mainly from neighbouring France). We used data from the \(26^{\text{th}}\) February of 2020 (first positive tested recorded in Geneva) to \(28^{\text{th}}\) February of 2022. Data was not collected after March \(1^{\text{st}}\) 2022 because contact declaration was stopped at this date in Switzerland.  
+
+<|ref|>text<|/ref|><|det|>[[116, 586, 878, 653]]<|/det|>
+This research received the agreement of the Cantonal Ethic Committee of Geneva (CCER protocol 2020- 01273). Participants had the possibility to refuse sharing their data for research through a form that was automatically sent. Those who did were removed from the analysis. Data are available upon request at https://edc.hcuge.ch/surveys/?s=TLT9EHE93C.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 667, 336, 682]]<|/det|>
+## Index cases and contacts  
+
+<|ref|>text<|/ref|><|det|>[[116, 685, 866, 718]]<|/det|>
+The register contains baseline, follow- up, and contact information of all SARS- CoV- 2 positive tested persons (index case) residing in the State of Geneva, Switzerland.  
+
+<|ref|>text<|/ref|><|det|>[[116, 728, 880, 866]]<|/det|>
+A contact was considered as infected by the index case if they had a positive COVID- 19 result within 10 days following their last contact with the index case. From February 2020 to end of April 2020, contact information was collected by interviewing the index case. From May 2020, index cases had the possibility to provide their contacts names and phone through an online form. Contacts were then approached using phone interviews. Additionally, an online form was implemented at the end of September 2020 to support the oral interviews, where the contacts had the ability to complete the required information themselves. From mid December 2021, the oral interviews could not be maintained, therefore contact information was only gathered from the online formula.  
+
+<|ref|>text<|/ref|><|det|>[[116, 876, 868, 910]]<|/det|>
+Contact information contained the type of contact setting between the index case and the declared contacts (see supplementary material), the date of the last contact between index and contact, the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 877, 202]]<|/det|>
+birth date, gender, date of subsequent or anterior positive PCR or antigenic test results, as well as the living address and the vaccination dates. Information about the index cases included date of SARS- CoV2 test result, gender, date of birth, living address, presence or absence of any symptoms (see supplementary material), presence or absence of cough (cough, dry cough or wet cough), personal vulnerability (based on if the person reported difficulty to make ends meet, lived in a highly subsidized housing, or if they asked explicitly to avoid police control, see supplementary material), vaccination dates and date of previous infections.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 215, 405, 231]]<|/det|>
+## Secondary attack rate (outcome)  
+
+<|ref|>text<|/ref|><|det|>[[115, 234, 882, 336]]<|/det|>
+The secondary attack rate \(^{34}\) , first described by Dr Chapin at the beginning of the last century, refer to the probability of infection among close contacts of an index case in a particular setting (work, household, ...) \(^{35}\) and is one of the key estimate of the transmissibility of the virus. Its raw estimation consists in dividing the number of contaminated contacts by the total number of susceptible contacts declared by the index cases. Adjusted estimation of SAR can be performed using linear regression methods (see statistical analysis).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 348, 408, 365]]<|/det|>
+## Immunity status (main predictor)  
+
+<|ref|>text<|/ref|><|det|>[[115, 384, 850, 418]]<|/det|>
+Immunity status was calculated at the date of the last contact between the index and the contact and was categorized in the following categories:  
+
+<|ref|>text<|/ref|><|det|>[[144, 427, 880, 581]]<|/det|>
+- Vaccinated more than 6 months or less than 6 months. This category included the complete vaccine schemes for the different vaccines available in Geneva, including booster doses, for which the last date of vaccination was more or less than 6 months.- Infected at least one time, more than 6 months or less than 6 months. This category included persons not vaccinated but having at least one positive PCR test result, more or less than one year ago.- Not vaccinated not infected (NVNI). This category included persons not vaccinated and not infected previously to the date of last contact between index case and the contact.- Hybrid infections: persons with complete vaccine scheme and previous infection  
+
+<|ref|>text<|/ref|><|det|>[[115, 592, 802, 608]]<|/det|>
+Addresses were geo- coded using the exhaustive list of all addresses of the State of Geneva.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 621, 193, 636]]<|/det|>
+## Controls  
+
+<|ref|>text<|/ref|><|det|>[[115, 639, 870, 742]]<|/det|>
+Categorization of the socio- economic condition of the neighbourhood area (417 official neighbourhood areas in the State of Geneva) was, similarly to previous work \(^{36}\) , based on an index provided by the centre for the analysis of territorial inequalities (see supplementary material). The statistical office of Geneva provided the type of building and number of inhabitant for each address. The building type were categorised in three categories: building, single houses, or collective structure. This last category included nursing homes, jails, asylums and fire- stations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 752, 877, 819]]<|/det|>
+Body mass index (BMI) was calculated from height and weight and was categorized in obese and non- obese categories. For age superior to 18 years, obese was considered for BMI above \(30 \text{kg} /\text{m}^2\) . For age below 18, we used the extended international body mass index cut- offs corresponding to the threshold of \(30 \text{kg} /\text{m}^2\) at 18 years old \(^{37}\) .  
+
+<|ref|>text<|/ref|><|det|>[[117, 830, 870, 881]]<|/det|>
+Tendency to test was estimated by counting the number of tests performed by each contact during the last 3 months preceding their contact with the index case. This number was categorized in three categories: 0, 1 and more than 2 (2+).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 85, 297, 101]]<|/det|>
+## SARS-CoV-2 variants  
+
+<|ref|>text<|/ref|><|det|>[[117, 103, 864, 155]]<|/det|>
+SARS- CoV- 2 variantsAs the ARGOS data did not contain information about the SARS- CoV- 2 variant type, we divided the study period into period of predominance of SARS- CoV- 2 variant of interest, based on the data provided by the Global Initiative on Sharing Avian Influenza Data \(^{38}\) in the Geneva region:  
+
+<|ref|>text<|/ref|><|det|>[[146, 164, 591, 233]]<|/det|>
+EU1 from 01- 06- 2020 to 01- 02- 2021 Alpha from 02- 02- 2021 to 01- 07- 2021 Delta from 02- 07- 2021 to 20- 12- 2021 Omicron from 21- 12- 2021 to 01- 03- 2022 (mainly BA.1)  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 245, 276, 261]]<|/det|>
+## Statistical analysis  
+
+<|ref|>text<|/ref|><|det|>[[116, 263, 880, 400]]<|/det|>
+SAR was estimated using generalized estimating equations predicting a binary outcome indicating if the contact was contaminated by the index or not. The clusters considered were the index cases, and we assumed an exchangeable correlation structure. We used a Gaussian identity link \(^{39}\) , which allows to estimate the relative proportion increase provoked by each covariate relative to a reference proportion of infected contacts, that is the SAR. Missing data were handled using multiple imputation with chained equations (20 samples, 5 iterations) at the person, infection or contact level (see supplementary material). The analysis was then performed independently on each imputed dataset, and the results were pooled according to the Rubin's rules.  
+
+<|ref|>text<|/ref|><|det|>[[117, 409, 880, 479]]<|/det|>
+All analysis has been performed using R 4.0.0 \(^{40}\) , using the geepack library \(^{41}\) for the general estimating equation, mice \(^{42}\) for the multiple imputation with chained equation and ggplot2 for the figures and graphs. The code used for the analysis- has been made available at the following Gitlab repository: https://gitlab.com/dmongin/scientific_articles/- /tree/main/Effect_of_mRNA_vaccination.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 496, 196, 514]]<|/det|>
+## Results  
+
+<|ref|>text<|/ref|><|det|>[[116, 517, 875, 621]]<|/det|>
+During the period of interest (01- 06- 2020 to 01- 03- 2022), 65'077 infections were recorded among persons living in Geneva and who declared at least one contact person. Among them, 9'890 refused to share their data for research. 15'327 declared contacts also refused to share their data, removing an additional 4'298 infections. The resulting dataset consisted in 50'889 index cases and 111'432 declared contacts. The mean number of declared contact per infected person was 2.2 overall, with a net decrease during the Omicron period (1.6 mean contacts per index, see table 1).  
+
+<|ref|>text<|/ref|><|det|>[[116, 630, 879, 717]]<|/det|>
+Index cases were at \(73\%\) adults between 18 and 64 years, \(22\%\) children and \(4.6\%\) adults older than 65 years. The proportion of children for the index cases tripled between the EU1 wave (11%) and the Delta wave (38%). Overall, children were overrepresented and adults \(>65\) years underrepresented in our cohort when compared to the demographics of the Geneva state (18.5% of children, and 16.5% of adults above 65 years, see supplementary table S1).  
+
+<|ref|>text<|/ref|><|det|>[[116, 726, 864, 777]]<|/det|>
+The vast majority of the index cases had symptoms (94%), among whom more than half had cough (58%). The majority of the contacts reported by the index were persons sharing their home (62%), this percentage increasing up to 78% during the Omicron wave.  
+
+<|ref|>text<|/ref|><|det|>[[116, 779, 875, 846]]<|/det|>
+Concerning the immunity status, the proportion of vaccinated index cases increased from around \(2\%\) during the alpha wave, up to \(52\%\) during the Omicron wave, of which \(25\%\) had their last dose more than 6 months before the infection. Contacts were less vaccinated (37%) and a higher proportion of them were previously infected (10%, compared to 2.9% for the index cases).  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 858, 245, 874]]<|/det|>
+## Overall results  
+
+<|ref|>text<|/ref|><|det|>[[116, 877, 870, 910]]<|/det|>
+Among the 111'432 declared contacts, 21'387 had a positive test result during the 10 days following the date of the last contact with the index case (raw SAR of 19.2%). This raw SAR increased almost
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 82, 867, 150]]<|/det|>
+linearly of 3 percent point per day when increasing the delay from 0 to 8 days, to then plateauing after 10 days (see supplementary figure S1). For the rest of the study, a delay of 10 days was considered. The raw SAR changed across variant and was \(16.5\%\) during the EU1 wave, \(21.2\%\) during the alpha wave, \(16.8\%\) during the delta wave, and \(26.3\%\) during the omicron wave.  
+
+<|ref|>text<|/ref|><|det|>[[115, 161, 872, 297]]<|/det|>
+The reference category was defined as follow: two asymptomatic adult men, neither vaccinated nor with an antecedent infection (NVNI), between the age of 18 and 65, having a house contact in a building in a wealthy neighbourhood, of which the index case being not obese and not a vulnerable person, the contact person having performed one test in the past three months, and being both not vaccinated not infected. For this reference category, the multivariable analysis yielded a SAR of \(33.3\%\) (95%CI: [30.9,35.7]) for the EU1 variant, \(31.0\%\) (95%CI: [27.8,34.1]) for the alpha variant, \(32.8\%\) (95%CI: [32.0,37.0]) for the delta variant and \(41.0\%\) (95%CI: [37.4,44.6]) for the omicron variant.  
+
+<|ref|>text<|/ref|><|det|>[[115, 308, 872, 427]]<|/det|>
+The main variables influencing the SAR (see figure 1) were the immune status of both the index case and the contact, the presence of symptoms or the presence of cough for the index case, the type of relation between the index case and their contacts, the age of the contacts, and the number of test the contact had in the 3 months before the contact date. The age of the index case, as well as the index housing type had a limited effect on the SAR. The gender of both index and contacts, the obesity of the index, the index vulnerability or its neighbourhood socio- economic condition did not affect the SAR.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 441, 250, 456]]<|/det|>
+## Immune status  
+
+<|ref|>text<|/ref|><|det|>[[115, 459, 880, 560]]<|/det|>
+Being previously infected for the index case decreased the SAR, with no obvious difference if the infection was recent, older than 6 months or hybrid (see supplementary table S2 and supplementary figure S2). The reduction of SAR induced by an infection of the index case was of - 9.9 adjusted percent points (pp) (95%CI: [- 14.7, - 6.5]) during the EU1 variant wave, - 8.6pp during the alpha wave [- 13.2, - 4.1], - 10.9pp [- 13.0, - 7.1] during the delta wave and of 4.3pp [- 7.3, - 1.3] during the Omicron wave.  
+
+<|ref|>text<|/ref|><|det|>[[115, 561, 875, 715]]<|/det|>
+The effect of previous infection was stronger for the contacts, with a greater effect when the date of infection was less than 6 months before the index- contact date. Previous infection of less than 6 month or more than 6 months respectively lowered the SAR of - 17.3pp [- 19.3, - 15.4] and - 13.3pp [- 15.4, - 11.8] for EU1, - 26.5 pp [- 28.1, - 24.8] and - 20.6pp [- 23.3, - 18.0] for alpha, - 30.7pp [- 32.1, - 29.4] and - 14.9pp [- 16.8, - 12.9] during delta and - 32.0pp [- 33.9, - 30.0] and - 4.4pp [- 7.6, - 1.2] during omicron. Considering an interaction between immune status and testing tendency, the protection caused by previous infection was around - 6pp stronger for all variants if the contacts were tested at least once during the 3 months preceding their encounter with the index (see supplementary table S3 and supplementary figure S4).  
+
+<|ref|>text<|/ref|><|det|>[[115, 717, 880, 803]]<|/det|>
+Being vaccinated for the index consistently lowered the SAR across VoCs mainly when the last dose of vaccination was less than 6 months before the index- contact date (- 5.0pp [- 9.5, - 0.4] during alpha, 5.6pp [- 6.9, - 4.3] during delta and - 6.6pp [- 8.2, - 4.9] during omicron). A small but significant protective effect of vaccination older than 6 month was observed during omicron (- 2.8pp [- 4.6, - 0.9]).  
+
+<|ref|>text<|/ref|><|det|>[[115, 785, 875, 905]]<|/det|>
+The recent vaccination of the contacts had a strong protective effect for alpha (- 13.5pp [- 16.0, - 11.0]) and delta variants (- 9.5pp [- 10.5, - 8.4]). In this multivariable model without interaction, recent contact vaccination increased the SAR during the omicron wave. This increase vanished when considering an interaction between the immune status and the number of test performed the last 3 months (2.2pp [- 1.8, 6.3] and 2.9pp [- 2.0, 7.8] if the contact performed 1 or more than 2 tests, respectively). If vaccination occurred more than 6 months before the last meeting between index and contact, it did not have a significant effect during the delta variant and even had a net tendency
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 82, 850, 132]]<|/det|>
+to increase the SAR with Omicron (increase of 13.3pp [11.2,15.4]). This increase remained similar even when considering an interaction between the immune status and the number of tests performed 3 months before.  
+
+<|ref|>text<|/ref|><|det|>[[117, 134, 880, 220]]<|/det|>
+Hybrid immunity had a higher protective effect than vaccination but lower than recent infection (- 20.1pp [- 25.8, - 14.3], - 21.7pp [- 22.9, - 20.6], and - 18.2p [- 20.4, - 16.1] for the alpha, delta and omicron wave respectively). The combined recent vaccination for both contact and index (interaction between both immune status) decreased the SAR by - 22pp [- 27, - 21] during alpha and - 17pp [- 14, - 18] during delta, but had no significant effect during omicron.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 232, 477, 248]]<|/det|>
+## Index Reported relationship with contact  
+
+<|ref|>text<|/ref|><|det|>[[116, 251, 875, 370]]<|/det|>
+When the contacts between index and contact took place outside the household, the SAR was substantially lower: it decreased by - 9.4pp [- 10.3, - 8.4], - 13.1pp [- 14.7, - 11.6], - 8.4pp [- 9.6, - 7.3], and - 7.2pp [- 8.8, - 5.5] for the EU1, alpha, delta and omicron variants when the index and its contacts were close relatives not living under the same roof. This diminution of SAR was more pronounced when contact and index were not closed relatives (e.g. professional or recreational relation): - 10.7pp [- 11.9, - 9.5], - 15.3pp [- 17.7, - 13.0], - 10.8pp [- 12.5, - 9.2] and - 11.5pp [- 13.9, - 9] for the EU1, alpha, delta and omicron variants respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 383, 211, 399]]<|/det|>
+## Symptoms  
+
+<|ref|>text<|/ref|><|det|>[[117, 401, 857, 468]]<|/det|>
+Cough increased the transmissibility, but with an amplitude decreasing with the new variants. A coughing index increased the SAR by 5.3pp [4.4, 6.2] and 5.5pp [3.9, 7.1] for the EU1 and alpha variant, but this effect was reduced to 2.2pp [1.0, 3.3] for the delta variant and 1.8pp [0.4, 3.3] for omicron.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 482, 153, 497]]<|/det|>
+## Age  
+
+<|ref|>text<|/ref|><|det|>[[116, 500, 880, 602]]<|/det|>
+Contact children had a lower SAR, especially for early variant: the SAR was - 11.7pp [- 12.4, - 10.9] lower and - 8.1pp [- 9.5, - 6.8] lower for EU1 and alpha, when the decrease was only - 0.2pp [- 1, - 0.6] and - 4.8pp [- 6.1, - 3.5] for delta and omicron. Contact older than 65 tended to be more contaminated at the beginning of the pandemic (4.1pp [2.4, 5.8] supplementary points for EU1), but this effect became non- significant for alpha and delta, and even reversed for omicron, where contact older than 65 years had a SAR - 6.5pp [- 9.6, - 3.4] lower than the reference category.  
+
+<|ref|>text<|/ref|><|det|>[[117, 603, 864, 671]]<|/det|>
+Concerning the age of the index, index children seemed to be more contagious over time (- 4.9pp [- 6.7, - 3.1], - 2.6pp [- 3.8, - 1.4] and - 2.1pp [- 3.9, - 0.4] during the alpha, delta and omicron waves respectively). Being an infected adult older than 65 years increased contagiosity only during the alpha wave (3.7pp [0.1, 7.3]).  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 682, 467, 700]]<|/det|>
+## Effect of testing during the past 90 days  
+
+<|ref|>text<|/ref|><|det|>[[116, 702, 864, 890]]<|/det|>
+The propensity of contacts to perform tests, measured by the number of test performed by the contacts the last 90 days preceding their last encounter with the index, had a large effect on SAR calculation. Those who did not perform any test during this period had a reduced SAR of - 16.3pp [- 17.4, - 15.1], - 11.1pp [- 12.7, - 9.5], - 13.3pp [- 14.6, - 12.0] and - 17.0pp [- 18.6, - 15.5] for the EU1, Alpha, delta and omicron respectively when compared to those who performed one test. Performing two tests or more did not clearly increase the SAR (an increase of 3.0pp [1.1, 4.8] only during the delta wave). The propensity of contacts to perform tests modified the effect of immunity on SAR when comparing univariable and multivariable adjustment. These results are detailed in supplementary material. After adjustment for an interaction between contact immunity of the number of test performed by contact during the past 90 days, testing more enhanced the effect of protection against infection conferred by recent infection and hybrid vaccination, but decreased this effect for
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 83, 808, 117]]<|/det|>
+other immune status, especially during the omicron wave (see supplementary figure S4 and supplementary table S3).  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 129, 288, 145]]<|/det|>
+## Gender differences  
+
+<|ref|>text<|/ref|><|det|>[[117, 148, 866, 216]]<|/det|>
+When stratifying for gender, multivariable models showed similar pattern of results. Nevertheless, previous infections of women index were more protective, and were associated with lower SAR for the EU1 and alpha variants. There was no difference for the previous infections of contacts, though they included both gender (see supplementary figure S3).  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 233, 231, 251]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[115, 255, 880, 444]]<|/det|>
+In this study of \(>50'000\) index cases and \(>110'000\) declared contacts, spanning four different SARS- CoV- 2 variants circulating over almost 2 years, we observe that the immunity conferred by vaccine or infection lowers both the transmission risk and the risk of being infected, and that the latter effect contributes more to the reduction of the virus propagation. The main immune factor lowering the secondary attack rate was natural infection, while vaccination had a more limited impact, even when recent enough. Although having a lower contribution to changes of secondary attack rate, the reduction of contagiousness conferred by vaccination appears to wane less in time and to be less sensitive to variant changes than the increase of the protection against infection. The other variables affecting the transmission of SARS- CoV- 2 were the age of the contact person, the presence of symptoms - especially cough - for the index, the setting of the encounter between index and contact (e.g., home, work) and the tendency of the contact to get tested.  
+
+<|ref|>text<|/ref|><|det|>[[115, 461, 880, 630]]<|/det|>
+Compared to non- vaccinated and never- infected person, vaccination was protective for both index and contacts, the effect for the index being smaller than for contacts, as reported previously \(^{43}\) . Because of the waning of the vaccine- induced immunity \(^{10}\) and the immune escape of successive variants when compared to the previous ones \(^{8,44}\) , the timing of the last vaccination and the VoC concerned were important, especially for the contact. Vaccination, even when performed less than 6 months before, did not add any protection to contacts during the omicron wave. On the other hand, the escaping capacities of omicron did not affect the reduction of contagiousness conferred by recent vaccination. This suggests that vaccine still lowers the viral load of persons infected by Omicron, in agreement with the fact that vaccine diminishes the occurrence of severe disease for this VoC \(^{13,45}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 633, 878, 736]]<|/det|>
+Although vaccination more than 6 months ago still had some effect for index cases, it added no protection for the contacts during delta wave, and even had an opposite effect (i.e. an increase of the SAR) during the omicron wave. This counter intuitive effect might be due to a combination of the strong immune escape of this variant \(^{46}\) and the tendency of vaccinated people to comply less with COVID- 19 mitigation strategies \(^{47}\) , such as physical distancing and mask recommendations. Such result has been observed in a previous study \(^{48}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 738, 880, 908]]<|/det|>
+Infected unvaccinated indexes had a reduced SAR across all variants. This reduction was higher than the one observed for vaccination for Delta, but not for Omicron, in agreement with recent measurement of viral load dynamics \(^{21}\) . Previous infection also showed a strong protective effect against being infected for the contacts, even more after adjusting for their tendency to test. This protection is reduced after 6 months. The reduction of transmission is rather small for early variants, in line with the recently observed slower immunity waning after infection when compared to vaccine \(^{7,16,18}\) , but is substantial for delta and omicron, due to their stronger potential for immune escape \(^{49}\) . This protection against infection was higher than vaccination for all variants (up to 7 times higher for Omicron, in agreement with recent estimate of Gazit and co- authors \(^{17}\) ). This higher and longer lasting protection of the infection when compared to vaccine induced immunity may find its root in a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 616, 100]]<|/det|>
+more global immune response and maybe specific IgA response \(^{50}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 101, 875, 272]]<|/det|>
+Hybrid immunity provided stronger protection and reduction of contagiousness than vaccines, as observed elsewhere \(^{51}\) , but no higher than recent infection, in agreement with a large Israeli study \(^{18}\) . The association of SAR and immune status, either due to recent vaccine or previous infection, changed notably between univariable and adjusted analyses. Though previous infection and recent vaccination were protective in univariable models, it became more so, for all VoCs, in the multivariable model. The main confounder of this association was the tendency of the contact to be tested, which modified the SAR of the non- immune population, our reference category. Indeed, the SAR was much higher among non- immune who tested compared to those who did not, whereas SAR was quite similar among previously infected or recently vaccinated people, irrespective of their tendency to get tested.  
+
+<|ref|>text<|/ref|><|det|>[[116, 289, 872, 340]]<|/det|>
+It is of note that the adjustment in our analysis corrects strongly the SAR value of each variant, resulting in a similar value for the EU1, alpha and delta variant, but a higher SAR for omicron, similar to what has been reported by large reviews \(^{20}\) .  
+
+<|ref|>text<|/ref|><|det|>[[116, 342, 873, 409]]<|/det|>
+The context of the encounter between index and contacts affected greatly the SAR, where more distant relations (work, leisure) led to lower SAR than housing relation, as noticed elsewhere \(^{26}\) . This confirms that the adaptations of measures for pandemic containment during the last waves, such as quarantine restricted to household, was appropriate.  
+
+<|ref|>text<|/ref|><|det|>[[115, 410, 875, 564]]<|/det|>
+Symptomatic indexes have consistently been shown to increase SAR since the beginning of the pandemic \(^{27,28}\) . However, the difference in SAR between symptomatic and asymptomatic is relatively small, suggesting that everyone should be careful to minimize their risk of transmitting the disease, even if not symptomatic. With respect to coughing, the impact of coughing, though significant, decreased for later VoCs. This could be due to a combination of a higher adherence to mask wearing within the population during these periods, and of changes in infection pathway. Indeed, since omicron infects mostly upper respiratory tract \(^{52}\) and produces a higher viral load \(^{53}\) , the higher quantity of virus expelled when naturally breathing or sneezing could explain the lower effect of coughing for this particular VoC.  
+
+<|ref|>text<|/ref|><|det|>[[115, 581, 875, 735]]<|/det|>
+As shown in previous studies \(^{28}\) , we also found that contact children had a lower SAR (both as index and contact) than adults. It has been postulated that difference in contact type, quantity of virus expelled, decreased receptor expression in the respiratory tract or age- related increase in innate immune response in children could explain this difference \(^{54 - 56}\) , but the tendency of children to be more asymptomatic \(^{57}\) could also play a role, as they tend to be less tested. However, this difference with adults decreased with delta and omicron variants compared with the other VoCs. This change is potentially due to both the preference of the new variants for this more immune and unvaccinated population \(^{58}\) and to a potential detection bias (children tended to be less tested at the beginning of the pandemic).  
+
+<|ref|>text<|/ref|><|det|>[[116, 737, 875, 805]]<|/det|>
+Adults older than 65 years had a slightly higher adjusted SAR during the early waves, as reported elsewhere \(^{54}\) . This effect disappeared later in time, probably due to multiple factors, such the implementation of physical distancing and protection, but also detection bias. The above- mentioned underrepresentation of this population in this study could also bias this result.  
+
+<|ref|>text<|/ref|><|det|>[[116, 823, 881, 908]]<|/det|>
+Interestingly, we found no association between living or personal socio- economic circumstances (SEC) and SAR. This result is in line with what was reported by a recent seroprevalence study in Geneva \(^{59}\) . It has to be noted that our study does not concern the first wave of SARS- CoV- 2 pandemic, and the association between COVID- 19 variables and socio- economic condition vary greatly among waves \(^{36}\) . Disparities across the social ladder of the society concerning COVID- 19 have been shown to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 81, 856, 133]]<|/det|>
+concern mainly the access to test \(^{36,60,61}\) , and the COVID- 19 mortality and morbidity \(^{62,63}\) . Even if dependence of SAR on some socio- economic variables have been shown in small samples in some countries \(^{64,65}\) , it may be dependent on a particular situation or time.  
+
+<|ref|>text<|/ref|><|det|>[[117, 143, 857, 229]]<|/det|>
+The associations between all variables and SAR were mostly similar between men and women, in agreement with seroprevalence studies in Switzerland \(^{58,66,67}\) and other SAR studies \(^{28}\) , which have shown that gender or sex affects access to healthcare, morbidity and mortality, but not the contagiousness of SARS- CoV2. Nevertheless, previous infections of women index were associated with lower SAR for the EU1 and alpha variants.  
+
+<|ref|>text<|/ref|><|det|>[[116, 239, 880, 359]]<|/det|>
+The strength of this study is based on the operational database gathering all SARS- CoV- 2 tests performed by a large population of indexes and their contacts, covering 2 years of pandemics and multiple variants of concern. Detailed information on cases and contacts were available, allowing adjustment for a wide range of covariates. In particular, the availability of vaccination status, for both index and contact, adds to the strength of the study. Finally, the canton of Geneva invested a lot of effort to reach vulnerable populations, including non- documented migrants, thus reducing potential selection bias.  
+
+<|ref|>text<|/ref|><|det|>[[116, 368, 880, 472]]<|/det|>
+This study also has limitations. As previously mentioned, people over 65 years old are underrepresented, while young people are overrepresented. Underrepresentation of old people may be due to the handling of contact tracing and isolation by their specific nursing home or healthcare facility. As a consequence, vaccinated people were also underrepresented in our cohort. This could potentially lead to selection bias, although we tried to adjust for most of the main factors potentially influencing the SAR.  
+
+<|ref|>text<|/ref|><|det|>[[116, 480, 881, 653]]<|/det|>
+The main limitation of this observational registry study is information and surveillance bias \(^{68}\) . Attack rate estimation depends on the tests being performed, since contacts will be considered positive only if they were tested. Indeed, we found that the tendency to get tested (number of tests in the 90 days before the date of contact) strongly influenced SAR, with people not testing in the preceding months having a much lower SAR. This propensity of the population to be tested varies over time and depends on the health policies implemented. This is especially the case for children, for whom the testing policies varied from almost no tests during the first waves, even when they were contacts (in part due to recommendations \(^{69}\) but also because they are often not symptomatic \(^{57}\) ) to compulsory autogenic testing in schools if more than two children were infected in a classroom by the end of 2021.  
+
+<|ref|>text<|/ref|><|det|>[[116, 654, 881, 860]]<|/det|>
+There are several potential effects of this bias. First, people from low socio- economic conditions may have avoided testing in order to escape quarantine \(^{36}\) . Although we adjusted for the propensity of the contact to test, we cannot exclude residual confounding partly explaining the absence of association between SEC and SAR. Similarly, previously infected people and vaccinated people tend to test less than non- vaccinated and non- infected persons, since they did not need a test result for their sanitary pass. The Swiss sanitary pass was introduced the \(26^{\text{th}}\) of June 2021, its rules were progressively toughened over time and, in December 2021, allowed only vaccinated or previously infected patients to use common social venues. Non- vaccinated and non- infected persons were thus more incited to test for SARS- CoV- 2 than infected or vaccinated persons. Although the adjustment for the propensity to test and for its interaction with the immune status confirmed and even strengthened the effect of immune status on SAR, we cannot completely rule out residual bias, inherent to any retrospective study.  
+
+<|ref|>text<|/ref|><|det|>[[116, 861, 858, 895]]<|/det|>
+Finally, our study did not assess variant by genotype results based on a PCR test, but was based on period of time of variant dominance. Due to an overlap between every variant change, this could
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 857, 117]]<|/det|>
+alter our result, but probably in a minimal way since variants became dominant quite quickly after they emerged.  
+
+<|ref|>text<|/ref|><|det|>[[114, 126, 884, 386]]<|/det|>
+We are now 3 years into this pandemic and the virus shows no sign of receding. Our study shows that mRNA vaccination alone, although effective for reducing severe outcomes or hospitalisations \(^{13,45}\) , had a limited effect but is not sufficient anymore to contains or moderate Sars- CoV- 2 propagation. Infections have important reduction effect on the virus transmission but they are accompanied, at a population level, by cumulative effects of Sars- CoV- 2 infections \(^{70}\) provoking potential immunity deficiency \(^{71}\) , long lasting symptoms \(^{72}\) , including cardiac \(^{73}\) and neurological \(^{74}\) damages. These health consequences concerning an increasing part of the population \(^{75}\) and the weakening of the health system due to overcrowding of hospitals and exhaustion of health personnel rule out the possibility of public health policies relying solely on natural infections. Public health policies would, on the contrary, focus on reducing the number of infections for all persons, vaccinated or not, with effective and socially acceptable non pharmaceutical interventions such as air purification \(^{76,77}\) , ventilation \(^{78,79}\) , or mask wearing \(^{80}\) . Indeed, since air purification and ventilation require no individual effort, they can be implemented everywhere, thus avoiding individual behavioural barriers. Finally, to be able to study the evolution of the SARS- CoV- 2 among the population, it is of prime importance to continue to monitor the infections in the community and the general population \(^{81}\) .
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 240, 104]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[113, 125, 880, 888]]<|/det|>
+1. Coronavirus Disease (COVID-19) Situation Reports. https://www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports.  
+2. Wang, H. et al. Estimating excess mortality due to the COVID-19 pandemic: a systematic analysis of COVID-19-related mortality, 2020–21. The Lancet 399, 1513–1536 (2022).  
+3. Shiels, M. S., Haque, A. T., Berrington de González, A. & Freedman, N. D. Leading Causes of Death in the US During the COVID-19 Pandemic, March 2020 to October 2021. JAMA Internal Medicine (2022) doi:10.1001/jamainternmed.2022.2476.  
+4. Chen, C. et al. Global Prevalence of Post COVID-19 Condition or Long COVID: A Meta-Analysis and Systematic Review. J Infect Dis jiac136 (2022) doi:10.1093/infdis/jiac136.  
+5. Smyth, D. S. et al. Tracking cryptic SARS-CoV-2 lineages detected in NYC wastewater. Nat Commun 13, 635 (2022).  
+6. Amoutzias, G. D. et al. The Remarkable Evolutionary Plasticity of Coronaviruses by Mutation and Recombination: Insights for the COVID-19 Pandemic and the Future Evolutionary Paths of SARS-CoV-2. Viruses 14, 78 (2022).  
+7. Hall, V. et al. Protection against SARS-CoV-2 after Covid-19 Vaccination and Previous Infection. N Engl J Med 386, 1207–1220 (2022).  
+8. Wang, P. et al. Antibody resistance of SARS-CoV-2 variants B.1.351 and B.1.1.7. Nature 593, 130–135 (2021).  
+9. Polack, F. P. et al. Safety and Efficacy of the BNT162b2 mRNA Covid-19 Vaccine. N Engl J Med 383, 2603–2615 (2020).  
+10. Feikin, D. R. et al. Duration of effectiveness of vaccines against SARS-CoV-2 infection and COVID-19 disease: results of a systematic review and meta-regression. The Lancet 399, 924–944 (2022).  
+11. Rosenberg, E. S. et al. Covid-19 Vaccine Effectiveness in New York State. N Engl J Med 386, 116–127 (2022).  
+12. Levin, E. G. et al. Waning Immune Humoral Response to BNT162b2 Covid-19 Vaccine over 6 Months. New England Journal of Medicine 385, e84 (2021).  
+13. Buchan, S. A. et al. Estimated Effectiveness of COVID-19 Vaccines Against Omicron or Delta Symptomatic Infection and Severe Outcomes. JAMA Network Open 5, e2232760 (2022).  
+14. Nordström, P., Ballin, M. & Nordström, A. Effectiveness of a fourth dose of mRNA COVID-19 vaccine against all-cause mortality in long-term care facility residents and in the oldest old: A nationwide, retrospective cohort study in Sweden. The Lancet Regional Health – Europe 21, (2022).  
+15. Patalon, T. et al. Waning effectiveness of the third dose of the BNT162b2 mRNA COVID-19 vaccine. Nat Commun 13, 3203 (2022).  
+16. Altarawneh, H. N. et al. Effects of Previous Infection and Vaccination on Symptomatic Omicron Infections. N Engl J Med 387, 21–34 (2022).  
+17. Gazit, S. et al. Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) Naturally Acquired Immunity versus Vaccine-induced Immunity, Reinfections versus Breakthrough Infections: A Retrospective Cohort Study. Clinical Infectious Diseases 75, e545–e551 (2022).  
+18. Goldberg, Y. et al. Protection and Waning of Natural and Hybrid Immunity to SARS-CoV-2. New England Journal of Medicine 386, 2201–2212 (2022).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 85, 875, 138]]<|/det|>
+19. Pilz, S., Theiler-Schwetz, V., Trummer, C., Krause, R. & Ioannidis, J. P. A. SARS-CoV-2 reinfections: Overview of efficacy and duration of natural and hybrid immunity. Environmental Research 209, 112911 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 140, 860, 192]]<|/det|>
+20. Madewell, Z. J., Yang, Y., Longini, I. M., Jr, Halloran, M. E. & Dean, N. E. Household Secondary Attack Rates of SARS-CoV-2 by Variant and Vaccination Status: An Updated Systematic Review and Meta-analysis. JAMA Network Open 5, e229317 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 195, 843, 246]]<|/det|>
+21. Woodbridge, Y., Amit, S., Huppert, A. & Kopelman, N. M. Viral load dynamics of SARS-CoV-2 Delta and Omicron variants following multiple vaccine doses and previous infection. Nat Commun 13, 6706 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 249, 860, 283]]<|/det|>
+22. Puhach, O. et al. Infectious viral load in unvaccinated and vaccinated individuals infected with ancestral, Delta or Omicron SARS-CoV-2. Nat Med 28, 1491-1500 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 286, 870, 320]]<|/det|>
+23. Karumanagoundar, K. et al. Secondary attack rate of COVID-19 among contacts and risk factors, Tamil Nadu, March-May 2020: a retrospective cohort study. BMJ Open 11, e051491 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 323, 860, 356]]<|/det|>
+24. Ng, O. T. et al. SARS-CoV-2 seroprevalence and transmission risk factors among high-risk close contacts: a retrospective cohort study. Lancet Infect Dis 21, 333-343 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 359, 820, 393]]<|/det|>
+25. Tian, T. & Huo, X. Secondary attack rates of COVID-19 in diverse contact settings, a meta-analysis. J Infect Dev Crites 14, 1361-1367 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 395, 860, 447]]<|/det|>
+26. Koh, W. C. et al. What do we know about SARS-CoV-2 transmission? A systematic review and meta-analysis of the secondary attack rate and associated risk factors. PLoS One 15, e0240205 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 450, 866, 484]]<|/det|>
+27. Madewell, Z. J., Yang, Y., Longini, I. M., Halloran, M. E. & Dean, N. E. Household Transmission of SARS-CoV-2: A Systematic Review and Meta-analysis. JAMA Netw Open 3, e2031756 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 487, 840, 539]]<|/det|>
+28. Madewell, Z. J., Yang, Y., Longini, I. M., Halloran, M. E. & Dean, N. E. Factors Associated With Household Transmission of SARS-CoV-2: An Updated Systematic Review and Meta-analysis. JAMA Netw Open e2122240-e2122240 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 542, 860, 576]]<|/det|>
+29. Ng, D. C.-E. et al. Risk factors associated with household transmission of SARS-CoV-2 in Negeri Sembilan, Malaysia. J Paediatr Child Health (2021) doi:10.1111/jpc.15821.  
+
+<|ref|>text<|/ref|><|det|>[[115, 579, 880, 649]]<|/det|>
+30. Elgersma, I. H., Telle, K., Elstrom, P., Skagseth, H. & Kacelnik, O. Transmission of severe acute respiratory coronavirus virus 2 (SARS-CoV-2) between hospital workers and members of their household: Nationwide, registry-based, cohort study from Norway. Infect Control Hosp Epidemiol 1-6 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 653, 833, 686]]<|/det|>
+31. Lyngse, F. P. et al. Nationwide study on SARS-CoV-2 transmission within households from lockdown to reopening, Denmark, 27 February 2020 to 1 August 2020. Euro surveill (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 689, 880, 722]]<|/det|>
+32. Genecand, C. et al. Cohort profile: Actionable Register of Geneva Outpatients and inpatients with SARS-CoV-2 (ARGOS). BMJ Open 11, e048946 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 725, 620, 740]]<|/det|>
+33. Bilan et état de la population du canton de Genève en 2021. 3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 743, 825, 777]]<|/det|>
+34. FOX, J. P. FAMILY-BASED EPIDEMIOLOGIC STUDIES: THE SECOND WADE HAMPTON FROST LECTURE12. American Journal of Epidemiology 99, 165-179 (1974).  
+
+<|ref|>text<|/ref|><|det|>[[115, 780, 870, 814]]<|/det|>
+35. Liu, Y., Eggo, R. M. & Kucharski, A. J. Secondary attack rate and superspreading events for SARS-CoV-2. Lancet 395, e47 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 817, 812, 869]]<|/det|>
+36. Mongin, D. et al. Neighbourhood socio-economic vulnerability and access to COVID-19 healthcare during the first two waves of the pandemic in Geneva, Switzerland: A gender perspective. eClinicalMedicine 46, (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 872, 844, 906]]<|/det|>
+37. Cole, T. J. & Lobstein, T. Extended international (IOTF) body mass index cut-offs for thinness, overweight and obesity. Pediatric Obesity 7, 284-294 (2012).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 85, 770, 101]]<|/det|>
+38. Hodcroft, E. B. CoVariants: SARS-CoV-2 Mutations and Variants of Interest. (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 103, 870, 137]]<|/det|>
+39. Cheung, Y. B. A modified least-squares regression approach to the estimation of risk difference. Am J Epidemiol 166, 1337-1344 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[113, 139, 830, 173]]<|/det|>
+40. R Core Team. R: A Language and Environment for Statistical Computing. (R Foundation for Statistical Computing, 2019).  
+
+<|ref|>text<|/ref|><|det|>[[113, 175, 875, 210]]<|/det|>
+41. Hojsgaard, S., Halekoh, U. & Yan, J. The R Package geepack for Generalized Estimating Equations. Journal of Statistical Software 15, 1-11 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[113, 212, 866, 246]]<|/det|>
+42. Buuren, S. van & Groothuis-Oudshoorn, K. mice: Multivariate Imputation by Chained Equations in R. Journal of Statistical Software 45, 1-67 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[113, 248, 856, 283]]<|/det|>
+43. Ng, O. T. et al. Impact of Delta Variant and Vaccination on SARS-CoV-2 Secondary Attack Rate Among Household Close Contacts. The Lancet Regional Health - Western Pacific 17, (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 285, 857, 319]]<|/det|>
+44. Thorne, L. G. et al. Evolution of enhanced innate immune evasion by SARS-CoV-2. Nature 602, 487-495 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[114, 321, 834, 374]]<|/det|>
+45. Stowe, J., Andrews, N., Kirsebom, F., Ramsay, M. & Bernal, J. L. Effectiveness of COVID-19 vaccines against Omicron and Delta hospitalisation, a test negative case-control study. Nat Commun 13, 5736 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[113, 377, 860, 411]]<|/det|>
+46. Willett, B. J. et al. SARS-CoV-2 Omicron is an immune escape variant with an altered cell entry pathway. Nat Microbiol 7, 1161-1179 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[113, 413, 875, 448]]<|/det|>
+47. Andersson, O., Campos-Mercade, P., Meier, A. N. & Wengström, E. Anticipation of COVID-19 vaccines reduces willingness to socially distance. Journal of Health Economics 80, 102530 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 450, 833, 484]]<|/det|>
+48. Israel, A. et al. Elapsed time since BNT162b2 vaccine and risk of SARS-CoV-2 infection: test negative design study. BMJ 375, e067873 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 486, 880, 521]]<|/det|>
+49. Jalali, N. et al. Increased household transmission and immune escape of the SARS-CoV-2 Omicron compared to Delta variants. Nat Commun 13, 5706 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[113, 523, 870, 558]]<|/det|>
+50. Pérez-Alós, L. et al. Modeling of waning immunity after SARS-CoV-2 vaccination and influencing factors. Nat Commun 13, 1614 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[113, 560, 840, 594]]<|/det|>
+51. Abu-Raddad, L. J. et al. Association of Prior SARS-CoV-2 Infection With Risk of Breakthrough Infection Following mRNA Vaccination in Qatar. JAMA 326, 1930-1939 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 596, 803, 630]]<|/det|>
+52. Meng, B. et al. Altered TMPRSS2 usage by SARS-CoV-2 Omicron impacts infectivity and fusogenicity. Nature 603, 706-714 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[113, 632, 870, 685]]<|/det|>
+53. Riediker, M. et al. Higher viral load and infectivity increase risk of aerosol transmission for Delta and Omicron variants of SARS-CoV-2. Swiss Medical Weekly (2022) doi:10.4414/smw.2022.w30133.  
+
+<|ref|>text<|/ref|><|det|>[[113, 688, 872, 722]]<|/det|>
+54. Stich, M. et al. Transmission of Severe Acute Respiratory Syndrome Coronavirus 2 in Households with Children, Southwest Germany, May-August 2020. Emerg Infect Dis 27, 3009-3019 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 725, 860, 777]]<|/det|>
+55. Reukers, D. F. M. et al. High Infection Secondary Attack Rates of Severe Acute Respiratory Syndrome Coronavirus 2 in Dutch Households Revealed by Dense Sampling. Clinical Infectious Diseases 74, 52-58 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[113, 780, 870, 814]]<|/det|>
+56. Bunyavanich, S., Do, A. & Vicencio, A. Nasal Gene Expression of Angiotensin-Converting Enzyme 2 in Children and Adults. JAMA 323, 2427-2429 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[113, 817, 866, 851]]<|/det|>
+57. Molteni, E. et al. Illness duration and symptom profile in symptomatic UK school-aged children tested for SARS-CoV-2. The Lancet Child & Adolescent Health 5, 708-718 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 853, 875, 907]]<|/det|>
+58. Zaballa, M.-E. et al. Seroprevalence of anti-SARS-CoV-2 antibodies and cross-variant neutralization capacity after the Omicron BA.2 wave in Geneva, Switzerland: a population-based study. Lancet Reg Health Eur 24, 100547 (2023).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 84, 880, 120]]<|/det|>
+59. Santa-Ramírez, H.-A. et al. Socio-economic determinants of SARS-CoV-2 infection: Results from a population-based cross-sectional serosurvey in Geneva, Switzerland. Frontiers in Public Health 10, (2022).  
+
+<|ref|>text<|/ref|><|det|>[[113, 112, 880, 870]]<|/det|>
+60. Riou, J. et al. Socioeconomic position and the COVID-19 care cascade from testing to mortality in Switzerland: a population-based analysis. The Lancet Public Health 0, (2021). 
+61. Vandentorren, S. et al. The effect of social deprivation on the dynamic of SARS-CoV-2 infection in France: a population-based analysis. The Lancet Public Health 0, (2022). 
+62. Williamson, E. J. et al. Factors associated with COVID-19-related death using OpenSAFELY. Nature 584, 430-436 (2020). 
+63. Bilal, U., Tabb, L. P., Barber, S. & Diez Roux, A. V. Spatial Inequities in COVID-19 Testing, Positivity, Confirmed Cases, and Mortality in 3 U.S. Cities. Ann Intern Med (2021) doi:10.7326/M20-3936. 
+64. Righolt, C. H., Zhang, G., Sever, E., Wilkinson, K. & Mahmud, S. M. Patterns and descriptors of COVID-19 testing and lab-confirmed COVID-19 incidence in Manitoba, Canada, March 2020-May 2021: A population-based study. The Lancet Regional Health - Americas 2, 100038 (2021). 
+65. Gentile, I. et al. COVID-19 Post-Exposure Evaluation (COPE) Study: Assessing the Role of Socio-Economic Factors in Household SARS-CoV-2 Transmission within Campania Region (Southern Italy). International Journal of Environmental Research and Public Health 19, 10262 (2022). 
+66. Amati, R. et al. Functional immunity against SARS-CoV-2 in the general population after a booster campaign and the Delta and Omicron waves, Switzerland, March 2022. Eurosurveillance 27, 2200561 (2022). 
+67. Stringhini, S. et al. Seroprevalence of anti-SARS-CoV-2 antibodies after the second pandemic peak. The Lancet Infectious Diseases 21, 600-601 (2021). 
+68. Tancredi, S., Anker, D., Rosella, L. & Chiolero, A. Elimination of covid-19: beware of surveillance bias. BMJ 374, n2126 (2021). 
+69. Testing strategy for children (0-12 years of age) and quarantining strategy of children and adolescents in Switzerland - Swiss National COVID-19 Science Task Force. https://sciencetaskforce.ch/en/policy-brief/testing-strategy-for-children-0-12-years-of-age-and-quarantining-strategy-of-children-and-adolescents-in-switzerland/. 
+70. Bowe, B., Xie, Y. & Al-Aly, Z. Acute and postacute sequelae associated with SARS-CoV-2 reinfection. Nat Med 28, 2398-2405 (2022). 
+71. Phetsuphanh, C. et al. Immunological dysfunction persists for 8 months following initial mild-to-moderate SARS-CoV-2 infection. Nat Immunol 23, 210-216 (2022). 
+72. Tran, V.-T., Porcher, R., Pane, I. & Ravaud, P. Course of post COVID-19 disease symptoms over time in the ComPaRe long COVID prospective e-cohort. Nat Commun 13, 1812 (2022). 
+73. Xie, Y., Xu, E., Bowe, B. & Al-Aly, Z. Long-term cardiovascular outcomes of COVID-19. Nat Med 28, 583-590 (2022). 
+74. Taquet, M. et al. Neurological and psychiatric risk trajectories after SARS-CoV-2 infection: an analysis of 2-year retrospective cohort studies including 1 284 437 patients. The Lancet Psychiatry 9, 815-827 (2022). 
+75. Global Burden of Disease Long COVID Collaborators. Estimated Global Proportions of Individuals With Persistent Fatigue, Cognitive, and Respiratory Symptom Clusters Following Symptomatic COVID-19 in 2020 and 2021. JAMA 328, 1604-1615 (2022).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 868, 118]]<|/det|>
+76. Thuresson, S. et al. Airborne Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) in Hospitals: Effects of Aerosol-Generating Procedures, HEPA-Filtration Units, Patient Viral Load, and Physical Distance. Clin. Infect. Dis. 75, E89-E96 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 120, 850, 155]]<|/det|>
+77. Conway Morris, A. et al. The Removal of Airborne Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) and Other Microbial Bioaerosols by Air Filtration on Coronavirus Disease 2019 (COVID-19) Surge Units. Clinical Infectious Diseases 75, e97-e101 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 157, 850, 192]]<|/det|>
+78. Stadnytskyi, V., Anfinrud, P. & Bax, A. Breathing, speaking, coughing or sneezing: What drives transmission of SARS-CoV-2? J Intern Med 290, 1010-1027 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 195, 855, 230]]<|/det|>
+79. Zhang, R., Li, Y., Zhang, A. L., Wang, Y. & Molina, M. J. Identifying airborne transmission as the dominant route for the spread of COVID-19. Proceedings of the National Academy of Sciences 117, 14857-14863 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 232, 855, 285]]<|/det|>
+80. Cowger, T. L. et al. Lifting Universal Masking in Schools — Covid-19 Incidence among Students and Staff. New England Journal of Medicine 387, 1935-1946 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 287, 855, 323]]<|/det|>
+81. Chiolero, A., Tancredi, S. & Cullati, S. Covid-19: Surveillance systems for the new normal. BMJ 376, o710 (2022).
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[116, 87, 315, 103]]<|/det|>
+# Figures and tables
+
+<|ref|>table_caption<|/ref|><|det|>[[116, 110, 847, 138]]<|/det|>
+Table1: socio-demographics characteristics of the index cases and declared contacts for the whole study period (Overall) and stratified per periods of variant predominant.
+
+<|ref|>table<|/ref|><|det|>[[101, 171, 975, 900]]<|/det|>
+
+<table><tr><td></td><td>Overall</td><td>EU1</td><td>alpha</td><td>delta</td><td>Omicron</td><td>Missing</td></tr><tr><td>Number of index cases</td><td>50889</td><td>18884</td><td>6692</td><td>11628</td><td>13685</td><td></td></tr><tr><td>Mean number of contact (SD) per index case</td><td>2.19 (1.90)</td><td>2.42 (2.30)</td><td>2.50 (1.83)</td><td>2.38 (1.87)</td><td>1.56 (0.93)</td><td></td></tr><tr><td>Total number of contacts</td><td>111432</td><td>45755</td><td>16755</td><td>27636</td><td>21286</td><td></td></tr><tr><td>Number of infected contacts within 10 days following<br>contact (%)</td><td>21387 (19.2)</td><td>7588 (16.6)</td><td>3546 (21.2)</td><td>4652 (16.8)</td><td>5601 (26.3)</td><td>0.0</td></tr><tr><td>Index Socio-economic-<br>neighbourhood</td><td></td><td></td><td></td><td></td><td></td><td>3.3</td></tr><tr><td>Healthy</td><td>39578 (36.4)</td><td>16066 (35.7)</td><td>5689 (34.3)</td><td>10715 (40.0)</td><td>7108 (34.7)</td><td></td></tr><tr><td>Slightly vulnerable</td><td>18127 (16.7)</td><td>7141 (15.9)</td><td>3025 (18.2)</td><td>4305 (16.1)</td><td>3656 (17.8)</td><td></td></tr><tr><td>Moderately vulnerable</td><td>19496 (17.9)</td><td>8200 (18.2)</td><td>3078 (18.6)</td><td>4513 (16.9)</td><td>3705 (18.1)</td><td></td></tr><tr><td>Highly vulnerable</td><td>31612 (29.1)</td><td>13582 (30.2)</td><td>4784 (28.9)</td><td>7223 (27.0)</td><td>6023 (29.4)</td><td></td></tr><tr><td>Index vulnerable person<br>(%)</td><td>8530(7.8)</td><td>3507(7.8)</td><td>1849 (11.1)</td><td>2149(8.0)</td><td>1025(5.0)</td><td>2.2</td></tr><tr><td>Index with symptoms (%)</td><td>98064 (94.1)</td><td>40877 (95.7)</td><td>15204 (91.4)</td><td>24538 (92.5)</td><td>17445 (95.5)</td><td>6.7</td></tr><tr><td>Index with cough (%)</td><td>60148 (57.7)</td><td>23657 (55.4)</td><td>9968 (59.9)</td><td>15124 (57.0)</td><td>11399 (62.4)</td><td>6.7</td></tr><tr><td>Index obesity (%)</td><td>9346 (10.5)</td><td>3892 (11.5)</td><td>1841 (11.6)</td><td>2052(8.3)</td><td>1561 (10.3)</td><td>20.1</td></tr><tr><td>Index women (%)</td><td>59604 (53.6)</td><td>24421 (53.4)</td><td>8973 (53.7)</td><td>14524 (52.8)</td><td>11686 (55.1)</td><td>0.2</td></tr><tr><td>Index age category (%)</td><td></td><td></td><td></td><td></td><td></td><td>0.0</td></tr><tr><td>18-65</td><td>81920 (73.5)</td><td>38236 (83.6)</td><td>12035 (71.8)</td><td>16101 (58.3)</td><td>15548 (73.1)</td><td></td></tr><tr><td>0-18</td><td>24358 (21.9)</td><td>4862 (10.6)</td><td>3980 (23.8)</td><td>10360 (37.5)</td><td>5156 (24.2)</td><td></td></tr><tr><td>65+</td><td>5135(4.6)</td><td>2656(5.8)</td><td>740(4.4)</td><td>1169(4.2)</td><td>570(2.7)</td><td></td></tr><tr><td>Index immune status (%)</td><td></td><td></td><td></td><td></td><td></td><td>0.0</td></tr><tr><td>Infected&lt;6 months</td><td>625(0.6)</td><td>262(0.6)</td><td>148(0.9)</td><td>44(0.2)</td><td>171(0.8)</td><td></td></tr><tr><td>Infected&gt;6 months</td><td>1286(1.2)</td><td>68(0.1)</td><td>58(0.3)</td><td>357(1.3)</td><td>803(3.8)</td><td></td></tr><tr><td>hybrid</td><td>1269(1.1)</td><td>0(0.0)</td><td>10(0.1)</td><td>148(0.5)</td><td>1111(5.2)</td><td></td></tr><tr><td>Non-vaccinated non<br>infected (NVNI)</td><td>89927(80.7)</td><td>45416(99.3)</td><td>16179(96.6)</td><td>19194(69.5)</td><td>9138(42.9)</td><td></td></tr><tr><td>Vaccinated&lt;6 months</td><td>11853(10.6)</td><td>9(0.0)</td><td>360(2.1)</td><td>6067(22.0)</td><td>5417(25.4)</td><td></td></tr><tr><td>Vaccinated&gt;6 months</td><td>6471(5.8)</td><td>0(0.0)</td><td>0(0.0)</td><td>1825(6.6)</td><td>4646(21.8)</td><td></td></tr><tr><td>Index housing type (%)</td><td></td><td></td><td></td><td></td><td></td><td>2.2</td></tr><tr><td>Building</td><td>88528(80.6)</td><td>36430(80.3)</td><td>13487(80.9)</td><td>21399(79.0)</td><td>17212(83.1)</td><td></td></tr><tr><td>Single house</td><td>18271(16.6)</td><td>7438(16.4)</td><td>2771(16.6)</td><td>5094(18.8)</td><td>2968(14.3)</td><td></td></tr><tr><td>Collective structure</td><td>3063(2.8)</td><td>1516(3.3)</td><td>420(2.5)</td><td>594(2.2)</td><td>533(2.6)</td><td></td></tr><tr><td>contact_type(%)</td><td></td><td></td><td></td><td></td><td></td><td>38.2</td></tr><tr><td>Same roof</td><td>43808(62.5)</td><td>16201(56.1)</td><td>10228(65.5)</td><td>13374(65.5)</td><td>4005(77.3)</td><td></td></tr><tr><td>Intimate or familial</td><td>19373(27.6)</td><td>8899(30.8)</td><td>4168(26.7)</td><td>5444(26.7)</td><td>862(16.6)</td><td></td></tr><tr><td>During the day</td><td>6908(9.9)</td><td>3777(13.1)</td><td>1211(7.8)</td><td>1606(7.9)</td><td>314(6.1)</td><td></td></tr><tr><td>Contact female (%)</td><td>56271(51.7)</td><td>22846(51.2)</td><td>8567(51.8)</td><td>14141(52.7)</td><td>10717(51.4)</td><td>2.3</td></tr><tr><td>Contact immune status (%)</td><td></td><td></td><td></td><td></td><td></td><td>9.5</td></tr></table>
+
+<--- Page Split --->
+<|ref|>table<|/ref|><|det|>[[101, 80, 975, 360]]<|/det|>
+
+<table><tr><td>Infected &lt; 6 months</td><td>5216(5.2)</td><td>1818(4.3)</td><td>871(5.6)</td><td>1394(6.1)</td><td>1133(5.5)</td><td></td></tr><tr><td>Infected &gt; 6 months</td><td>1727(1.7)</td><td>199(0.5)</td><td>185(1.2)</td><td>592(2.6)</td><td>751(3.7)</td><td></td></tr><tr><td>hybrid</td><td>3114(3.1)</td><td>0(0.0)</td><td>53(0.3)</td><td>1212(5.3)</td><td>1849(9.0)</td><td></td></tr><tr><td>Non-vaccinated non infected (NVNI)</td><td>76800(76.0)</td><td>40103(95.2)</td><td>13657(87.2)</td><td>12130(53.3)</td><td>10910(53.3)</td><td></td></tr><tr><td>Vaccinated &lt; 6 months</td><td>10271(10.2)</td><td>13(0.0)</td><td>891(5.7)</td><td>6360(28.0)</td><td>3007(14.7)</td><td></td></tr><tr><td>Vaccinated &gt; 6 months</td><td>3879(3.8)</td><td>0(0.0)</td><td>0(0.0)</td><td>1066(4.7)</td><td>2813(13.7)</td><td></td></tr><tr><td>Contact age category (%)</td><td></td><td></td><td></td><td></td><td></td><td>9.9</td></tr><tr><td>18-65</td><td>63497(63.2)</td><td>29161(69.3)</td><td>9266(59.2)</td><td>13334(58.7)</td><td>11736(58.6)</td><td></td></tr><tr><td>0-18</td><td>31733(31.6)</td><td>10421(24.8)</td><td>5518(35.3)</td><td>8128(35.8)</td><td>7666(38.3)</td><td></td></tr><tr><td>65+</td><td>5252(5.2)</td><td>2495(5.9)</td><td>858(5.5)</td><td>1269(5.6)</td><td>630(3.1)</td><td></td></tr><tr><td>Number of tests last 3 months (%)</td><td></td><td></td><td></td><td></td><td></td><td>0.0</td></tr><tr><td>0</td><td>80713(72.4)</td><td>36439(79.6)</td><td>11964(71.4)</td><td>19573(70.8)</td><td>12737(59.8)</td><td></td></tr><tr><td>1</td><td>21964(19.7)</td><td>7966(17.4)</td><td>3531(21.1)</td><td>5305(19.2)</td><td>5162(24.3)</td><td></td></tr><tr><td>2+</td><td>8755(7.9)</td><td>1350(3.0)</td><td>1260(7.5)</td><td>2758(10.0)</td><td>3387(15.9)</td><td></td></tr></table>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[81, 117, 919, 300]]<|/det|>
+Figure 1: Adjusted Secondary Attack Rate (SAR) stratified per variant (EU1, alpha, delta and omicron), with the reference value indicated with a vertical line (reference of each covariate is indicated in bold in parenthesis). This multivariate analysis considers the effect of the index case gender, age, obesity, presence of symptoms, presence of cough, immunity status, neighbourhood socio-economic condition, vulnerability and type of living; the link between the index case and its contacts, and for the contact persons, their gender, age, number of tests performed the three months before the contact date with the index case, and their immunity status. The reference index case - contact relation of this multivariate analysis is the contact between two men of age below 65 living at the same place, the index being not vaccinated not infected (NVNI), not obese, living in a wealthy neighbourhood and being not a vulnerable person, living in a housing building, and the contact person being a NVNI adult men who performed one SARS- CoV- 2 test during the last 3 month preceding the contact.  
+
+<|ref|>image<|/ref|><|det|>[[130, 303, 870, 870]]<|/det|>
+
+<--- Page Split --->
+<|ref|>table_caption<|/ref|><|det|>[[115, 84, 875, 184]]<|/det|>
+Table 2: Coefficients of the multivariate generalized estimating equation [Confidence Interval], providing the additional effect on the reference secondary attack rate (first line), for the 4 periods of dominance of the variants EU1, alpha, delta, and omicron. The reference category for each categorical variable is indicated in bold in parenthesis. The left column indicates is the variable concern the index case, the contact, or their relation. p values are indicated with *. *: 0.01<p<0.05, **:0.001<p<0.01, ***: p<0.001 
+
+<|ref|>table<|/ref|><|det|>[[32, 195, 963, 911]]<|/det|>
+<table><tr><td colspan="2"></td><td>EU1</td><td>alpha</td><td>delta</td><td>omicron</td></tr><tr><td rowspan="3">Index</td><td>Reference</td><td>33.3*** [30.9,35.7]</td><td>31.0*** [27.8,34.1]</td><td>34.5*** [32.0,37.0]</td><td>41.0*** [37.4,44.6]</td></tr><tr><td>Immunity: previously infected (NVNI)</td><td>-9.9*** [-13.3,-6.6]</td><td>-8.6*** [-13.1,-4.1]</td><td>-10.9*** [-13.9,-7.9]</td><td>-4.3** [-7.3,-1.3]</td></tr><tr><td>Immunity: vaccinated &lt; 6 months (NVNI)</td><td></td><td>-5.0* [-9.5,-0.4]</td><td>-5.6*** [-6.9,-4.3]</td><td>-6.6*** [-8.2,-4.9]</td></tr><tr><td>Index</td><td>Immunity: vaccinated &gt; 6 months (NVNI)</td><td></td><td></td><td>-1.3 [-3.5,0.9]</td><td>-2.8** [-4.6,-0.9]</td></tr><tr><td>Index</td><td>women (men)</td><td>-0.3 [-1.2,0.5]</td><td>0.1 [-1.4,1.5]</td><td>0.1 [-1.0,1.1]</td><td>-1.7* [-3.0,-0.4]</td></tr><tr><td>Index</td><td>age 0-17 (18-64)</td><td>-0.7 [-2.1,0.6]</td><td>-4.9*** [-6.6,-3.1]</td><td>-2.6*** [-3.8,-1.3]</td><td>-2.1* [-3.9,-0.4]</td></tr><tr><td>Index</td><td>age 65+ (18-64)</td><td>1.2 [-0.6,3.0]</td><td>3.8* [0.1,7.4]</td><td>1.5 [-0.9,4.0]</td><td>2.0 [-2.0,6.1]</td></tr><tr><td>Index</td><td>Obese (not obese)</td><td>1.1 [-0.4,2.5]</td><td>2.1 [-0.3,4.5]</td><td>0.5 [-1.5,2.4]</td><td>-0.8 [-3.0,1.4]</td></tr><tr><td>Index</td><td>Symptoms (no symptoms)</td><td>2.0 [-0.1,4.0]</td><td>7.6*** [5.1,10.0]</td><td>5.2*** [3.4,7.0]</td><td>4.5** [1.5,7.6]</td></tr><tr><td>Index</td><td>Cough (no cough)</td><td>5.3*** [4.4,6.2]</td><td>5.5*** [3.9,7.1]</td><td>2.2*** [1.0,3.3]</td><td>1.8* [0.4,3.3]</td></tr><tr><td>Index</td><td>neighbourhood poverty: sligh (wealthy)</td><td>-0.3 [-1.6,1.0]</td><td>0.8 [-1.4,3.0]</td><td>-1.1 [-2.6,0.5]</td><td>0.2 [-1.7,2.2]</td></tr><tr><td>Index</td><td>neighbourhood poverty: moderate (wealthy)</td><td>-0.9 [-2.1,0.4]</td><td>-0.9 [-3.1,1.4]</td><td>-0.7 [-2.3,0.8]</td><td>-0.4 [-2.4,1.5]</td></tr><tr><td>Index</td><td>neighbourhood poverty: high (wealthy)</td><td>-0.4 [-1.5,0.7]</td><td>0.0 [-1.9,2.0]</td><td>-1.3 [-2.7,0.1]</td><td>-0.5 [-2.3,1.2]</td></tr><tr><td>Index</td><td>vulnerable (not vulnerable)</td><td>-0.7 [-2.3,0.8]</td><td>-1.2 [-3.6,1.3]</td><td>0.6 [-1.5,2.6]</td><td>-2.1 [-4.9,0.8]</td></tr><tr><td>Index</td><td>living: single house (building)</td><td>0.0 [-1.2,1.3]</td><td>-1.3 [-3.4,0.8]</td><td>0.0 [-1.4,1.5]</td><td>-0.9 [-2.9,1.0]</td></tr><tr><td>Index</td><td>living: collective structure (building)</td><td>0.7 [-1.8,3.2]</td><td>-1.9 [-6.3,2.6]</td><td>-5.1** [-8.3,-1.8]</td><td>-0.1 [-4.3,4.1]</td></tr><tr><td>Index - Contact</td><td>intimate/family (housing)</td><td>-9.4*** [-10.3,-8.4]</td><td>-13.1*** [-14.7,-11.6]</td><td>-8.4*** [-9.6,-7.3]</td><td>-8.0*** [-9.6,-6.3]</td></tr><tr><td>Index - Contact</td><td>pro/school/daily (housing)</td><td>-10.7*** [-11.9,-9.5]</td><td>-15.3*** [-17.7,-13.0]</td><td>-10.8*** [-12.5,-9.2]</td><td>-11.5*** [-13.9,-9.0]</td></tr><tr><td>Contact</td><td>Immunity: previously infected &lt; 6 months (NVNI)</td><td>-13.3*** [-15.4,-11.3]</td><td>-26.5*** [-28.1,-24.8]</td><td>-30.7*** [-32.1,-29.4]</td><td>-32.0*** [-33.9,-30.0]</td></tr><tr><td>Contact</td><td>Immunity: previously infected &gt; 6 months (NVNI)</td><td>-17.3*** [-19.3,-15.4]</td><td>-20.6*** [-23.3,-18.0]</td><td>-14.9*** [-16.8,-12.9]</td><td>-4.4** [-7.6,-1.2]</td></tr><tr><td>Contact</td><td>Immunity: vaccinated &lt; 6 months (NVNI)</td><td></td><td>-13.5*** [-16.0,-11.0]</td><td>-9.5*** [-10.5,-8.4]</td><td>7.0*** [5.0,8.9]</td></tr><tr><td>Contact</td><td>Immunity: vaccinated &gt; 6 months (NVNI)</td><td></td><td></td><td>0.8 [-1.6,3.3]</td><td>13.3*** [11.2,15.4]</td></tr></table>
+
+<--- Page Split --->
+<|ref|>table<|/ref|><|det|>[[31, 80, 963, 234]]<|/det|>
+<table><tr><td colspan="2"></td><td>EU1</td><td>alpha</td><td>delta</td><td>omicron</td></tr><tr><td>Contact</td><td>Immunity: hybrid (NVNI)</td><td></td><td>-20.1*** [-25.8,-14.3]</td><td>-21.7*** [-22.9,-20.6]</td><td>-18.2*** [-20.4,-16.1]</td></tr><tr><td>Contact</td><td>women (men)</td><td>0.2 [-0.4,0.8]</td><td>-1.0 [-2.1,0.1]</td><td>-0.2 [-1.0,0.6]</td><td>-0.4 [-1.5,0.7]</td></tr><tr><td>Contact</td><td>age: 0-17 (18-64)</td><td>-11.7*** [-12.5,-11.0]</td><td>-8.2*** [-9.5,-6.8]</td><td>-0.2 [-1.3,0.8]</td><td>-4.8*** [-6.2,-3.4]</td></tr><tr><td>Contact</td><td>65+ (18-64)</td><td>4.1*** [2.4,5.8]</td><td>1.2 [-1.7,4.1]</td><td>-1.4 [-3.2,0.3]</td><td>-6.5*** [-9.6,-3.4]</td></tr><tr><td>Contact</td><td>N test 30days: 0 (1)</td><td>-16.3*** [-17.4,-15.1]</td><td>-11.1*** [-12.7,-9.5]</td><td>-13.3*** [-14.6,-12.0]</td><td>-17.0*** [-18.6,-15.5]</td></tr><tr><td>Contact</td><td>N test 30days: 2+ (1)</td><td>1.3 [-1.2,3.9]</td><td>1.7 [-1.0,4.4]</td><td>3.0** [1.1,4.8]</td><td>-0.1 [-2.0,1.8]</td></tr></table>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 100, 407, 118]]<|/det|>
+## Statements & Declarations  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 126, 190, 141]]<|/det|>
+## Funding  
+
+<|ref|>text<|/ref|><|det|>[[117, 145, 803, 177]]<|/det|>
+This research was supported by the research project SELFISH, financed by the Swiss National Science Foundation, grant number 51NF40- 160590 (LIVES Center international research project call).  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 184, 296, 200]]<|/det|>
+## Competing Interests  
+
+<|ref|>text<|/ref|><|det|>[[117, 204, 510, 220]]<|/det|>
+The authors have no competing interest to declare  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 234, 291, 250]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[117, 253, 670, 269]]<|/det|>
+We thank the Geneva Directorate of Health for collecting and providing the data.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 275, 252, 290]]<|/det|>
+## Ethics approval  
+
+<|ref|>text<|/ref|><|det|>[[115, 294, 850, 326]]<|/det|>
+Research received the agreement of the Cantonal Ethic Committee of Geneva (CCER protocol 2020- 01273). Individuals who refused to share their data were removed from the analysis.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 472, 150]]<|/det|>
+adjustedsecondaryattackratesuppmatV2. pdf
+
+<--- Page Split --->

preprint/preprint__7e878f0cd9d5b59940e0554b3ad286805751e8d5ee47e9c5c90a04d0070bacd8/images_list.json

@@ -0,0 +1,77 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 Improvements in constraining subsurface coupling using earthquake focal mechanisms.",
+    "footnote": [],
+    "bbox": [
+      [
+        100,
+        133,
+        912,
+        500
+      ]
+    ],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2 Seismicity distribution along the CSAF. a Map view and b cross-section view of relocated earthquakes from 1984 to 2015 (Waldhauser and Schaff, 2008, extended to later years) along the central San Andreas Fault. Events in the box XX' are colored by faulting style. Small beach balls denote \\(M \\geq 4.0\\) earthquakes. Large beach balls denote the 2003 \\(M_{w}6.5\\) San Simeon and 2004 \\(M_{w}6.0\\) Parkfield earthquakes' focal mechanisms. White stars denote historic major earthquakes. Note vertical exaggeration \\(\\mathrm{VE} = 3.13\\) in b. The local coordinate system has its origin at latitude \\(35.867^{\\circ}\\mathrm{N}\\) , longitude \\(120.447^{\\circ}\\mathrm{W}\\) and is oriented \\(\\mathrm{N42^{\\circ}W}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        252,
+        92,
+        737,
+        660
+      ]
+    ],
+    "page_idx": 30
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 Spatiotemporal variation of seismicity and its relationship with the main fault. a Rotated map view, b cross-section view, and c spatiotemporal variations of seismicity (black dots), \\(M \\geq 4.0\\) earthquakes (blue beachballs), and repeating earthquakes (26; red dots) along the Central San Andreas Fault. The local coordinate system has its origin at latitude \\(35.867^{\\circ}\\mathrm{N}\\) , longitude \\(120.447^{\\circ}\\mathrm{W}\\) and is oriented \\(N42^{\\circ}\\mathrm{W}\\) . d Spatial variations of fault strike (red line) and dip (blue line) angles determined using PCA analysis from relocated seismicity around each 15-km-long 15-km-deep fault segment stepping at 1-km intervals along the fault. The reference fault strike and dip are \\(N138^{\\circ}\\mathrm{E}\\) and \\(90^{\\circ}\\) (vertical), respectively. Cumulative Density Functions (CDF) of the e horizontal distance of events from main fault and f azimuthal difference from main fault strike",
+    "footnote": [],
+    "bbox": [
+      [
+        87,
+        87,
+        884,
+        597
+      ]
+    ],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 Modeled and observed on-fault displacement. a Fault model 1 setup with freely slipping zones (blue), locked sections (gray) and constant-rate creeping zones (red). The deep creeping zone driving the shallow creep extends from \\(15\\mathrm{km}\\) to \\(2000\\mathrm{km}\\) depth and far beyond the lateral ends of the CSAF. The shallow fault is fully coupled beyond \\(166\\mathrm{km}\\) NW and -17km SE of the fault. The size of each fault patch is \\(3\\times 3\\mathrm{km}\\) . White stars denote \\(M\\geq 4.0\\) earthquakes. b Modeled and c observed fault creep rate estimated from the occurrence of repeating earthquakes. d Modeled and e observed fault slip direction estimated from the rake and dip of repeating earthquake focal",
+    "footnote": [],
+    "bbox": [
+      [
+        90,
+        90,
+        884,
+        660
+      ]
+    ],
+    "page_idx": 33
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 Off-fault stress field and kinematics. Comparison of a the angle between the main fault and the modeled off-fault maximum horizontal stress orientation \\((\\theta)\\) 1.5 km NE of the main fault with b the angle between the main fault and the observed maximum horizontal stress orientation calculated from \\(M\\geq 1.0\\) focal mechanisms located within 2 km around the fault trace. White dots denote the locations of focal mechanisms used in stress inversion. c The percentage of oblique-reverse-faulting events \\((P e r c_{r a k e > 0};\\) red curve) for M1.0 focal mechanisms located within 2 km around the fault trace. d Point-to-point comparison of \\(\\theta\\) in b and \\(P e r c_{r a k e > 0}\\) in c. e Schematic illustration showing the variation of fault zone structure, weakness, and stress field.",
+    "footnote": [],
+    "bbox": [
+      [
+        88,
+        90,
+        905,
+        590
+      ]
+    ],
+    "page_idx": 35
+  }
+]

preprint/preprint__7e878f0cd9d5b59940e0554b3ad286805751e8d5ee47e9c5c90a04d0070bacd8/preprint__7e878f0cd9d5b59940e0554b3ad286805751e8d5ee47e9c5c90a04d0070bacd8.mmd

@@ -0,0 +1,403 @@
+
+# 3D architecture and complex behavior along the simple central San Andreas fault  
+
+Yifang Cheng  
+
+chengyi@berkeley.edu  
+
+University of California, Berkeley https://orcid.org/0000- 0002- 6507- 5607Roland BürgmannUniversity of California, Berkeley https://orcid.org/0000- 0002- 3560- 044XRichard AllenUniversity of California Berkeley https://orcid.org/0000- 0003- 4293- 9772  
+
+## Article  
+
+# Keywords:  
+
+Posted Date: November 30th, 2023  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 3678641/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 25th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 49454- z.
+
+<--- Page Split --->
+
+# 3D architecture and complex behavior along the simple central San Andreas fault  
+
+# Authors  
+
+Yifang Cheng, \(^{1,2*}\) Roland Bürgmann, \(^{1,2}\) Richard M. Allen \(^{1,2}\)  
+
+# Affiliations  
+
+\(^{1}\) Department of Earth and Planetary Science, University of California, Berkeley, CA, USA  
+
+\(^{2}\) Berkeley Seismological Laboratory, University of California, Berkeley, Berkeley, California,  
+
+U.S.A.  
+
+\(^{*}\) Yifang Cheng Email: chengyif@berkeley.edu
+
+<--- Page Split --->
+
+## Abstract  
+
+The central San Andreas Fault (CSAF) exhibits a simple linear large- scale fault geometry, yet seismic and aseismic deformation features vary in a complex way along the fault. Here we investigate fault zone behaviors using geodetic observation, seismicity and microearthquake focal mechanisms. We employ an improved focal- mechanism characterization method using relative earthquake radiation patterns on 75,164 M \(\geq 1\) earthquakes along a 2- km- wide, 190- km- long segment of the CSAF, from 1984 to 2015. The data reveal the 3D fine- scale structure and interseismic kinematics of the CSAF. Our findings indicate that the first- order spatial variations in interseismic fault creep rate, creep direction, and the fault zone stress field can be explained by a simple fault coupling model. The inferred 3D mechanical properties of a mechanically weak and poorly coupled fault zone provide a unified understanding of the complex fine- scale kinematics, indicating distributed slip deficits facilitating small- to- moderate earthquakes, localized stress heterogeneities, and complex multi- scale ruptures along the fault. Through this detailed mapping, we aim to relate the fine- scale fault architecture to potential future faulting behavior along the CSAF.
+
+<--- Page Split --->
+
+## Introduction  
+
+Fault zones have geometric complexities and are multiscale systems consisting of a localized fault core accommodating the primary slip (on- fault), a highly fractured damage zone around the fault core accommodating a small fraction of deformation (off- fault), and the surrounding host rock \(^{1,2,3}\) . Fault zones can deform seismically, such as in large damaging earthquakes, small earthquakes including earthquake clusters and repeating events, and tectonic tremor \(^{4}\) , and aseismically, such as by transient slow slip events, steady creep, and afterslip \(^{5,6}\) . The partitioning of seismic and aseismic fault slip is usually quantified using geodetically determined fault coupling, which is defined as the ratio of the inferred slip deficit rate and the long- term slip rate, with a value of 1 corresponding to fully locked while a value of 0 indicates a freely creeping fault. The spatial distribution of fault coupling and implied slip deficits can help us better understand fault zone properties, determine the seismic potential of faults, and constrain the return period and the maximum- possible magnitude of earthquakes \(^{7,8,9}\) . Fault sections with larger kinematic coupling at depth generally correspond to lower surface creep rates observed from geodetic investigations \(^{10,11,12,13}\) , lower recurrence rates of repeating earthquakes \(^{14,15,16}\) , lower b- values \(^{17}\) , and a larger fraction of clustered events \(^{18}\) .  
+
+However, gaps remain in our understanding of the seismic and aseismic deformation of faults due to several problems. It is often assumed that all patches on a fault slip in the same direction without considering the variation of slip directions that we can expect in areas with abrupt changes of fault geometry or creep rate \(^{19}\) . Secondly, analyses of seismicity often focus on statistical parameters estimated from earthquake occurrences instead of the underlying physical properties of fault zones. Moreover, the effects of distributed fault- zone deformation and on- fault/off- fault interactions are usually ignored, although on- fault and off- fault deformations are tightly interlinked and coevolve with strong feedback loops over multiple spatial and temporal scales \(^{20}\) (Fig. 1a). In this work, we analyze the distribution of
+
+<--- Page Split --->
+
+fault orientations and slip directions in the fault zone from a comprehensive catalog of earthquake focal mechanisms. This helps to resolve fine- scale on- fault slip directions, provides physical parameters for seismicity analysis, and allows for differentiating on- and off- fault seismic deformation (Fig. 1b).  
+
+The 190- km- long central segment of the San Andreas Fault (CSAF) in California (Fig. 2) offers a unique opportunity to investigate the fine- scale complexities of seismic and aseismic fault zone processes and their interconnected nature. The CSAF is characterized by a long- term slip rate of \(\sim 33 - 35\) mm/yr and varying surface creep at \(\leq 30\) mm/yr<sup>21,22</sup>. It lacks large historical earthquakes, while \(M_{w} \sim 7 - 8\) earthquakes repeatedly occur to the north and south (e.g., the 1906 San Francisco earthquake and the 1857 Fort Tejon earthquake; Fig. 2a and Supplementary Fig. 1). Understanding the potential for large earthquakes along the CSAF is important for seismic hazard assessment of the SAF and other large creeping fault zones in the world<sup>23</sup>. The CSAF also provides a natural laboratory for studies of seismic and aseismic fault zone processes because of the availability of a dense, continuous monitoring network. Decades of monitoring show that the CSAF exhibits heterogeneous seismic and aseismic deformation patterns and variable fault coupling inferred from varying along- fault surface displacements<sup>11,12,13,21</sup>, the occurrence of moderate earthquakes (e.g. 1922, 1934, 1966, and 2004 M6.0 Parkfield earthquake), abundant small earthquakes<sup>24</sup>, and repeating earthquakes whose recurrence rate is proportional to the fault creep rate at depth<sup>25,26,27,28</sup>. In order to obtain focal mechanisms of the above- mentioned small earthquakes, we use a recently developed relative focal mechanism calculation algorithm<sup>29</sup> and obtained high- quality focal mechanisms (uncertainty \(< 35\) degree) of \(\sim 80\%\) of the \(M_{l} \geq 1\) catalog events<sup>24</sup>. The diverse fault deformation patterns, abundant earthquakes, and their focal mechanisms provide a great opportunity to comprehensively characterize and understand multiscale fault zone deformation processes along the CSAF and their connections (Figs. 1 and 2).
+
+<--- Page Split --->
+
+## Results  
+
+## Overview of seismicity data  
+
+In this study, we integrate the location, time, magnitude, and focal mechanism of \(75,164 M_{l} \geq 1.0\) earthquakes, \(145 M_{l} \geq 4.0\) earthquakes, and 355 repeating earthquakes from 1980 to 2015 to place high- resolution constraints on the fault structure of the CSAF (Fig. 3). Most \(M_{l} \geq 4.0\) sequences are located along the San Juan Bautista (SJB) and Parkfield (PK) transition zones and a few \(M_{l} \geq 4.0\) earthquakes are located near Bitterwater (BW). In contrast, repeating earthquakes occur more frequently between Melendy Ranch (MR) and the San Andreas Fault Observatory at Depth (SAFOD), implying high aseismic creep rates. Since most repeating earthquakes are concentrated in a narrow zone and appear to delineate the major fault strands, we first use repeating earthquakes to obtain the horizontal location of major fault strands and then use both repeating earthquakes and the surrounding \(M_{l} \geq 1.0\) earthquakes within 1 km epicentral distance (Fig. 3a- b) to estimate the location and the orientation of the main fault using principal component analysis \(^{30}\) (see Methods section). The whole fault is nearly vertical with strike angles varying between N120°E to N140°E (Fig. 3d). Our final catalog includes 99.7% of the repeating earthquake sequences and 97.8% of \(M_{l} \geq 4.0\) earthquakes, as well as 83.4% of \(M_{l} \geq 1.0\) events located within 1km NE from the main fault strand (Fig. 3e and Supplementary Fig. 2). To better quantify the fault zone structure, we pick the focal mechanism nodal plane closer to the main fault orientation and calculate the azimuthal difference between the nodal plane and the main fault. Note that this calculation assumes that the nodal plane closer to the main fault orientation is the real fault plane and ignores left- lateral faults at high angles to the main fault. For these near- fault earthquakes, we obtain the percentages of \(M_{l} \geq 4.0\) earthquakes, repeating earthquake sequences, and \(M_{l} \geq 1.0\) earthquakes with less than 20° azimuthal difference from the main fault, which are 92.9%, 94.1%, and 83.3%, respectively (Fig. 3f). Therefore, in the following analysis we assume that \(M_{l} \geq 4.0\) earthquakes and
+
+<--- Page Split --->
+
+repeating earthquake sequences are located on the main fault (Supplementary Fig. 3), while \(M_{l} \geq 1.0\) earthquakes have variable orientations and are located both on- and off- fault. To focus on the interseismic period and disregard the co- and early postseismic deformation due to the 1989 M6.9 Loma Prieta and 2004 \(M_{w}6.0\) Parkfield earthquakes, we exclude earthquakes within the first 3 years after these events when calculating the occurrence rates of repeating earthquakes and focal mechanism properties.  
+
+## Fault model setup  
+
+The abundant small earthquake focal mechanisms provide a great opportunity to determine the fine- scale fault slip directions, which can be directly compared with the output of kinematic fault models of the partially coupled CSAF employing shear- stress- free boundary conditions on the fault. As shown in Fig. 1c and 1d, if there is a locked fault patch at depth, the fault slip rate decreases slightly around the locked patch, and the surrounding slip directions exhibit opposite rotations near the two ends of the locked patch, providing a powerful constraint on the location and size of the locked patches at depth. Therefore, in this study, we forward model the interseismic fault deformation of the CSAF and compare the obtained slip kinematics with the fault zone properties estimated from seismicity.  
+
+Based on the definition of fault coupling and previous observations, most moderate- to- large earthquakes are located in high- coupling areas while most repeating earthquakes are interpreted as small patches that repeatedly break and are surrounded and driven to failure by aseismically slipping sections of the fault (low- coupling areas). Therefore, based on the spatial distribution of \(M_{l} \geq 4.0\) events and repeating earthquakes (Fig. 3a), we design a simple forward model that consists of several locked patches near SJB, BW, and PK (in areas of low earthquake- density and near the \(M_{l} \geq 4.0\) earthquakes; Fig. 3b) and an otherwise freely slipping shallow fault in the top 15 km depth driven by a buried fault plane with a 34 mm/yr interseismic deep creep rate \(^{22}\) beneath it (model A in Fig 4a). The fault is bounded by fully locked fault segments at the two ends, corresponding to the locked sections of the SAF
+
+<--- Page Split --->
+
+that hosted the 1857 M7.9 Fort Tejon to the SE as well as the 1906 M7.8 San Francisco and 1989 M6.9 Loma Prieta earthquakes to the NW. We then compute the fault displacement rate and slip directions on the freely slipping patches using a boundary element method (Poly3D; see Methods section) \(^{31}\) .  
+
+## Observed and modeled along-fault displacements  
+
+We compare the forward modeling results with multiple geophysical observations. Besides the traditionally used surface creep rate (red curve and green symbols in Fig. 4f) \(^{11,21,32}\) , we also estimate the 2D variations of creep rate and creep direction on the NE side of the main fault using the occurrence rate and the slip direction of repeating earthquakes, respectively (see Methods section; Supplementary Figs. 3- 5). Fig. 4 shows the comparison of the modeled and observed creep rate using the occurrence rate of repeating earthquakes (Fig. 4b- c), the modeled and observed creep direction using the focal mechanisms of repeating earthquakes (Fig. 4d- e), and the modeled and observed surface creep rate using creepmeter, alignment array and InSAR data (Fig. 4f) \(^{11,21,32}\) . The modeled results are overall consistent with the observed subsurface fault creep rate variations, fault slip directions at depth, and surface creep rates with correlation coefficients of 0.47 (Fig. 4b- c), 0.55 (Fig. 4d- e), and 0.91 (Fig. 4f), respectively (Supplementary Fig. 6). The presence of locked fault patches results in gradually decreasing creep rates around the patches. For example, the observed creep rates from repeating earthquakes and measured surface creep rates are lower than 25 mm/yr and 15 mm/yr, respectively, near the SJB and PK transition zones (Fig. 4b, 4c, 4f). In the central creeping section from MR to the SAFOD, the modeled and observed creep rates approach the deep creep rate ( \(\sim 34 \mathrm{mm / yr}\) ). The repeater- derived creep directions show a more heterogeneous distribution around the locked patches with an upward creep to the NW of the patch and downward creep to the SE of the patch (Fig. 4d- e), providing valuable additional constraints on fault coupling at depth. For example, there is no clear evidence of the existence of locked fault patches near BW based on just the variation of creep rates (Fig. 4f). In contrast, the abrupt changes
+
+<--- Page Split --->
+
+of the vertical creep component (Fig. 4d-e) and the observed \(M_{l} \geq 4.0\) sequences near BW (Fig. 4a) indicate considerable stress accumulation and the existence of a deep locked fault patch near BW.  
+
+## Off-fault structure and kinematics  
+
+In addition to on- fault displacements, there are also many small earthquakes located around the fault and most of them are located to the NE of the fault trace (Fig. 3 and Supplementary Fig. 2). Therefore, we obtain the along- fault cross- sectional distribution of the modeled azimuthal differences \((\theta)\) between the off- fault maximum horizontal stress orientation \((SH_{\mathrm{max}})\) and the main fault 1.5 km NE of the main fault plane (see Methods section; Fig. 5a). The correlation coefficient between the modeled off- fault stress distribution and the observed stress field estimated using \(M_{l} \geq 1.0\) focal mechanisms located within 2 km of the main fault (Fig. 5b) is 0.33 (Supplementary Fig. 6d), which is considerable considering the effects of background tectonic stress and the strong stress variations across the fault. Overall, the NW part of the fault shows low \(\theta\) value, and the SE part has high \(\theta\) value. However, both modeled and observed variations of stress orientation near BW exhibit an opposite pattern with high \(\theta\) value to the NW of BW and low \(\theta\) value to the SE of BW, consistent with the existence of a large deep locked patch near the BW. The consistency between the observed stress field and the modeled off- fault stress field variations suggests that on- fault coupling heterogeneity can cause significant stress perturbation in the surrounding area.  
+
+We further investigate the off- fault structure and kinematics by calculating the cross- section distribution of the percentage of oblique- reverse faulting events \((P e r c_{r a k e > 0})\) (Fig. 5c). There is a significant negative correlation between with \(P e r c_{r a k e > 0}\) with a correlation coefficient of - 0.43 When \(SH_{\mathrm{max}}\) is oriented at a high angle to the main fault, most events are oblique- normal faulting events with optimal fault orientations at a high angle to the main fault (case 1 in Fig. 5d- e). When \(SH_{\mathrm{max}}\) is at about 45 degrees, there are a comparable number of oblique- normal and oblique- reverse faulting events,
+
+<--- Page Split --->
+
+suggesting strike- slip faulting is the preferred faulting style (case 2 in Fig. 5d- e). When \(SH_{\mathrm{max}}\) in the fault zone is at a low angle to the main fault, most earthquakes are oblique- reverse faulting events with optimal fault orientations showing high angles to the main fault (case 3 in Fig. 5d- e). In all cases, small- scale faults tend to have horizontal slip directions parallel to the main fault orientation and some of their fault strike orientations exhibit a high angle to the main- fault orientation.  
+
+## Discussion  
+
+To evaluate the proposed fault model, we also perform modeling using a simple creeping fault model without the small, locked patches but with fully locked segments on the two ends of the CSAF (model B, Supplementary Fig. 9a) and compared the modeling results using model B with the observations (Supplementary Fig. 10). Without the locked patches on the CSAF, the modeled creep rate is generally higher than the observed creep rate but the modeled and observed creep rates still show a first- order correlation (Supplementary Fig. 10a, 10c). In contrast, there is almost no correlation between the observed and modeled creep directions (Supplementary Fig. 10b, 10d), suggesting that the vertical fault slip component is highly sensitive to the small- scale fault coupling heterogeneity and can help to constrain the fault coupling model at depth. The solved fine- scale fault zone properties and fault coupling model provide great opportunity to improve our understanding of the physical mechanisms of small, locked asperities that were not noticed before.  
+
+There are some misfits between the modeled and observed fault zone properties due to the uncertainty of input data and the simplicity of the first- order fault coupling model. The range of modeled creep directions is \(\pm 6\) degrees from the horizontal direction and is much smaller than that of the creep directions observed from repeating earthquake focal mechanisms \((\pm 30\) degrees). The different ranges of modeled and observed creep directions might be caused by local variations of the fault dipping angle (Supplementary Figs. 8, 11), the considerable uncertainties of small- earthquake focal mechanism
+
+<--- Page Split --->
+
+solutions (Hardebeck and Shearer, 2002; \~20 degree in this study), and fault coupling heterogeneities smaller than the modeled \(3 \times 3\) km patch size (Supplementary Fig. 5). In contrast to the fault creep direction, the range of the modeled \(\theta\) is 5- 85 degrees is larger than the observed stress orientation (20- 70 degrees). This is because the inverted stress field is obtained from focal mechanisms of a large number of earthquakes (> 100) along a given fault patch including many on- fault earthquake focal mechanisms, compared to only a few (<10) repeating earthquake sequences in each fault patch, which is averaged over a long period and is less sensitive to focal mechanism uncertainties and local fault zone heterogeneities. Moreover, the observed surface creep rate has shown to be different from the modeled surface creep rate between MR and BW, which might be due to the temporal oscillation of surface creep rate, the simple first- order fault coupling model, etc. Better fitting might be achieved using a more advanced adjoint inversion method, but that is beyond the scope of this paper.  
+
+The creep rate distribution obtained from our first- order fault coupling model is overall consistent with those inverted from geodetic observations (Fig. S12) \(^{12,13}\) , with low fault coupling on the central creeping section between 25 and 100 km along the fault and relatively high fault coupling on the other part of the fault, but with better- constrained fault slip distributions at depth. One interesting feature is that many of these models show a deep low- coupling area in the central creeping section but with somewhat different locations. In this study, we constrain the location of the deep locked patch near BW using additional observations from earthquakes at depth, including \(M_{l} \geq 4.0\) earthquakes (Figs. 3, 4a), reduced estimated creep rates (Fig. 4c), as well as abrupt changes of creep direction (Fig. 4e), off- fault stress orientation (Fig. 5b), and earthquake faulting style (Figs. 2, 4c). Based on the modeling result, the differential creep rate between the locked patch near BW and the surrounding high- creep- rate patches is much higher compared with the other locked patches in the SJB and PK transition zones, suggesting a higher rate of elastic strain accumulation. However, there are only 2 instrumentally observed \(M_{l} \geq 4.0\)
+
+<--- Page Split --->
+
+\((M_{w}4.1\) and \(M_{w}5.3)\) earthquakes near the BW locked patch in the past 100 years compared with other locked patches near MR and PK. In contrast, there were six \(M_{w} > 5.5\) earthquakes near BW between the 1857 Fort Tejon and the 1906 San Francisco earthquakes (Supplementary Fig. 1) \(^{33}\) . The infrequent occurrences of large- magnitude earthquakes near the BW locked patch in the past century might be caused by the changes of absolute stress level and the recurrence times of moderate- to- large magnitude events caused by the occurrence of the 1857 Fort Tejon and 1906 San Francisco earthquakes \(^{34}\) . Similar changes of the occurrence rate of large- magnitude earthquakes before and after the 1857 Fort Tejon and 1906 San Francisco earthquakes are also seen along other sections of the fault (Supplementary Fig. 1) \(^{33}\) . Since the main fault orientation is simple in this area (Fig. 3a, 3d), the significant variations of fault locking may be mainly due to the variations of material properties, such as lithology \(^{37,38,39}\) , local temperature anomalies \(^{35}\) , and pore fluid pressure \(^{40}\) .  
+
+The seismic potential inferred from the fault coupling model along the CSAF is of interest from a hazard perspective because the CSAF accommodates nearly all the plate motion in this part of California. Assuming a deep slip rate of \(34 \mathrm{mm} / \mathrm{yr}^{21}\) , the modeled moment deficit rate is \(1.32 \times 10^{18} \mathrm{N} \cdot \mathrm{m} / \mathrm{yr}\) , while \(1.48 \times 10^{18} \mathrm{N} \cdot \mathrm{m} / \gamma \mathrm{r}\) are released by aseismic slip. The stored moment over a 150- year period is equivalent to a moment magnitude \((M_{w})\) 7.5 earthquake which agrees with previous studies \(^{12,13}\) . If we assume that the areas of partial coupling catch up the moment deficit only by aseismic slip, such as by accelerated afterslip following earthquakes on the SAF, the modeled moment deficit rate is \(4.04 \times 10^{17} \mathrm{N} \cdot \mathrm{m} / \gamma \mathrm{r}\) and the 150- year accumulated moment is equivalent to a \(M_{w}7.2\) earthquake. Since the locked patches are distributed along the creeping section with low slip deficit around them, these small locked patches are more likely to rupture independently as moderate earthquakes (e.g., the 2004 Parkfield earthquake) with a substantial amount of slip deficit on the surrounding fault being taken up by transient afterslip of major ruptures (e.g., following the 1906 Great San Francisco, 1989 Loma
+
+<--- Page Split --->
+
+Prieta, and 2004 Parkfield earthquakes \(^{15,27}\) . An important but controversial question is whether major ruptures can dynamically penetrate deep into the central creeping section \(^{5,41,42}\) . The deep locked patch near BW may weaken the barrier effect of the central creeping section and increase the possibility for major earthquakes to rupture through the whole creeping section. However, the stressing rate would be required to be high enough to penetrate multiple creeping sections between the small, locked patches along the CSAF. Note that while we can't fully rule out this scenario, there is no direct evidence that great ruptures have made it across the CSAF creeping section \(^{42}\) . Therefore, a substantial portion of the accumulated moment deficit is more likely to be released by the afterslip of surrounding major earthquakes and moderate- magnitude earthquakes on the CSAF.  
+
+The fine- scale fault coupling heterogeneity can also affect the large- scale stress distribution by perturbing the stress field near the fault (Fig. 4a- b). The rotation of \(SH_{max}\) from high angles away from the main fault to low angles in the vicinity of the SAF has been observed and studied for decades \(^{43,44,45}\) . Rice \(^{46}\) proposed a high pore- pressure fracture zone model that can decrease the effective fault- normal compressive stress and cause stress rotation. However, this can only be applied to narrow, high pore- pressure fault zones \(^{47,48}\) . Some studies explain the stress rotation as the combined effect of a weak fault and finite- width weak lower crust \(^{47,49}\) , which requires strong lateral variations of lower crustal properties. Scholz \(^{48}\) suggested that the fault- parallel shear stress decreases with distance from the SAF due to the frictional resistance to strike- slip motion under a strong- fault hypothesis, which conflicts with the absence of localized shear heating near the main fault \(^{50,51}\) . In this study, we found that even though the whole CSAF is rapidly creeping with limited accumulated shear stress overall, fine- scale fault coupling heterogeneity can still cause substantial stress perturbations near the fault. Since the along- fault strike- slip motion and the fine- scale heterogeneous fault zone deformation generally co- exist along the
+
+<--- Page Split --->
+
+whole SAF, this provides an additional explanation to the observed stress rotation near the main fault that is not related to the strong- or weak- fault hypotheses.  
+
+In addition to stress orientation, the fine- scale fault zone coupling heterogeneity can also cause local stress concentrations around the major fault, facilitating multi- scale earthquake ruptures near the major fault and promoting the evolution of fault zone architecture. When earthquake source dimensions significantly exceed the fault zone width, their ruptures tend to align with the main fault orientation, showing consistent focal mechanisms ( \(M_{t} \geq 4\) earthquakes in Fig. 3e- f). When earthquake source dimensions are smaller than or comparable to the fault zone width, earthquakes in the fault zone exhibit more diverse focal mechanisms. However, they still tend to slip in a fault- parallel direction with a notable portion of secondary faults showing high angle to the main fault ( \(M_{t}1 - 4\) earthquakes in Figs. 3e- f, 5d- e). These small high- angle faults might be related to phenomena such as microfracturing near the fault tips, relay breaching and splay- faulting, promoting the weakening of rock adjacent to the developing fault \(^{52}\) . The weakened rock in the fault zone potentially promotes the inclination of small subsidiary fault to slip along a fault- parallel direction, the linkage of fault segments, and the occurrences of moderate- and- large earthquakes along the major fault segments (Figs. 2- 3) \(^{52}\) . The narrow weak fault zone can also cause localized stress and slip concentrations, which may lead to a zone of localized stress rotation in the vicinity of the SAF. Since the narrow weak fault zone is much easier to deform than the surrounding strong host rock, moderate and large magnitude earthquakes along the CSAF are more likely to be right- lateral ruptures along the main fault orientation instead of complex ruptures that are frequently observed in wide, diffuse, and immature fault zones, like the 2019 \(M_{w}7.1\) Ridgecrest earthquake \(^{53}\) and the 2016 \(M_{w}7.8\) Kaikoura earthquake \(^{54}\) .  
+
+One important thing to note is that neither the above- mentioned fault coupling distribution, the large- scale stress rotations, nor the relative weakness of the narrow fault zone indicate the absolute level
+
+<--- Page Split --->
+
+of fault strength. Both frictional strength and fault coupling can be affected by variations of the stress field on the fault and variations of material properties, such as lithology \(^{37,38,39}\) , temperature \(^{35}\) , and pore fluid pressure \(^{55}\) . However, there is no clear correlation between fault coupling and fault strength \(^{56,57}\) . The variation of fault geometry and surface roughness \(^{58}\) can also significantly affect the fault strength. One possible way to estimate the strength of the CSAF might be the direct estimation of absolute stress levels before moderate earthquake ruptures along the CSAF based on changes of small earthquake focal mechanisms \(^{29}\) . For example, we can search over possible absolute stress levels, model the stress rotation in the fault zone before and after the 2004 Parkfield earthquake using finite source slip models \(^{59,60}\) , and compare the results with the focal mechanism observations \(^{61}\) .  
+
+In summary, we use high- resolution locations and focal mechanisms of repeating earthquakes, \(M_{l} \geq 4\) earthquakes, and \(M_{l}1 - 4\) earthquakes to illuminate fine- scale on- fault creep rates and directions as well as the stress field, structure, and earthquake slip variations in the fault zone. Our results reveal closely connected on- fault and off- fault deformation processes. All observed fine- scale kinematic features can be reconciled with a simple fault coupling model, inferred to be surrounded by a narrow, mechanically weak zone. Our study demonstrates the value of integrating small- earthquake focal mechanisms into fault zone analysis, to better resolve the detailed fault orientations, slip directions, and stress field variations associated with various aseismic and seismic processes. The resolved fault coupling heterogeneity, the surrounding narrow weak fault zone with complex internal structure, as well as their interactions have important implications for revealing multi- scale fault zone deformation, understanding large- scale stress field variation, and estimating the slip budget, timing, and patterns of future major earthquakes. The analyses performed in this study can be applied to other transform and subduction fault zones around the world for better understanding of multi- scale fault geometry and kinematics and improved estimation of future seismic hazard.
+
+<--- Page Split --->
+
+## Methods  
+
+## Fault slip modeling  
+
+The variation of fault displacement and stress around the fault depend on the mechanical response of the fault to the shear stress loading from the lower crust. Therefore, we used the boundary- element code Poly3D<sup>31</sup> to simulate fault slip processes during the inter- seismic period. The fault lies in a uniform elastic half- space with a Poisson's ratio of 0.25 and shear modulus of 30 GPa. It is vertical both in and below the seismogenic layer with 3×3 km patch size. In the top 15 km depth. Only zero- slip or zero- shear- stress boundary conditions on patches of the shallow fault are defined, rather than frictional properties. The fault is mostly freely slipping between - 17 km and 166 km with several small frictionally locked patches and is fully coupled beyond 166 km NW and - 17km SE of the fault (model A; Fig. 4a). Between 15 km and 2000 km depth, the fault slips steadily with a 34 mm/yr inter- seismic slip rate between - 500 km to 500 km horizontal distance along the fault. To evaluate the effect of the small, locked fault patched, we also consider a model without small frictionally locked patches between - 17 km and 166 km along the fault but with a fully creeping fault bounded by locked segments at the two ends of the fault (model B; Supplementary Figs. 9, 10).  
+
+## \(M_{l}\geq 1\) earthquake focal mechanism calculation  
+
+In this study, we calculate the focal mechanisms along the CSAF using the REFOC algorithm<sup>29</sup>, which uses first- motion polarities, S- /P- wave amplitude ratios to obtain the initial earthquake focal mechanisms and further constrain them using the P- wave amplitude ratios and S- wave amplitude ratios of closely located earthquakes within 3 km hypocentral distance. We utilize the polarities, P- and S- wave phases manually picked by data analysts, and the relocated earthquake catalog archived by the Northern California Earthquake Data Center (NCEDC). We use the same P- wave velocity models<sup>29</sup> and the S- wave velocity models are derived from P- wave velocity models by assuming that the P- /S- wave
+
+<--- Page Split --->
+
+velocity ratio equals 1.732. We apply a 1- 10 Hz band- pass- filter to earthquake waveform data from NCEDC, choose 0.5s before to 1.5s after P- and S- wave arrivals as the signal windows and 2.5s to 0.5s before P- wave arrivals as noise windows, and take the difference between maximum and minimum amplitude values in each time window to be the estimated signal and noise amplitudes. If the time difference between P- and S- wave arrivals is larger than 2s and the signal- to- noise ratio (SNR) is larger than 3, we use the P- and S- wave amplitudes to obtain S-/P- wave amplitude ratios for each individual event as well as the inter- event P- wave amplitude ratios and S- wave amplitude ratios. We obtained 52,211 out of 65,492 ( \(\sim 80\%\) ) \(M_{l} \geq 1\) earthquakes with at least 8 polarities and focal mechanism uncertainties less than 35 degree. The catalog is available via (https://data.mendeley.com/datasets/34szj3jm6k/1).  
+
+## Repeating earthquake focal mechanism calculation  
+
+Since repeating earthquakes share highly similar waveforms and locations, they also share similar rupture processes and earthquake focal mechanisms. Therefore, we can utilize the similarity of repeating earthquakes to better constrain the focal mechanism of each repeating earthquake sequence. For each repeating earthquake sequence, we obtain available first- motion polarities and S-/P- wave amplitude ratios from all earthquakes in the sequence at each station, calculate the median values of polarities and S-/P- wave amplitude ratios, and assign the values to the station. Then we use the median polarities and S-/P- wave amplitude ratios to calculate the focal mechanism of each repeating earthquake sequence using the HASH algorithm<sup>62</sup>. By doing this process, we can both reduce the errors caused by manual picking and temporal noises in the waveforms and make full use of all available stations in the study time period for focal mechanism calculation. We obtained focal mechanisms of 386 repeating earthquake sequences and the catalog is available via (https://data.mendeley.com/datasets/34szj3jm6k/1).
+
+<--- Page Split --->
+
+## Fault geometry estimation using repeating earthquake and seismicity  
+
+In this study, we determine the main fault geometry (representative strike and dip and average fault- zone width of each segment) using earthquake locations at depth. Since small earthquakes may occur around the main fault instead of on- fault, we first determine the horizontal location of the primary fault strand using the locations of repeating earthquake sequences with at least 2 repeaters (red dots in Fig. 3a- b) \(^{28}\) , which generally indicate localized aseismic slip of major faults. We then choose earthquakes within 1- km epicentral distance from the horizontal fault trace to determine the 3D fault geometry by applying the principal component analysis \(^{30}\) to minimize the orthogonal hypocentral distances to the fitted fault plane in each fault segment. For each 15- km- long 15- km- deep fault segment stepping at 1- km intervals along the fault trace, we calculate strike and dip of the plane that minimizes the distance between earthquake hypocenters and the plane and assign the values to the center of the fault segment (Fig. 3d).  
+
+## Fault creep rate estimation using repeating earthquakes  
+
+Repeating earthquakes are events that repeatedly rupture particular fault patches \(^{16}\) , which can be detected by waveform similarity \(^{25,63}\) and can be used to illuminate the spatiotemporal variations of fault creep rate at depth \(^{26,64}\) . Here, we use the repeating earthquake sequences with more than 10 repeaters from a Northern California repeating earthquake catalog \(^{28}\) to estimate the creep rate variations along the CSAF.  
+
+The cumulative fault slip of and surrounding an earthquake patch over one seismic cycle in a repeating earthquake sequence can be estimated following the empirical scaling relationship  
+
+\[d = 10^{\alpha}M_{0}^{\beta},\]  
+
+where \(d\) is slip in centimeter and \(M_{0}\) is seismic moment in dyne·cm, converted from the NCSN preferred magnitude \(M_{l}\) using the empirical relationship \(\log (M_{0}) = 1.6M_{l} + 15.8^{65}\) . The empirical values and are
+
+<--- Page Split --->
+
+\(\alpha = - 2.36\) and \(\beta = 0.16\) based on comparison with the geodetically inferred creep rate at Parkfield<sup>26</sup>.  
+
+To compare with the modeling results, we obtain the averaged slip rate of all repeating earthquakes in each \(3 \times 3 \mathrm{~km}\) fault patch in Fig. 4a and assign the value as the slip rate of the fault patch (Fig. 3c,  
+
+Supplementary Figs. 4a, 5a). If the \(3 \times 3 \mathrm{~km}\) fault patch lacks repeating earthquake focal mechanisms, we will broaden our search to include a \(9 \times 9 \mathrm{~km}\) fault area centered around the same location.  
+
+## Local fault slip direction estimation using repeating earthquake focal mechanisms  
+
+Since repeating earthquakes are closely located around the main fault with strike orientation highly consistent with the main fault (Fig. 2), we assume that most repeating earthquakes are located on the main fault and represent the local fault slip behaviors. We first obtain the nodal plane whose strike angle has a smaller azimuthal difference from the main fault orientation. We then combine the dip and rake angles of the best- fitting nodal plane to estimate the slip direction of each repeating earthquake sequence (Supplementary Figs. 3, 4). Since we use the NE side of the main fault as the reference and the rake angle represent the moving direction of the hanging wall, we treat the rake direction as the slip direction when the nodal plane is dipping to the NE and use the opposite of the rake direction as slip direction when the nodal plane is dipping to the SW (Supplementary Fig. 4). To compare with the modeling results, we obtain the averaged slip direction of all repeating earthquakes in each \(3 \times 3 \mathrm{~km}\) fault patch in Fig. 4a and assign the value as the slip direction of the fault patch (Fig. 3e, Supplementary Fig. 5). If the \(3 \times 3 \mathrm{~km}\) fault patch lacks repeating earthquake focal mechanisms, we will broaden our search to include a \(9 \times 9 \mathrm{~km}\) fault area centered around the same location.  
+
+## Estimations of stress orientation \((SH_{\mathrm{max}})\) and faulting style \((Perc_{rake > 0})\) using \(M_{l} \geq 1\) focal mechanisms  
+
+Focal mechanisms contain valuable information about fault geometry, kinematics and stress state in the crust. We use \(24,915 M_{l} \geq 1\) focal mechanisms located within \(2 \mathrm{~km}\) from the main fault to
+
+<--- Page Split --->
+
+estimate stress orientation \(SH_{\mathrm{max}}\) and fault style \(Perc_{rake > 0}\) . We only use focal mechanisms with more than 8 polarities and \(< 35\) degrees uncertainties for quality control. To compare with the modeling results, we obtain these values in each \(3 \times 3 \mathrm{~km}\) fault patch in Fig. 4a when there are more than 100 focal mechanisms in the grid (Fig. 5b). If the \(3 \times 3 \mathrm{~km}\) fault patch lacks sufficient focal mechanisms, we will broaden our search to include a \(9 \times 9 \mathrm{~km}\) fault area centered around the same location.  
+
+For stress orientation estimation, we iteratively inverse stress using the STRESSINVERSE program<sup>66</sup>. This method is modified from Michael's method (1987) that jointly inverse stress and fault orientations by selecting the nodal plane with higher value of instability \(I^{67}\) . For earthquake faulting style, it is usually classified into normal, reverse, or strike- slip earthquakes based on rake angles<sup>68</sup>. Here, in order to represent the faulting style of a group of focal mechanisms, we simplify the classification into two types: oblique- reverse- faulting events with rake angle larger than \(0^{\circ}\) and oblique- normal- faulting events with rake angle equal to or smaller than \(0^{\circ}\) so that the summation of the percentages of oblique- normal- faulting and oblique- reverse- faulting events equals \(100\%\) . In this study, we use the percentage of oblique- reverse- faulting events \((Perc_{rake > 0})\) to represent the earthquake faulting style.  
+
+## Data and materials availability:  
+
+The InSAR surface slip- rate estimates were obtained from Jolivet et al., (2015)<sup>10</sup>. The repeating earthquake catalog was obtained from Waldhauser and Schaff (2021)<sup>28</sup>. The relocated earthquake catalog was obtained from Waldhauser and Schaff (2008)<sup>24</sup>. The earthquake phase information and seismic waveforms are taken from the Northern California Earthquake Data Center, Northern California Seismic Network (https://ncedc.org/ncsn/). The estimated earthquake focal mechanisms are available at Mendeley Data (https://data.mendeley.com/datasets/34szj3jm6k/1).
+
+<--- Page Split --->
+
+## References  
+
+1. Chester, F. M. & Chester, J. S. Ultracataclasite structure and friction processes of the Punchbowl fault, San Andreas system, California. Tectonophysics, 295(1-2), 199-221 (1998).  
+
+2. Ben-Zion, Y. & Sammis, C. G. Characterization of fault zones. Pure and applied geophysics, 160(3), 677-715 (2003).  
+
+3. Mitchell, T. M. & Faulkner, D. R. The nature and origin of off-fault damage surrounding strike-slip fault zones with a wide range of displacements: A field study from the Atacama fault system, northern Chile. Journal of Structural Geology, 31(8), 802-816, (2009).  
+
+4. Peng, Z. & Gomberg, J. An integrated perspective of the continuum between earthquakes and slow-slip phenomena. Nature geoscience, 3(9), 599-607 (2010).  
+
+5. Harris, R. A. Large earthquakes and creeping faults. Reviews of Geophysics, 55(1), 169-198 (2017).  
+
+6. Bürgmann, R. The geophysics, geology and mechanics of slow fault slip. Earth and Planetary Science Letters, 495, 112-134 (2018).  
+
+7. Perfettini, H., Avouac, J. P., Tavera, H., Kositsky, A., Nocquet, J. M., Bondoux, F., Chlieh, M., Sladen, A., Audin, L., Farber, D. L. & Soler, P. Seismic and aseismic slip on the Central Peru megathrust. Nature, 465(7294), 78-81 (2010).  
+
+8. Chlieh, M., Mothes, P. A., Nocquet, J. M., Jarrin, P., Charvis, P., Cisneros, D., Font, Y., Collot, J. Y., Villegas-Lanza, J. C., Rolandone, F., Vallée, M., Regnier, M., Segovia M., Martin X. & Yepes, H. Distribution of discrete seismic asperities and aseismic slip along the Ecuadorian megathrust. Earth and Planetary Science Letters, 400, 292-301 (2014).  
+
+9. Avouac, J. P. From geodetic imaging of seismic and aseismic fault slip to dynamic modeling of the seismic cycle. Annual Review of Earth and Planetary Sciences, 43, 233-271 (2015).
+
+<--- Page Split --->
+
+10. Malservisi, R., Gans, C. & Furlong, K. P. Numerical modeling of strike-slip creeping faults and implications for the Hayward fault, California. Tectonophysics, 361(1-2), 121-137 (2003).  
+
+11. Jolivet, R., Simons, M., Agram, P. S., Duputel, Z. & Shen, Z. K. Aseismic slip and seismogenic coupling along the central San Andreas Fault. Geophysical Research Letters, 42(2), 297-306 (2015).  
+
+12. Maurer, J. & Johnson, K. Fault coupling and potential for earthquakes on the creeping section of the central San Andreas Fault. Journal of Geophysical Research: Solid Earth, 119(5), 4414-4428 (2014).  
+
+13. Ryder, I. & Bürgmann, R. Spatial variations in slip deficit on the central San Andreas Fault from InSAR. Geophysical Journal International, 175(3), 837-852 (2008).  
+
+14. Malservisi, R., Furlong, K. P. & Gans, C. R. Microseismicity and creeping faults: Hints from modeling the Hayward fault, California (USA). Earth and Planetary Science Letters, 234(3-4), 421-435 (2005).  
+
+15. Khoshmanesh, M., Shirzaei, M. & Nadeau, R. M. Time-dependent model of aseismic slip on the central San Andreas Fault from InSAR time series and repeating earthquakes. Journal of Geophysical Research: Solid Earth, 120(9), 6658-6679 (2015).  
+
+16. Uchida, N. & Bürgmann, R. Repeating earthquakes. Annual Review of Earth and Planetary Sciences, 47(1), 305-332 (2019).  
+
+17. Tormann, T., Wiemer, S., Metzger, S., Michael, A. & Hardebeck, J. L. Size distribution of Parkfield's microearthquakes reflects changes in surface creep rate. Geophysical Journal International, 193(3), 1474-1478 (2013).  
+
+18. Liu, Y. K., Ross, Z. E., Cochran, E. S. & Lapusta, N. A unified perspective of seismicity and fault coupling along the San Andreas Fault. Science advances, 8(8), eabk1167 (2022).
+
+<--- Page Split --->
+
+19. Bürgmann, R., Fielding, E. & Sukhatme, J. Slip along the Hayward fault, California, estimated from space-based synthetic aperture radar interferometry. Geology, 26(6), 559-562 (1998).  
+
+20. Mia, M. S., Abdelmeguid, M. & Elbanna, A. E. Spatio-Temporal Clustering of Seismicity Enabled by Off-Fault Plasticity. Geophysical Research Letters, 49(8), e2021GL097601 (2022).  
+
+21. Titus, S. J., DeMets, C. & Tikoff, B. Thirty-five-year creep rates for the creeping segment of the San Andreas fault and the effects of the 2004 Parkfield earthquake: Constraints from alignment arrays, continuous global positioning system, and creepmeters. Bulletin of the Seismological Society of America, 96(4B), S250-S268 (2006).  
+
+22. Titus, S. J., Dyson, M., DeMets, C., Tikoff, B., Rolandone, F. & Bürgmann, R. Geologic versus geodetic deformation adjacent to the San Andreas fault, central California. Bulletin, 123(5-6), 794-820 (2011).  
+
+23. Noda, H. & Lapusta, N. Stable creeping fault segments can become destructive as a result of dynamic weakening. Nature, 493(7433), 518-521 (2013).  
+
+24. Waldhauser, F. & Schaff, D. P. Large-scale relocation of two decades of Northern California seismicity using cross-correlation and double-difference methods. Journal of Geophysical Research: Solid Earth, 113(B8) (2008).  
+
+25. Nadeau, R. M., Foxall, W. & McEvilly, T. V. Clustering and periodic recurrence of microearthquakes on the San Andreas fault at Parkfield, California. Science, 267(5197), 503-507 (1995).  
+
+26. Nadeau, R. M. & Johnson, L. R. Seismological studies at Parkfield VI: Moment release rates and estimates of source parameters for small repeating earthquakes. Bulletin of the Seismological Society of America, 88(3), 790-814 (1998).
+
+<--- Page Split --->
+
+27. Turner, R. C., Shirzaei, M., Nadeau, R. M. & Bürgmann, R. Slow and Go: Pulsing slip rates on the creeping section of the San Andreas Fault. Journal of Geophysical Research: Solid Earth, 120(8), 5940-5951 (2015).  
+
+28. Waldhauser, F. & Schaff, D. P. A comprehensive search for repeating earthquakes in northern California: Implications for fault creep, slip rates, slip partitioning, and transient stress. Journal of Geophysical Research: Solid Earth, 126(11), e2021JB022495 (2021).  
+
+29. Cheng, Y., Allen, R. M. & Taira, T. A. A new focal mechanism calculation algorithm (REFOC) using inter-event relative radiation patterns: Application to the earthquakes in the Parkfield area, Journal of Geophysical Research: Solid Earth, 128(3), e2022JB025006 (2023).  
+
+30. Abdi, H. & Williams, L. J. Principal component analysis. Wiley interdisciplinary reviews: computational statistics, 2(4), 433-459 (2010).  
+
+31. Thomas, A. Poly3D: A three-dimensional, polygonal element, displacement boundary element computer program with applications to fractures, faults, and cavities in the Earth's crust [MS thesis] (1993).  
+
+32. Schulz, S. S., Mavko, G. M., Burford, R. O. & Stuart, W. D. Long-term fault creep observations in central California. Journal of Geophysical Research: Solid Earth, 87(B8), 6977-6982 (1982).  
+
+33. Toppozada, T. R., Branum, D. M., Reichle, M. S. & Hallstrom, C. L. San Andreas fault zone, California: M≥ 5.5 earthquake history. Bulletin of the Seismological Society of America, 92(7), 2555-2601 (2002).  
+
+34. Ben-Zion, Y., Rice, J. R. & Dmowska, R. Interaction of the San Andreas fault creeping segment with adjacent great rupture zones and earthquake recurrence at Parkfield. Journal of Geophysical Research: Solid Earth, 98(B2), 2135-2144 (1993).
+
+<--- Page Split --->
+
+35. Sibson, R. H. Fault zone models, heat flow, and the depth distribution of earthquakes in the continental crust of the United States. Bulletin of the Seismological Society of America, 72(1), 151-163 (1982).  
+
+36. Dieterich, J. H. Time-dependent friction and the mechanics of stick-slip. In Rock friction and earthquake prediction (pp. 790-806). Birkhäuser, Basel (1978).  
+
+37. Lockner, D. A., Morrow, C., Moore, D. & Hickman, S. Low strength of deep San Andreas fault gouge from SAFOD core. Nature, 472(7341), 82-85 (2011).  
+
+38. Carpenter, B. M., Marone, C. & Saffer, D. M. Weakness of the San Andreas Fault revealed by samples from the active fault zone. Nature Geoscience, 4(4), 251-254 (2011).  
+
+39. Carpenter, B. M., Saffer, D. M. & Marone, C. Frictional properties of the active San Andreas Fault at SAFOD: Implications for fault strength and slip behavior. Journal of Geophysical Research: Solid Earth, 120(7), 5273-5289 (2015).  
+
+40. Khoshmanesh, M. & Shirzaei, M. Episodic creep events on the San Andreas Fault caused by pore pressure variations. Nature geoscience, 11(8), 610-614 (2018).  
+
+41. Noda, H. & Lapusta, N. Stable creeping fault segments can become destructive as a result of dynamic weakening. Nature, 493(7433), 518-521 (2013).  
+
+42. Schwartz, D. P. Past and future fault rupture lengths in seismic source characterization—The long and short of it. Bulletin of the Seismological Society of America, 108(5A), 2493-2520 (2018).  
+
+43. Zoback, M. D., Zoback, M. L., Mount, V. S., Suppe, J., Eaton, J. P., Healy, J. H., Oppenheimer, D., Reasenberg, P., Jones, L., Raleigh, C. B., Wong, I. G., Scotti, O. & Wentworth, C. New evidence on the state of stress of the San Andreas fault system, Science, 238, 1105-1111, (1987).
+
+<--- Page Split --->
+
+44. Provost, A. S. & Houston, H. Stress orientations in northern and central California: Evidence for the evolution of frictional strength along the San Andreas plate boundary system. Journal of Geophysical Research: Solid Earth, 108(B3), (2003).  
+
+45. Hardebeck, J. L. & Hauksson, E. Role of fluids in faulting inferred from stress field signatures. Science, 285(5425), 236-239 (1999).  
+
+46. Rice, J. R. Fault stress states, pore pressure distributions, and the weakness of the San Andreas fault. In International geophysics (Vol. 51, pp. 475-503). Academic Press (1992).  
+
+47. Chéry, J., Zoback, M. D. & Hassani, R. An integrated mechanical model of the San Andreas fault in central and northern California. Journal of Geophysical Research: Solid Earth, 106(B10), 22051-22066 (2001).  
+
+48. Scholz, C. H. A fault in the 'weak San Andreas' theory. Nature, 406(6793), 234-234 (2000).  
+
+49. Lynch, J. C. & Richards, M. A. Finite element models of stress orientations in well-developed strike-slip fault zones: Implications for the distribution of lower crustal strain. Journal of Geophysical Research: Solid Earth, 106(B11), 26707-26729 (2001).  
+
+50. Brune, J. N., Henyey, T. L. & Roy, R. F. Heat flow, stress, and rate of slip along the San Andreas fault, California. Journal of Geophysical Research, 74(15), 3821-3827 (1969).  
+
+51. Lachenbruch, A. H. & Sass, J. H. Heat flow and energetics of the San Andreas fault zone. Journal of Geophysical Research: Solid Earth, 85(B11), 6185-6222 (1980).  
+
+52. Wibberley, C. A., Yielding, G., & Di Toro, G. Recent advances in the understanding of fault zone internal structure: a review. Geological Society, London, Special Publications, 299(1), 5-33 (2008).
+
+<--- Page Split --->
+
+53. Cheng, Y. & Ben-Zion, Y. Variations of earthquake properties before, during, and after the 2019 M7.1 Ridgecrest, CA, earthquake. Geophysical Research Letters, 47(18), e2020GL089650 (2020).  
+
+54. Xu, W., Feng, G., Meng, L., Zhang, A., Ampuero, J. P., Bürgmann, R. & Fang, L. Transpressional rupture cascade of the 2016 Mw 7.8 Kaikoura earthquake, New Zealand. Journal of Geophysical Research: Solid Earth, 123(3), 2396-2409 (2018).  
+
+55. Kodaira, S., Iidaka, T., Kato, A., Park, J. O., Iwasaki, T. & Kaneda, Y. High pore fluid pressure may cause silent slip in the Nankai Trough. Science, 304(5675), 1295-1298 (2004).  
+
+56. Saffer, D. M. & Marone, C. Comparison of smectite-and illite-rich gouge frictional properties: application to the updip limit of the seismogenic zone along subduction megathrusts. Earth and Planetary Science Letters, 215(1-2), 219-235 (2003).  
+
+57. d'Alessio, M. A., Williams, C. F. & Bürgmann, R. Frictional strength heterogeneity and surface heat flow: Implications for the strength of the creeping San Andreas fault. Journal of Geophysical Research: Solid Earth, 111(B5) (2006).  
+
+58. Xu, S., Fukuyama, E., Yamashita, F., Kawakata, H., Mizoguchi, K. & Takizawa, S. Fault strength and rupture process controlled by fault surface topography. Nature Geoscience, 16(1), 94-100 (2023).  
+
+59. Jiang, J., Bock, Y. & Klein, E. Coevolving early afterslip and aftershock signatures of a San Andreas fault rupture. Science Advances, 7(15), eabc1606 (2021).  
+
+60. Okada, Y. Internal deformation due to shear and tensile faults in a half-space. Bulletin of the seismological society of America, 82(2), 1018-1040 (1992).  
+
+61. Hardebeck, J. L. & Okada, T. Temporal stress changes caused by earthquakes: A review. Journal of Geophysical Research: Solid Earth, 123(2), 1350-1365 (2018).
+
+<--- Page Split --->
+
+62. Hardebeck, J. L. & Shearer, P. M. Using S/P amplitude ratios to constrain the focal mechanisms of small earthquakes. Bulletin of the Seismological Society of America, 93(6), 2434-2444 (2003). 
+
+63. Vidale, J. E., Ellsworth, W. L., Cole, A. & Marone, C. Variations in rupture process with recurrence interval in a repeated small earthquake. Nature, 368(6472), 624-626 (1994). 
+
+64. Nadeau, R. M. & McEvilly, T. V. Fault slip rates at depth from recurrence intervals of repeating microearthquakes. Science, 285(5428), 718-721 (1999). 
+
+65. Wyss, M., Sammis, C. G., Nadeau, R. M. & Wiemer, S. Fractal dimension and b-value on creeping and locked patches of the San Andreas fault near Parkfield, California. Bulletin of the Seismological Society of America, 94(2), 410-421 (2004). 
+
+66. Vavryčuk, V. Iterative joint inversion for stress and fault orientations from focal mechanisms. Geophysical Journal International, 199(1), 69-77 (2014). 
+
+67. Lund, B. & Slunga, R. Stress tensor inversion using detailed microearthquake information and stability constraints: Application to Ölfus in southwest Iceland. Journal of Geophysical Research: Solid Earth, 104(B7), 14947-14964 (1999). 
+
+68. Yang, W., Hauksson, E. & Shearer, P. M. Computing a large refined catalog of focal mechanisms for southern California (1981–2010): Temporal stability of the style of faulting. Bulletin of the Seismological Society of America, 102(3), 1179-1194 (2012).
+
+<--- Page Split --->
+
+Acknowledgments: We thank Shiqing Xu, Taka'aki Taira, Naoki Uchida, and Saeko Kita for valuable comments and suggestions. Funding: This work was funded by the California Governor's Office of Emergency Services (Cal OES), Agreement Number 6172- 2018. Author contributions: Y.C. and R.B. designed the study. Y.C. processed all datasets and performed modeling and analyses. Y.C., R.B., and R.A. all contribute to discussions. Y.C. lead the writing of the manuscript with contributions from R.B. and R.A. Competing interests: The authors declare that they have no competing interests.
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Fig. 1 Improvements in constraining subsurface coupling using earthquake focal mechanisms. </center>  
+
+Schematic comparison of data and assumptions used in a previous geodetic inversion studies \(^{11,12,13,15}\) and b this study relying on the spatial distribution of earthquakes and slip directions from focal mechanisms. Dark red areas denote locked patches and light red areas denote creeping areas. Modeled fault c creep rate and d creep direction around a locked patch. Yellow stars and beachballs denote earthquake locations and focal mechanisms, respectively. Black and blue beachballs are on- fault and off- fault earthquakes, respectively. Red arrows show the assumed fault slip direction.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Fig. 2 Seismicity distribution along the CSAF. a Map view and b cross-section view of relocated earthquakes from 1984 to 2015 (Waldhauser and Schaff, 2008, extended to later years) along the central San Andreas Fault. Events in the box XX' are colored by faulting style. Small beach balls denote \(M \geq 4.0\) earthquakes. Large beach balls denote the 2003 \(M_{w}6.5\) San Simeon and 2004 \(M_{w}6.0\) Parkfield earthquakes' focal mechanisms. White stars denote historic major earthquakes. Note vertical exaggeration \(\mathrm{VE} = 3.13\) in b. The local coordinate system has its origin at latitude \(35.867^{\circ}\mathrm{N}\) , longitude \(120.447^{\circ}\mathrm{W}\) and is oriented \(\mathrm{N42^{\circ}W}\) . </center>
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Fig. 3 Spatiotemporal variation of seismicity and its relationship with the main fault. a Rotated map view, b cross-section view, and c spatiotemporal variations of seismicity (black dots), \(M \geq 4.0\) earthquakes (blue beachballs), and repeating earthquakes (26; red dots) along the Central San Andreas Fault. The local coordinate system has its origin at latitude \(35.867^{\circ}\mathrm{N}\) , longitude \(120.447^{\circ}\mathrm{W}\) and is oriented \(N42^{\circ}\mathrm{W}\) . d Spatial variations of fault strike (red line) and dip (blue line) angles determined using PCA analysis from relocated seismicity around each 15-km-long 15-km-deep fault segment stepping at 1-km intervals along the fault. The reference fault strike and dip are \(N138^{\circ}\mathrm{E}\) and \(90^{\circ}\) (vertical), respectively. Cumulative Density Functions (CDF) of the e horizontal distance of events from main fault and f azimuthal difference from main fault strike </center>
+
+<--- Page Split --->
+
+for earthquakes located within 1 km horizontal distance from the main fault based on focalmechanism nodal plane closest to the main fault. Black, blue, and red curves denote \(M \geq 1.0\) earthquakes, \(M \geq 4.0\) earthquakes, and repeating earthquake sequences, respectively. SJB: San Juan Bautista, MR: Melendy Ranch; BW: Bitterwater; SC: Slack Canyon; SAFOD: San Andreas Fault Observatory at Depth; MM: Middle Mountain; PK: Parkfield; GH: Gold Hill.
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Fig. 4 Modeled and observed on-fault displacement. a Fault model 1 setup with freely slipping zones (blue), locked sections (gray) and constant-rate creeping zones (red). The deep creeping zone driving the shallow creep extends from \(15\mathrm{km}\) to \(2000\mathrm{km}\) depth and far beyond the lateral ends of the CSAF. The shallow fault is fully coupled beyond \(166\mathrm{km}\) NW and -17km SE of the fault. The size of each fault patch is \(3\times 3\mathrm{km}\) . White stars denote \(M\geq 4.0\) earthquakes. b Modeled and c observed fault creep rate estimated from the occurrence of repeating earthquakes. d Modeled and e observed fault slip direction estimated from the rake and dip of repeating earthquake focal </center>
+
+<--- Page Split --->
+
+mechanisms. Colored dots and grids in c and e represent the values from each individual repeating earthquake sequence and those averaged in each spatial bin. Positive creep directions indicate a NE- side- up dip- slip component. f Modeled (blue) and observed surface creep rate (red) estimated from InSAR data (Jolivet et al., 2014). Green circles and squares denote average surface creep rates from creepmeters and alignment arrays (circle: 30; square: 19), respectively.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Fig. 5 Off-fault stress field and kinematics. Comparison of a the angle between the main fault and the modeled off-fault maximum horizontal stress orientation \((\theta)\) 1.5 km NE of the main fault with b the angle between the main fault and the observed maximum horizontal stress orientation calculated from \(M\geq 1.0\) focal mechanisms located within 2 km around the fault trace. White dots denote the locations of focal mechanisms used in stress inversion. c The percentage of oblique-reverse-faulting events \((P e r c_{r a k e > 0};\) red curve) for M1.0 focal mechanisms located within 2 km around the fault trace. d Point-to-point comparison of \(\theta\) in b and \(P e r c_{r a k e > 0}\) in c. e Schematic illustration showing the variation of fault zone structure, weakness, and stress field. </center>
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- CSAFSupplementaryMaterials1127.pdf
+
+<--- Page Split --->

preprint/preprint__7e878f0cd9d5b59940e0554b3ad286805751e8d5ee47e9c5c90a04d0070bacd8/preprint__7e878f0cd9d5b59940e0554b3ad286805751e8d5ee47e9c5c90a04d0070bacd8_det.mmd

@@ -0,0 +1,528 @@
+<|ref|>title<|/ref|><|det|>[[42, 107, 877, 175]]<|/det|>
+# 3D architecture and complex behavior along the simple central San Andreas fault  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 140, 214]]<|/det|>
+Yifang Cheng  
+
+<|ref|>text<|/ref|><|det|>[[55, 222, 290, 240]]<|/det|>
+chengyi@berkeley.edu  
+
+<|ref|>text<|/ref|><|det|>[[44, 268, 700, 381]]<|/det|>
+University of California, Berkeley https://orcid.org/0000- 0002- 6507- 5607Roland BürgmannUniversity of California, Berkeley https://orcid.org/0000- 0002- 3560- 044XRichard AllenUniversity of California Berkeley https://orcid.org/0000- 0003- 4293- 9772  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 421, 103, 439]]<|/det|>
+## Article  
+
+<|ref|>title<|/ref|><|det|>[[44, 460, 135, 478]]<|/det|>
+# Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 497, 350, 516]]<|/det|>
+Posted Date: November 30th, 2023  
+
+<|ref|>text<|/ref|><|det|>[[44, 535, 474, 555]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 3678641/v1  
+
+<|ref|>text<|/ref|><|det|>[[42, 572, 914, 615]]<|/det|>
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 634, 533, 653]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 689, 916, 732]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on June 25th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 49454- z.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[81, 117, 881, 140]]<|/det|>
+# 3D architecture and complex behavior along the simple central San Andreas fault  
+
+<|ref|>title<|/ref|><|det|>[[84, 283, 157, 300]]<|/det|>
+# Authors  
+
+<|ref|>text<|/ref|><|det|>[[144, 315, 620, 336]]<|/det|>
+Yifang Cheng, \(^{1,2*}\) Roland Bürgmann, \(^{1,2}\) Richard M. Allen \(^{1,2}\)  
+
+<|ref|>title<|/ref|><|det|>[[84, 525, 183, 543]]<|/det|>
+# Affiliations  
+
+<|ref|>text<|/ref|><|det|>[[144, 565, 860, 586]]<|/det|>
+\(^{1}\) Department of Earth and Planetary Science, University of California, Berkeley, CA, USA  
+
+<|ref|>text<|/ref|><|det|>[[144, 608, 890, 628]]<|/det|>
+\(^{2}\) Berkeley Seismological Laboratory, University of California, Berkeley, Berkeley, California,  
+
+<|ref|>text<|/ref|><|det|>[[114, 644, 170, 660]]<|/det|>
+U.S.A.  
+
+<|ref|>text<|/ref|><|det|>[[144, 720, 536, 740]]<|/det|>
+\(^{*}\) Yifang Cheng Email: chengyif@berkeley.edu
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[84, 91, 162, 108]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[79, 121, 911, 565]]<|/det|>
+The central San Andreas Fault (CSAF) exhibits a simple linear large- scale fault geometry, yet seismic and aseismic deformation features vary in a complex way along the fault. Here we investigate fault zone behaviors using geodetic observation, seismicity and microearthquake focal mechanisms. We employ an improved focal- mechanism characterization method using relative earthquake radiation patterns on 75,164 M \(\geq 1\) earthquakes along a 2- km- wide, 190- km- long segment of the CSAF, from 1984 to 2015. The data reveal the 3D fine- scale structure and interseismic kinematics of the CSAF. Our findings indicate that the first- order spatial variations in interseismic fault creep rate, creep direction, and the fault zone stress field can be explained by a simple fault coupling model. The inferred 3D mechanical properties of a mechanically weak and poorly coupled fault zone provide a unified understanding of the complex fine- scale kinematics, indicating distributed slip deficits facilitating small- to- moderate earthquakes, localized stress heterogeneities, and complex multi- scale ruptures along the fault. Through this detailed mapping, we aim to relate the fine- scale fault architecture to potential future faulting behavior along the CSAF.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[84, 91, 196, 108]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[80, 128, 912, 606]]<|/det|>
+Fault zones have geometric complexities and are multiscale systems consisting of a localized fault core accommodating the primary slip (on- fault), a highly fractured damage zone around the fault core accommodating a small fraction of deformation (off- fault), and the surrounding host rock \(^{1,2,3}\) . Fault zones can deform seismically, such as in large damaging earthquakes, small earthquakes including earthquake clusters and repeating events, and tectonic tremor \(^{4}\) , and aseismically, such as by transient slow slip events, steady creep, and afterslip \(^{5,6}\) . The partitioning of seismic and aseismic fault slip is usually quantified using geodetically determined fault coupling, which is defined as the ratio of the inferred slip deficit rate and the long- term slip rate, with a value of 1 corresponding to fully locked while a value of 0 indicates a freely creeping fault. The spatial distribution of fault coupling and implied slip deficits can help us better understand fault zone properties, determine the seismic potential of faults, and constrain the return period and the maximum- possible magnitude of earthquakes \(^{7,8,9}\) . Fault sections with larger kinematic coupling at depth generally correspond to lower surface creep rates observed from geodetic investigations \(^{10,11,12,13}\) , lower recurrence rates of repeating earthquakes \(^{14,15,16}\) , lower b- values \(^{17}\) , and a larger fraction of clustered events \(^{18}\) .  
+
+<|ref|>text<|/ref|><|det|>[[82, 626, 912, 892]]<|/det|>
+However, gaps remain in our understanding of the seismic and aseismic deformation of faults due to several problems. It is often assumed that all patches on a fault slip in the same direction without considering the variation of slip directions that we can expect in areas with abrupt changes of fault geometry or creep rate \(^{19}\) . Secondly, analyses of seismicity often focus on statistical parameters estimated from earthquake occurrences instead of the underlying physical properties of fault zones. Moreover, the effects of distributed fault- zone deformation and on- fault/off- fault interactions are usually ignored, although on- fault and off- fault deformations are tightly interlinked and coevolve with strong feedback loops over multiple spatial and temporal scales \(^{20}\) (Fig. 1a). In this work, we analyze the distribution of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[81, 88, 901, 180]]<|/det|>
+fault orientations and slip directions in the fault zone from a comprehensive catalog of earthquake focal mechanisms. This helps to resolve fine- scale on- fault slip directions, provides physical parameters for seismicity analysis, and allows for differentiating on- and off- fault seismic deformation (Fig. 1b).  
+
+<|ref|>text<|/ref|><|det|>[[80, 199, 910, 855]]<|/det|>
+The 190- km- long central segment of the San Andreas Fault (CSAF) in California (Fig. 2) offers a unique opportunity to investigate the fine- scale complexities of seismic and aseismic fault zone processes and their interconnected nature. The CSAF is characterized by a long- term slip rate of \(\sim 33 - 35\) mm/yr and varying surface creep at \(\leq 30\) mm/yr<sup>21,22</sup>. It lacks large historical earthquakes, while \(M_{w} \sim 7 - 8\) earthquakes repeatedly occur to the north and south (e.g., the 1906 San Francisco earthquake and the 1857 Fort Tejon earthquake; Fig. 2a and Supplementary Fig. 1). Understanding the potential for large earthquakes along the CSAF is important for seismic hazard assessment of the SAF and other large creeping fault zones in the world<sup>23</sup>. The CSAF also provides a natural laboratory for studies of seismic and aseismic fault zone processes because of the availability of a dense, continuous monitoring network. Decades of monitoring show that the CSAF exhibits heterogeneous seismic and aseismic deformation patterns and variable fault coupling inferred from varying along- fault surface displacements<sup>11,12,13,21</sup>, the occurrence of moderate earthquakes (e.g. 1922, 1934, 1966, and 2004 M6.0 Parkfield earthquake), abundant small earthquakes<sup>24</sup>, and repeating earthquakes whose recurrence rate is proportional to the fault creep rate at depth<sup>25,26,27,28</sup>. In order to obtain focal mechanisms of the above- mentioned small earthquakes, we use a recently developed relative focal mechanism calculation algorithm<sup>29</sup> and obtained high- quality focal mechanisms (uncertainty \(< 35\) degree) of \(\sim 80\%\) of the \(M_{l} \geq 1\) catalog events<sup>24</sup>. The diverse fault deformation patterns, abundant earthquakes, and their focal mechanisms provide a great opportunity to comprehensively characterize and understand multiscale fault zone deformation processes along the CSAF and their connections (Figs. 1 and 2).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[85, 92, 150, 108]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[85, 125, 320, 144]]<|/det|>
+## Overview of seismicity data  
+
+<|ref|>text<|/ref|><|det|>[[78, 152, 912, 888]]<|/det|>
+In this study, we integrate the location, time, magnitude, and focal mechanism of \(75,164 M_{l} \geq 1.0\) earthquakes, \(145 M_{l} \geq 4.0\) earthquakes, and 355 repeating earthquakes from 1980 to 2015 to place high- resolution constraints on the fault structure of the CSAF (Fig. 3). Most \(M_{l} \geq 4.0\) sequences are located along the San Juan Bautista (SJB) and Parkfield (PK) transition zones and a few \(M_{l} \geq 4.0\) earthquakes are located near Bitterwater (BW). In contrast, repeating earthquakes occur more frequently between Melendy Ranch (MR) and the San Andreas Fault Observatory at Depth (SAFOD), implying high aseismic creep rates. Since most repeating earthquakes are concentrated in a narrow zone and appear to delineate the major fault strands, we first use repeating earthquakes to obtain the horizontal location of major fault strands and then use both repeating earthquakes and the surrounding \(M_{l} \geq 1.0\) earthquakes within 1 km epicentral distance (Fig. 3a- b) to estimate the location and the orientation of the main fault using principal component analysis \(^{30}\) (see Methods section). The whole fault is nearly vertical with strike angles varying between N120°E to N140°E (Fig. 3d). Our final catalog includes 99.7% of the repeating earthquake sequences and 97.8% of \(M_{l} \geq 4.0\) earthquakes, as well as 83.4% of \(M_{l} \geq 1.0\) events located within 1km NE from the main fault strand (Fig. 3e and Supplementary Fig. 2). To better quantify the fault zone structure, we pick the focal mechanism nodal plane closer to the main fault orientation and calculate the azimuthal difference between the nodal plane and the main fault. Note that this calculation assumes that the nodal plane closer to the main fault orientation is the real fault plane and ignores left- lateral faults at high angles to the main fault. For these near- fault earthquakes, we obtain the percentages of \(M_{l} \geq 4.0\) earthquakes, repeating earthquake sequences, and \(M_{l} \geq 1.0\) earthquakes with less than 20° azimuthal difference from the main fault, which are 92.9%, 94.1%, and 83.3%, respectively (Fig. 3f). Therefore, in the following analysis we assume that \(M_{l} \geq 4.0\) earthquakes and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[80, 88, 910, 250]]<|/det|>
+repeating earthquake sequences are located on the main fault (Supplementary Fig. 3), while \(M_{l} \geq 1.0\) earthquakes have variable orientations and are located both on- and off- fault. To focus on the interseismic period and disregard the co- and early postseismic deformation due to the 1989 M6.9 Loma Prieta and 2004 \(M_{w}6.0\) Parkfield earthquakes, we exclude earthquakes within the first 3 years after these events when calculating the occurrence rates of repeating earthquakes and focal mechanism properties.  
+
+<|ref|>sub_title<|/ref|><|det|>[[85, 272, 239, 291]]<|/det|>
+## Fault model setup  
+
+<|ref|>text<|/ref|><|det|>[[80, 306, 912, 572]]<|/det|>
+The abundant small earthquake focal mechanisms provide a great opportunity to determine the fine- scale fault slip directions, which can be directly compared with the output of kinematic fault models of the partially coupled CSAF employing shear- stress- free boundary conditions on the fault. As shown in Fig. 1c and 1d, if there is a locked fault patch at depth, the fault slip rate decreases slightly around the locked patch, and the surrounding slip directions exhibit opposite rotations near the two ends of the locked patch, providing a powerful constraint on the location and size of the locked patches at depth. Therefore, in this study, we forward model the interseismic fault deformation of the CSAF and compare the obtained slip kinematics with the fault zone properties estimated from seismicity.  
+
+<|ref|>text<|/ref|><|det|>[[80, 592, 910, 896]]<|/det|>
+Based on the definition of fault coupling and previous observations, most moderate- to- large earthquakes are located in high- coupling areas while most repeating earthquakes are interpreted as small patches that repeatedly break and are surrounded and driven to failure by aseismically slipping sections of the fault (low- coupling areas). Therefore, based on the spatial distribution of \(M_{l} \geq 4.0\) events and repeating earthquakes (Fig. 3a), we design a simple forward model that consists of several locked patches near SJB, BW, and PK (in areas of low earthquake- density and near the \(M_{l} \geq 4.0\) earthquakes; Fig. 3b) and an otherwise freely slipping shallow fault in the top 15 km depth driven by a buried fault plane with a 34 mm/yr interseismic deep creep rate \(^{22}\) beneath it (model A in Fig 4a). The fault is bounded by fully locked fault segments at the two ends, corresponding to the locked sections of the SAF
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[82, 88, 904, 180]]<|/det|>
+that hosted the 1857 M7.9 Fort Tejon to the SE as well as the 1906 M7.8 San Francisco and 1989 M6.9 Loma Prieta earthquakes to the NW. We then compute the fault displacement rate and slip directions on the freely slipping patches using a boundary element method (Poly3D; see Methods section) \(^{31}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[84, 202, 500, 221]]<|/det|>
+## Observed and modeled along-fault displacements  
+
+<|ref|>text<|/ref|><|det|>[[80, 231, 912, 890]]<|/det|>
+We compare the forward modeling results with multiple geophysical observations. Besides the traditionally used surface creep rate (red curve and green symbols in Fig. 4f) \(^{11,21,32}\) , we also estimate the 2D variations of creep rate and creep direction on the NE side of the main fault using the occurrence rate and the slip direction of repeating earthquakes, respectively (see Methods section; Supplementary Figs. 3- 5). Fig. 4 shows the comparison of the modeled and observed creep rate using the occurrence rate of repeating earthquakes (Fig. 4b- c), the modeled and observed creep direction using the focal mechanisms of repeating earthquakes (Fig. 4d- e), and the modeled and observed surface creep rate using creepmeter, alignment array and InSAR data (Fig. 4f) \(^{11,21,32}\) . The modeled results are overall consistent with the observed subsurface fault creep rate variations, fault slip directions at depth, and surface creep rates with correlation coefficients of 0.47 (Fig. 4b- c), 0.55 (Fig. 4d- e), and 0.91 (Fig. 4f), respectively (Supplementary Fig. 6). The presence of locked fault patches results in gradually decreasing creep rates around the patches. For example, the observed creep rates from repeating earthquakes and measured surface creep rates are lower than 25 mm/yr and 15 mm/yr, respectively, near the SJB and PK transition zones (Fig. 4b, 4c, 4f). In the central creeping section from MR to the SAFOD, the modeled and observed creep rates approach the deep creep rate ( \(\sim 34 \mathrm{mm / yr}\) ). The repeater- derived creep directions show a more heterogeneous distribution around the locked patches with an upward creep to the NW of the patch and downward creep to the SE of the patch (Fig. 4d- e), providing valuable additional constraints on fault coupling at depth. For example, there is no clear evidence of the existence of locked fault patches near BW based on just the variation of creep rates (Fig. 4f). In contrast, the abrupt changes
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[82, 88, 875, 145]]<|/det|>
+of the vertical creep component (Fig. 4d-e) and the observed \(M_{l} \geq 4.0\) sequences near BW (Fig. 4a) indicate considerable stress accumulation and the existence of a deep locked fault patch near BW.  
+
+<|ref|>sub_title<|/ref|><|det|>[[85, 167, 377, 186]]<|/det|>
+## Off-fault structure and kinematics  
+
+<|ref|>text<|/ref|><|det|>[[81, 201, 901, 680]]<|/det|>
+In addition to on- fault displacements, there are also many small earthquakes located around the fault and most of them are located to the NE of the fault trace (Fig. 3 and Supplementary Fig. 2). Therefore, we obtain the along- fault cross- sectional distribution of the modeled azimuthal differences \((\theta)\) between the off- fault maximum horizontal stress orientation \((SH_{\mathrm{max}})\) and the main fault 1.5 km NE of the main fault plane (see Methods section; Fig. 5a). The correlation coefficient between the modeled off- fault stress distribution and the observed stress field estimated using \(M_{l} \geq 1.0\) focal mechanisms located within 2 km of the main fault (Fig. 5b) is 0.33 (Supplementary Fig. 6d), which is considerable considering the effects of background tectonic stress and the strong stress variations across the fault. Overall, the NW part of the fault shows low \(\theta\) value, and the SE part has high \(\theta\) value. However, both modeled and observed variations of stress orientation near BW exhibit an opposite pattern with high \(\theta\) value to the NW of BW and low \(\theta\) value to the SE of BW, consistent with the existence of a large deep locked patch near the BW. The consistency between the observed stress field and the modeled off- fault stress field variations suggests that on- fault coupling heterogeneity can cause significant stress perturbation in the surrounding area.  
+
+<|ref|>text<|/ref|><|det|>[[82, 700, 905, 898]]<|/det|>
+We further investigate the off- fault structure and kinematics by calculating the cross- section distribution of the percentage of oblique- reverse faulting events \((P e r c_{r a k e > 0})\) (Fig. 5c). There is a significant negative correlation between with \(P e r c_{r a k e > 0}\) with a correlation coefficient of - 0.43 When \(SH_{\mathrm{max}}\) is oriented at a high angle to the main fault, most events are oblique- normal faulting events with optimal fault orientations at a high angle to the main fault (case 1 in Fig. 5d- e). When \(SH_{\mathrm{max}}\) is at about 45 degrees, there are a comparable number of oblique- normal and oblique- reverse faulting events,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[81, 88, 907, 250]]<|/det|>
+suggesting strike- slip faulting is the preferred faulting style (case 2 in Fig. 5d- e). When \(SH_{\mathrm{max}}\) in the fault zone is at a low angle to the main fault, most earthquakes are oblique- reverse faulting events with optimal fault orientations showing high angles to the main fault (case 3 in Fig. 5d- e). In all cases, small- scale faults tend to have horizontal slip directions parallel to the main fault orientation and some of their fault strike orientations exhibit a high angle to the main- fault orientation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[85, 272, 192, 293]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[80, 310, 907, 682]]<|/det|>
+To evaluate the proposed fault model, we also perform modeling using a simple creeping fault model without the small, locked patches but with fully locked segments on the two ends of the CSAF (model B, Supplementary Fig. 9a) and compared the modeling results using model B with the observations (Supplementary Fig. 10). Without the locked patches on the CSAF, the modeled creep rate is generally higher than the observed creep rate but the modeled and observed creep rates still show a first- order correlation (Supplementary Fig. 10a, 10c). In contrast, there is almost no correlation between the observed and modeled creep directions (Supplementary Fig. 10b, 10d), suggesting that the vertical fault slip component is highly sensitive to the small- scale fault coupling heterogeneity and can help to constrain the fault coupling model at depth. The solved fine- scale fault zone properties and fault coupling model provide great opportunity to improve our understanding of the physical mechanisms of small, locked asperities that were not noticed before.  
+
+<|ref|>text<|/ref|><|det|>[[81, 709, 914, 906]]<|/det|>
+There are some misfits between the modeled and observed fault zone properties due to the uncertainty of input data and the simplicity of the first- order fault coupling model. The range of modeled creep directions is \(\pm 6\) degrees from the horizontal direction and is much smaller than that of the creep directions observed from repeating earthquake focal mechanisms \((\pm 30\) degrees). The different ranges of modeled and observed creep directions might be caused by local variations of the fault dipping angle (Supplementary Figs. 8, 11), the considerable uncertainties of small- earthquake focal mechanism
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[79, 88, 911, 460]]<|/det|>
+solutions (Hardebeck and Shearer, 2002; \~20 degree in this study), and fault coupling heterogeneities smaller than the modeled \(3 \times 3\) km patch size (Supplementary Fig. 5). In contrast to the fault creep direction, the range of the modeled \(\theta\) is 5- 85 degrees is larger than the observed stress orientation (20- 70 degrees). This is because the inverted stress field is obtained from focal mechanisms of a large number of earthquakes (> 100) along a given fault patch including many on- fault earthquake focal mechanisms, compared to only a few (<10) repeating earthquake sequences in each fault patch, which is averaged over a long period and is less sensitive to focal mechanism uncertainties and local fault zone heterogeneities. Moreover, the observed surface creep rate has shown to be different from the modeled surface creep rate between MR and BW, which might be due to the temporal oscillation of surface creep rate, the simple first- order fault coupling model, etc. Better fitting might be achieved using a more advanced adjoint inversion method, but that is beyond the scope of this paper.  
+
+<|ref|>text<|/ref|><|det|>[[81, 488, 912, 896]]<|/det|>
+The creep rate distribution obtained from our first- order fault coupling model is overall consistent with those inverted from geodetic observations (Fig. S12) \(^{12,13}\) , with low fault coupling on the central creeping section between 25 and 100 km along the fault and relatively high fault coupling on the other part of the fault, but with better- constrained fault slip distributions at depth. One interesting feature is that many of these models show a deep low- coupling area in the central creeping section but with somewhat different locations. In this study, we constrain the location of the deep locked patch near BW using additional observations from earthquakes at depth, including \(M_{l} \geq 4.0\) earthquakes (Figs. 3, 4a), reduced estimated creep rates (Fig. 4c), as well as abrupt changes of creep direction (Fig. 4e), off- fault stress orientation (Fig. 5b), and earthquake faulting style (Figs. 2, 4c). Based on the modeling result, the differential creep rate between the locked patch near BW and the surrounding high- creep- rate patches is much higher compared with the other locked patches in the SJB and PK transition zones, suggesting a higher rate of elastic strain accumulation. However, there are only 2 instrumentally observed \(M_{l} \geq 4.0\)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[80, 88, 912, 460]]<|/det|>
+\((M_{w}4.1\) and \(M_{w}5.3)\) earthquakes near the BW locked patch in the past 100 years compared with other locked patches near MR and PK. In contrast, there were six \(M_{w} > 5.5\) earthquakes near BW between the 1857 Fort Tejon and the 1906 San Francisco earthquakes (Supplementary Fig. 1) \(^{33}\) . The infrequent occurrences of large- magnitude earthquakes near the BW locked patch in the past century might be caused by the changes of absolute stress level and the recurrence times of moderate- to- large magnitude events caused by the occurrence of the 1857 Fort Tejon and 1906 San Francisco earthquakes \(^{34}\) . Similar changes of the occurrence rate of large- magnitude earthquakes before and after the 1857 Fort Tejon and 1906 San Francisco earthquakes are also seen along other sections of the fault (Supplementary Fig. 1) \(^{33}\) . Since the main fault orientation is simple in this area (Fig. 3a, 3d), the significant variations of fault locking may be mainly due to the variations of material properties, such as lithology \(^{37,38,39}\) , local temperature anomalies \(^{35}\) , and pore fluid pressure \(^{40}\) .  
+
+<|ref|>text<|/ref|><|det|>[[80, 488, 911, 897]]<|/det|>
+The seismic potential inferred from the fault coupling model along the CSAF is of interest from a hazard perspective because the CSAF accommodates nearly all the plate motion in this part of California. Assuming a deep slip rate of \(34 \mathrm{mm} / \mathrm{yr}^{21}\) , the modeled moment deficit rate is \(1.32 \times 10^{18} \mathrm{N} \cdot \mathrm{m} / \mathrm{yr}\) , while \(1.48 \times 10^{18} \mathrm{N} \cdot \mathrm{m} / \gamma \mathrm{r}\) are released by aseismic slip. The stored moment over a 150- year period is equivalent to a moment magnitude \((M_{w})\) 7.5 earthquake which agrees with previous studies \(^{12,13}\) . If we assume that the areas of partial coupling catch up the moment deficit only by aseismic slip, such as by accelerated afterslip following earthquakes on the SAF, the modeled moment deficit rate is \(4.04 \times 10^{17} \mathrm{N} \cdot \mathrm{m} / \gamma \mathrm{r}\) and the 150- year accumulated moment is equivalent to a \(M_{w}7.2\) earthquake. Since the locked patches are distributed along the creeping section with low slip deficit around them, these small locked patches are more likely to rupture independently as moderate earthquakes (e.g., the 2004 Parkfield earthquake) with a substantial amount of slip deficit on the surrounding fault being taken up by transient afterslip of major ruptures (e.g., following the 1906 Great San Francisco, 1989 Loma
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[81, 87, 896, 390]]<|/det|>
+Prieta, and 2004 Parkfield earthquakes \(^{15,27}\) . An important but controversial question is whether major ruptures can dynamically penetrate deep into the central creeping section \(^{5,41,42}\) . The deep locked patch near BW may weaken the barrier effect of the central creeping section and increase the possibility for major earthquakes to rupture through the whole creeping section. However, the stressing rate would be required to be high enough to penetrate multiple creeping sections between the small, locked patches along the CSAF. Note that while we can't fully rule out this scenario, there is no direct evidence that great ruptures have made it across the CSAF creeping section \(^{42}\) . Therefore, a substantial portion of the accumulated moment deficit is more likely to be released by the afterslip of surrounding major earthquakes and moderate- magnitude earthquakes on the CSAF.  
+
+<|ref|>text<|/ref|><|det|>[[80, 416, 907, 860]]<|/det|>
+The fine- scale fault coupling heterogeneity can also affect the large- scale stress distribution by perturbing the stress field near the fault (Fig. 4a- b). The rotation of \(SH_{max}\) from high angles away from the main fault to low angles in the vicinity of the SAF has been observed and studied for decades \(^{43,44,45}\) . Rice \(^{46}\) proposed a high pore- pressure fracture zone model that can decrease the effective fault- normal compressive stress and cause stress rotation. However, this can only be applied to narrow, high pore- pressure fault zones \(^{47,48}\) . Some studies explain the stress rotation as the combined effect of a weak fault and finite- width weak lower crust \(^{47,49}\) , which requires strong lateral variations of lower crustal properties. Scholz \(^{48}\) suggested that the fault- parallel shear stress decreases with distance from the SAF due to the frictional resistance to strike- slip motion under a strong- fault hypothesis, which conflicts with the absence of localized shear heating near the main fault \(^{50,51}\) . In this study, we found that even though the whole CSAF is rapidly creeping with limited accumulated shear stress overall, fine- scale fault coupling heterogeneity can still cause substantial stress perturbations near the fault. Since the along- fault strike- slip motion and the fine- scale heterogeneous fault zone deformation generally co- exist along the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[83, 89, 886, 145]]<|/det|>
+whole SAF, this provides an additional explanation to the observed stress rotation near the main fault that is not related to the strong- or weak- fault hypotheses.  
+
+<|ref|>text<|/ref|><|det|>[[80, 163, 910, 821]]<|/det|>
+In addition to stress orientation, the fine- scale fault zone coupling heterogeneity can also cause local stress concentrations around the major fault, facilitating multi- scale earthquake ruptures near the major fault and promoting the evolution of fault zone architecture. When earthquake source dimensions significantly exceed the fault zone width, their ruptures tend to align with the main fault orientation, showing consistent focal mechanisms ( \(M_{t} \geq 4\) earthquakes in Fig. 3e- f). When earthquake source dimensions are smaller than or comparable to the fault zone width, earthquakes in the fault zone exhibit more diverse focal mechanisms. However, they still tend to slip in a fault- parallel direction with a notable portion of secondary faults showing high angle to the main fault ( \(M_{t}1 - 4\) earthquakes in Figs. 3e- f, 5d- e). These small high- angle faults might be related to phenomena such as microfracturing near the fault tips, relay breaching and splay- faulting, promoting the weakening of rock adjacent to the developing fault \(^{52}\) . The weakened rock in the fault zone potentially promotes the inclination of small subsidiary fault to slip along a fault- parallel direction, the linkage of fault segments, and the occurrences of moderate- and- large earthquakes along the major fault segments (Figs. 2- 3) \(^{52}\) . The narrow weak fault zone can also cause localized stress and slip concentrations, which may lead to a zone of localized stress rotation in the vicinity of the SAF. Since the narrow weak fault zone is much easier to deform than the surrounding strong host rock, moderate and large magnitude earthquakes along the CSAF are more likely to be right- lateral ruptures along the main fault orientation instead of complex ruptures that are frequently observed in wide, diffuse, and immature fault zones, like the 2019 \(M_{w}7.1\) Ridgecrest earthquake \(^{53}\) and the 2016 \(M_{w}7.8\) Kaikoura earthquake \(^{54}\) .  
+
+<|ref|>text<|/ref|><|det|>[[83, 839, 907, 894]]<|/det|>
+One important thing to note is that neither the above- mentioned fault coupling distribution, the large- scale stress rotations, nor the relative weakness of the narrow fault zone indicate the absolute level
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[81, 88, 912, 389]]<|/det|>
+of fault strength. Both frictional strength and fault coupling can be affected by variations of the stress field on the fault and variations of material properties, such as lithology \(^{37,38,39}\) , temperature \(^{35}\) , and pore fluid pressure \(^{55}\) . However, there is no clear correlation between fault coupling and fault strength \(^{56,57}\) . The variation of fault geometry and surface roughness \(^{58}\) can also significantly affect the fault strength. One possible way to estimate the strength of the CSAF might be the direct estimation of absolute stress levels before moderate earthquake ruptures along the CSAF based on changes of small earthquake focal mechanisms \(^{29}\) . For example, we can search over possible absolute stress levels, model the stress rotation in the fault zone before and after the 2004 Parkfield earthquake using finite source slip models \(^{59,60}\) , and compare the results with the focal mechanism observations \(^{61}\) .  
+
+<|ref|>text<|/ref|><|det|>[[81, 409, 900, 886]]<|/det|>
+In summary, we use high- resolution locations and focal mechanisms of repeating earthquakes, \(M_{l} \geq 4\) earthquakes, and \(M_{l}1 - 4\) earthquakes to illuminate fine- scale on- fault creep rates and directions as well as the stress field, structure, and earthquake slip variations in the fault zone. Our results reveal closely connected on- fault and off- fault deformation processes. All observed fine- scale kinematic features can be reconciled with a simple fault coupling model, inferred to be surrounded by a narrow, mechanically weak zone. Our study demonstrates the value of integrating small- earthquake focal mechanisms into fault zone analysis, to better resolve the detailed fault orientations, slip directions, and stress field variations associated with various aseismic and seismic processes. The resolved fault coupling heterogeneity, the surrounding narrow weak fault zone with complex internal structure, as well as their interactions have important implications for revealing multi- scale fault zone deformation, understanding large- scale stress field variation, and estimating the slip budget, timing, and patterns of future major earthquakes. The analyses performed in this study can be applied to other transform and subduction fault zones around the world for better understanding of multi- scale fault geometry and kinematics and improved estimation of future seismic hazard.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[85, 91, 161, 108]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[85, 125, 251, 144]]<|/det|>
+## Fault slip modeling  
+
+<|ref|>text<|/ref|><|det|>[[82, 157, 912, 600]]<|/det|>
+The variation of fault displacement and stress around the fault depend on the mechanical response of the fault to the shear stress loading from the lower crust. Therefore, we used the boundary- element code Poly3D<sup>31</sup> to simulate fault slip processes during the inter- seismic period. The fault lies in a uniform elastic half- space with a Poisson's ratio of 0.25 and shear modulus of 30 GPa. It is vertical both in and below the seismogenic layer with 3×3 km patch size. In the top 15 km depth. Only zero- slip or zero- shear- stress boundary conditions on patches of the shallow fault are defined, rather than frictional properties. The fault is mostly freely slipping between - 17 km and 166 km with several small frictionally locked patches and is fully coupled beyond 166 km NW and - 17km SE of the fault (model A; Fig. 4a). Between 15 km and 2000 km depth, the fault slips steadily with a 34 mm/yr inter- seismic slip rate between - 500 km to 500 km horizontal distance along the fault. To evaluate the effect of the small, locked fault patched, we also consider a model without small frictionally locked patches between - 17 km and 166 km along the fault but with a fully creeping fault bounded by locked segments at the two ends of the fault (model B; Supplementary Figs. 9, 10).  
+
+<|ref|>sub_title<|/ref|><|det|>[[85, 620, 491, 640]]<|/det|>
+## \(M_{l}\geq 1\) earthquake focal mechanism calculation  
+
+<|ref|>text<|/ref|><|det|>[[82, 655, 904, 885]]<|/det|>
+In this study, we calculate the focal mechanisms along the CSAF using the REFOC algorithm<sup>29</sup>, which uses first- motion polarities, S- /P- wave amplitude ratios to obtain the initial earthquake focal mechanisms and further constrain them using the P- wave amplitude ratios and S- wave amplitude ratios of closely located earthquakes within 3 km hypocentral distance. We utilize the polarities, P- and S- wave phases manually picked by data analysts, and the relocated earthquake catalog archived by the Northern California Earthquake Data Center (NCEDC). We use the same P- wave velocity models<sup>29</sup> and the S- wave velocity models are derived from P- wave velocity models by assuming that the P- /S- wave
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[80, 88, 900, 420]]<|/det|>
+velocity ratio equals 1.732. We apply a 1- 10 Hz band- pass- filter to earthquake waveform data from NCEDC, choose 0.5s before to 1.5s after P- and S- wave arrivals as the signal windows and 2.5s to 0.5s before P- wave arrivals as noise windows, and take the difference between maximum and minimum amplitude values in each time window to be the estimated signal and noise amplitudes. If the time difference between P- and S- wave arrivals is larger than 2s and the signal- to- noise ratio (SNR) is larger than 3, we use the P- and S- wave amplitudes to obtain S-/P- wave amplitude ratios for each individual event as well as the inter- event P- wave amplitude ratios and S- wave amplitude ratios. We obtained 52,211 out of 65,492 ( \(\sim 80\%\) ) \(M_{l} \geq 1\) earthquakes with at least 8 polarities and focal mechanism uncertainties less than 35 degree. The catalog is available via (https://data.mendeley.com/datasets/34szj3jm6k/1).  
+
+<|ref|>sub_title<|/ref|><|det|>[[85, 445, 516, 465]]<|/det|>
+## Repeating earthquake focal mechanism calculation  
+
+<|ref|>text<|/ref|><|det|>[[80, 480, 904, 885]]<|/det|>
+Since repeating earthquakes share highly similar waveforms and locations, they also share similar rupture processes and earthquake focal mechanisms. Therefore, we can utilize the similarity of repeating earthquakes to better constrain the focal mechanism of each repeating earthquake sequence. For each repeating earthquake sequence, we obtain available first- motion polarities and S-/P- wave amplitude ratios from all earthquakes in the sequence at each station, calculate the median values of polarities and S-/P- wave amplitude ratios, and assign the values to the station. Then we use the median polarities and S-/P- wave amplitude ratios to calculate the focal mechanism of each repeating earthquake sequence using the HASH algorithm<sup>62</sup>. By doing this process, we can both reduce the errors caused by manual picking and temporal noises in the waveforms and make full use of all available stations in the study time period for focal mechanism calculation. We obtained focal mechanisms of 386 repeating earthquake sequences and the catalog is available via (https://data.mendeley.com/datasets/34szj3jm6k/1).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[84, 90, 666, 111]]<|/det|>
+## Fault geometry estimation using repeating earthquake and seismicity  
+
+<|ref|>text<|/ref|><|det|>[[81, 123, 904, 494]]<|/det|>
+In this study, we determine the main fault geometry (representative strike and dip and average fault- zone width of each segment) using earthquake locations at depth. Since small earthquakes may occur around the main fault instead of on- fault, we first determine the horizontal location of the primary fault strand using the locations of repeating earthquake sequences with at least 2 repeaters (red dots in Fig. 3a- b) \(^{28}\) , which generally indicate localized aseismic slip of major faults. We then choose earthquakes within 1- km epicentral distance from the horizontal fault trace to determine the 3D fault geometry by applying the principal component analysis \(^{30}\) to minimize the orthogonal hypocentral distances to the fitted fault plane in each fault segment. For each 15- km- long 15- km- deep fault segment stepping at 1- km intervals along the fault trace, we calculate strike and dip of the plane that minimizes the distance between earthquake hypocenters and the plane and assign the values to the center of the fault segment (Fig. 3d).  
+
+<|ref|>sub_title<|/ref|><|det|>[[84, 515, 557, 536]]<|/det|>
+## Fault creep rate estimation using repeating earthquakes  
+
+<|ref|>text<|/ref|><|det|>[[81, 548, 905, 708]]<|/det|>
+Repeating earthquakes are events that repeatedly rupture particular fault patches \(^{16}\) , which can be detected by waveform similarity \(^{25,63}\) and can be used to illuminate the spatiotemporal variations of fault creep rate at depth \(^{26,64}\) . Here, we use the repeating earthquake sequences with more than 10 repeaters from a Northern California repeating earthquake catalog \(^{28}\) to estimate the creep rate variations along the CSAF.  
+
+<|ref|>text<|/ref|><|det|>[[84, 723, 850, 777]]<|/det|>
+The cumulative fault slip of and surrounding an earthquake patch over one seismic cycle in a repeating earthquake sequence can be estimated following the empirical scaling relationship  
+
+<|ref|>equation<|/ref|><|det|>[[459, 790, 565, 816]]<|/det|>
+\[d = 10^{\alpha}M_{0}^{\beta},\]  
+
+<|ref|>text<|/ref|><|det|>[[81, 831, 912, 890]]<|/det|>
+where \(d\) is slip in centimeter and \(M_{0}\) is seismic moment in dyne·cm, converted from the NCSN preferred magnitude \(M_{l}\) using the empirical relationship \(\log (M_{0}) = 1.6M_{l} + 15.8^{65}\) . The empirical values and are
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[81, 88, 895, 108]]<|/det|>
+\(\alpha = - 2.36\) and \(\beta = 0.16\) based on comparison with the geodetically inferred creep rate at Parkfield<sup>26</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[82, 123, 894, 180]]<|/det|>
+To compare with the modeling results, we obtain the averaged slip rate of all repeating earthquakes in each \(3 \times 3 \mathrm{~km}\) fault patch in Fig. 4a and assign the value as the slip rate of the fault patch (Fig. 3c,  
+
+<|ref|>text<|/ref|><|det|>[[82, 194, 907, 250]]<|/det|>
+Supplementary Figs. 4a, 5a). If the \(3 \times 3 \mathrm{~km}\) fault patch lacks repeating earthquake focal mechanisms, we will broaden our search to include a \(9 \times 9 \mathrm{~km}\) fault area centered around the same location.  
+
+<|ref|>sub_title<|/ref|><|det|>[[84, 273, 774, 295]]<|/det|>
+## Local fault slip direction estimation using repeating earthquake focal mechanisms  
+
+<|ref|>text<|/ref|><|det|>[[80, 306, 912, 748]]<|/det|>
+Since repeating earthquakes are closely located around the main fault with strike orientation highly consistent with the main fault (Fig. 2), we assume that most repeating earthquakes are located on the main fault and represent the local fault slip behaviors. We first obtain the nodal plane whose strike angle has a smaller azimuthal difference from the main fault orientation. We then combine the dip and rake angles of the best- fitting nodal plane to estimate the slip direction of each repeating earthquake sequence (Supplementary Figs. 3, 4). Since we use the NE side of the main fault as the reference and the rake angle represent the moving direction of the hanging wall, we treat the rake direction as the slip direction when the nodal plane is dipping to the NE and use the opposite of the rake direction as slip direction when the nodal plane is dipping to the SW (Supplementary Fig. 4). To compare with the modeling results, we obtain the averaged slip direction of all repeating earthquakes in each \(3 \times 3 \mathrm{~km}\) fault patch in Fig. 4a and assign the value as the slip direction of the fault patch (Fig. 3e, Supplementary Fig. 5). If the \(3 \times 3 \mathrm{~km}\) fault patch lacks repeating earthquake focal mechanisms, we will broaden our search to include a \(9 \times 9 \mathrm{~km}\) fault area centered around the same location.  
+
+<|ref|>sub_title<|/ref|><|det|>[[82, 771, 859, 825]]<|/det|>
+## Estimations of stress orientation \((SH_{\mathrm{max}})\) and faulting style \((Perc_{rake > 0})\) using \(M_{l} \geq 1\) focal mechanisms  
+
+<|ref|>text<|/ref|><|det|>[[82, 840, 911, 896]]<|/det|>
+Focal mechanisms contain valuable information about fault geometry, kinematics and stress state in the crust. We use \(24,915 M_{l} \geq 1\) focal mechanisms located within \(2 \mathrm{~km}\) from the main fault to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[81, 88, 904, 250]]<|/det|>
+estimate stress orientation \(SH_{\mathrm{max}}\) and fault style \(Perc_{rake > 0}\) . We only use focal mechanisms with more than 8 polarities and \(< 35\) degrees uncertainties for quality control. To compare with the modeling results, we obtain these values in each \(3 \times 3 \mathrm{~km}\) fault patch in Fig. 4a when there are more than 100 focal mechanisms in the grid (Fig. 5b). If the \(3 \times 3 \mathrm{~km}\) fault patch lacks sufficient focal mechanisms, we will broaden our search to include a \(9 \times 9 \mathrm{~km}\) fault area centered around the same location.  
+
+<|ref|>text<|/ref|><|det|>[[80, 272, 914, 576]]<|/det|>
+For stress orientation estimation, we iteratively inverse stress using the STRESSINVERSE program<sup>66</sup>. This method is modified from Michael's method (1987) that jointly inverse stress and fault orientations by selecting the nodal plane with higher value of instability \(I^{67}\) . For earthquake faulting style, it is usually classified into normal, reverse, or strike- slip earthquakes based on rake angles<sup>68</sup>. Here, in order to represent the faulting style of a group of focal mechanisms, we simplify the classification into two types: oblique- reverse- faulting events with rake angle larger than \(0^{\circ}\) and oblique- normal- faulting events with rake angle equal to or smaller than \(0^{\circ}\) so that the summation of the percentages of oblique- normal- faulting and oblique- reverse- faulting events equals \(100\%\) . In this study, we use the percentage of oblique- reverse- faulting events \((Perc_{rake > 0})\) to represent the earthquake faulting style.  
+
+<|ref|>sub_title<|/ref|><|det|>[[85, 589, 353, 608]]<|/det|>
+## Data and materials availability:  
+
+<|ref|>text<|/ref|><|det|>[[81, 621, 910, 818]]<|/det|>
+The InSAR surface slip- rate estimates were obtained from Jolivet et al., (2015)<sup>10</sup>. The repeating earthquake catalog was obtained from Waldhauser and Schaff (2021)<sup>28</sup>. The relocated earthquake catalog was obtained from Waldhauser and Schaff (2008)<sup>24</sup>. The earthquake phase information and seismic waveforms are taken from the Northern California Earthquake Data Center, Northern California Seismic Network (https://ncedc.org/ncsn/). The estimated earthquake focal mechanisms are available at Mendeley Data (https://data.mendeley.com/datasets/34szj3jm6k/1).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[84, 91, 181, 108]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[110, 123, 911, 184]]<|/det|>
+1. Chester, F. M. & Chester, J. S. Ultracataclasite structure and friction processes of the Punchbowl fault, San Andreas system, California. Tectonophysics, 295(1-2), 199-221 (1998).  
+
+<|ref|>text<|/ref|><|det|>[[112, 194, 787, 250]]<|/det|>
+2. Ben-Zion, Y. & Sammis, C. G. Characterization of fault zones. Pure and applied geophysics, 160(3), 677-715 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[112, 262, 901, 353]]<|/det|>
+3. Mitchell, T. M. & Faulkner, D. R. The nature and origin of off-fault damage surrounding strike-slip fault zones with a wide range of displacements: A field study from the Atacama fault system, northern Chile. Journal of Structural Geology, 31(8), 802-816, (2009).  
+
+<|ref|>text<|/ref|><|det|>[[112, 366, 885, 422]]<|/det|>
+4. Peng, Z. & Gomberg, J. An integrated perspective of the continuum between earthquakes and slow-slip phenomena. Nature geoscience, 3(9), 599-607 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[112, 435, 872, 491]]<|/det|>
+5. Harris, R. A. Large earthquakes and creeping faults. Reviews of Geophysics, 55(1), 169-198 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[112, 504, 890, 561]]<|/det|>
+6. Bürgmann, R. The geophysics, geology and mechanics of slow fault slip. Earth and Planetary Science Letters, 495, 112-134 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[112, 573, 897, 666]]<|/det|>
+7. Perfettini, H., Avouac, J. P., Tavera, H., Kositsky, A., Nocquet, J. M., Bondoux, F., Chlieh, M., Sladen, A., Audin, L., Farber, D. L. & Soler, P. Seismic and aseismic slip on the Central Peru megathrust. Nature, 465(7294), 78-81 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[112, 679, 899, 806]]<|/det|>
+8. Chlieh, M., Mothes, P. A., Nocquet, J. M., Jarrin, P., Charvis, P., Cisneros, D., Font, Y., Collot, J. Y., Villegas-Lanza, J. C., Rolandone, F., Vallée, M., Regnier, M., Segovia M., Martin X. & Yepes, H. Distribution of discrete seismic asperities and aseismic slip along the Ecuadorian megathrust. Earth and Planetary Science Letters, 400, 292-301 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[112, 818, 900, 875]]<|/det|>
+9. Avouac, J. P. From geodetic imaging of seismic and aseismic fault slip to dynamic modeling of the seismic cycle. Annual Review of Earth and Planetary Sciences, 43, 233-271 (2015).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 88, 904, 147]]<|/det|>
+10. Malservisi, R., Gans, C. & Furlong, K. P. Numerical modeling of strike-slip creeping faults and implications for the Hayward fault, California. Tectonophysics, 361(1-2), 121-137 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[113, 158, 904, 250]]<|/det|>
+11. Jolivet, R., Simons, M., Agram, P. S., Duputel, Z. & Shen, Z. K. Aseismic slip and seismogenic coupling along the central San Andreas Fault. Geophysical Research Letters, 42(2), 297-306 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[113, 263, 911, 353]]<|/det|>
+12. Maurer, J. & Johnson, K. Fault coupling and potential for earthquakes on the creeping section of the central San Andreas Fault. Journal of Geophysical Research: Solid Earth, 119(5), 4414-4428 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[113, 366, 907, 424]]<|/det|>
+13. Ryder, I. & Bürgmann, R. Spatial variations in slip deficit on the central San Andreas Fault from InSAR. Geophysical Journal International, 175(3), 837-852 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[113, 436, 896, 529]]<|/det|>
+14. Malservisi, R., Furlong, K. P. & Gans, C. R. Microseismicity and creeping faults: Hints from modeling the Hayward fault, California (USA). Earth and Planetary Science Letters, 234(3-4), 421-435 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[113, 541, 907, 633]]<|/det|>
+15. Khoshmanesh, M., Shirzaei, M. & Nadeau, R. M. Time-dependent model of aseismic slip on the central San Andreas Fault from InSAR time series and repeating earthquakes. Journal of Geophysical Research: Solid Earth, 120(9), 6658-6679 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[113, 645, 869, 703]]<|/det|>
+16. Uchida, N. & Bürgmann, R. Repeating earthquakes. Annual Review of Earth and Planetary Sciences, 47(1), 305-332 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[113, 715, 860, 807]]<|/det|>
+17. Tormann, T., Wiemer, S., Metzger, S., Michael, A. & Hardebeck, J. L. Size distribution of Parkfield's microearthquakes reflects changes in surface creep rate. Geophysical Journal International, 193(3), 1474-1478 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[113, 819, 877, 877]]<|/det|>
+18. Liu, Y. K., Ross, Z. E., Cochran, E. S. & Lapusta, N. A unified perspective of seismicity and fault coupling along the San Andreas Fault. Science advances, 8(8), eabk1167 (2022).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 88, 897, 145]]<|/det|>
+19. Bürgmann, R., Fielding, E. & Sukhatme, J. Slip along the Hayward fault, California, estimated from space-based synthetic aperture radar interferometry. Geology, 26(6), 559-562 (1998).  
+
+<|ref|>text<|/ref|><|det|>[[110, 157, 897, 214]]<|/det|>
+20. Mia, M. S., Abdelmeguid, M. & Elbanna, A. E. Spatio-Temporal Clustering of Seismicity Enabled by Off-Fault Plasticity. Geophysical Research Letters, 49(8), e2021GL097601 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[110, 227, 904, 354]]<|/det|>
+21. Titus, S. J., DeMets, C. & Tikoff, B. Thirty-five-year creep rates for the creeping segment of the San Andreas fault and the effects of the 2004 Parkfield earthquake: Constraints from alignment arrays, continuous global positioning system, and creepmeters. Bulletin of the Seismological Society of America, 96(4B), S250-S268 (2006).  
+
+<|ref|>text<|/ref|><|det|>[[110, 366, 900, 459]]<|/det|>
+22. Titus, S. J., Dyson, M., DeMets, C., Tikoff, B., Rolandone, F. & Bürgmann, R. Geologic versus geodetic deformation adjacent to the San Andreas fault, central California. Bulletin, 123(5-6), 794-820 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[110, 472, 877, 528]]<|/det|>
+23. Noda, H. & Lapusta, N. Stable creeping fault segments can become destructive as a result of dynamic weakening. Nature, 493(7433), 518-521 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[110, 541, 879, 632]]<|/det|>
+24. Waldhauser, F. & Schaff, D. P. Large-scale relocation of two decades of Northern California seismicity using cross-correlation and double-difference methods. Journal of Geophysical Research: Solid Earth, 113(B8) (2008).  
+
+<|ref|>text<|/ref|><|det|>[[110, 645, 909, 736]]<|/det|>
+25. Nadeau, R. M., Foxall, W. & McEvilly, T. V. Clustering and periodic recurrence of microearthquakes on the San Andreas fault at Parkfield, California. Science, 267(5197), 503-507 (1995).  
+
+<|ref|>text<|/ref|><|det|>[[110, 749, 910, 840]]<|/det|>
+26. Nadeau, R. M. & Johnson, L. R. Seismological studies at Parkfield VI: Moment release rates and estimates of source parameters for small repeating earthquakes. Bulletin of the Seismological Society of America, 88(3), 790-814 (1998).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 88, 897, 180]]<|/det|>
+27. Turner, R. C., Shirzaei, M., Nadeau, R. M. & Bürgmann, R. Slow and Go: Pulsing slip rates on the creeping section of the San Andreas Fault. Journal of Geophysical Research: Solid Earth, 120(8), 5940-5951 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[111, 193, 900, 285]]<|/det|>
+28. Waldhauser, F. & Schaff, D. P. A comprehensive search for repeating earthquakes in northern California: Implications for fault creep, slip rates, slip partitioning, and transient stress. Journal of Geophysical Research: Solid Earth, 126(11), e2021JB022495 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[111, 297, 905, 389]]<|/det|>
+29. Cheng, Y., Allen, R. M. & Taira, T. A. A new focal mechanism calculation algorithm (REFOC) using inter-event relative radiation patterns: Application to the earthquakes in the Parkfield area, Journal of Geophysical Research: Solid Earth, 128(3), e2022JB025006 (2023).  
+
+<|ref|>text<|/ref|><|det|>[[111, 401, 855, 458]]<|/det|>
+30. Abdi, H. & Williams, L. J. Principal component analysis. Wiley interdisciplinary reviews: computational statistics, 2(4), 433-459 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[111, 472, 893, 563]]<|/det|>
+31. Thomas, A. Poly3D: A three-dimensional, polygonal element, displacement boundary element computer program with applications to fractures, faults, and cavities in the Earth's crust [MS thesis] (1993).  
+
+<|ref|>text<|/ref|><|det|>[[111, 577, 904, 633]]<|/det|>
+32. Schulz, S. S., Mavko, G. M., Burford, R. O. & Stuart, W. D. Long-term fault creep observations in central California. Journal of Geophysical Research: Solid Earth, 87(B8), 6977-6982 (1982).  
+
+<|ref|>text<|/ref|><|det|>[[111, 646, 895, 736]]<|/det|>
+33. Toppozada, T. R., Branum, D. M., Reichle, M. S. & Hallstrom, C. L. San Andreas fault zone, California: M≥ 5.5 earthquake history. Bulletin of the Seismological Society of America, 92(7), 2555-2601 (2002).  
+
+<|ref|>text<|/ref|><|det|>[[111, 750, 910, 842]]<|/det|>
+34. Ben-Zion, Y., Rice, J. R. & Dmowska, R. Interaction of the San Andreas fault creeping segment with adjacent great rupture zones and earthquake recurrence at Parkfield. Journal of Geophysical Research: Solid Earth, 98(B2), 2135-2144 (1993).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 88, 888, 179]]<|/det|>
+35. Sibson, R. H. Fault zone models, heat flow, and the depth distribution of earthquakes in the continental crust of the United States. Bulletin of the Seismological Society of America, 72(1), 151-163 (1982).  
+
+<|ref|>text<|/ref|><|det|>[[110, 192, 884, 249]]<|/det|>
+36. Dieterich, J. H. Time-dependent friction and the mechanics of stick-slip. In Rock friction and earthquake prediction (pp. 790-806). Birkhäuser, Basel (1978).  
+
+<|ref|>text<|/ref|><|det|>[[110, 262, 900, 319]]<|/det|>
+37. Lockner, D. A., Morrow, C., Moore, D. & Hickman, S. Low strength of deep San Andreas fault gouge from SAFOD core. Nature, 472(7341), 82-85 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[110, 332, 888, 389]]<|/det|>
+38. Carpenter, B. M., Marone, C. & Saffer, D. M. Weakness of the San Andreas Fault revealed by samples from the active fault zone. Nature Geoscience, 4(4), 251-254 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[110, 401, 880, 494]]<|/det|>
+39. Carpenter, B. M., Saffer, D. M. & Marone, C. Frictional properties of the active San Andreas Fault at SAFOD: Implications for fault strength and slip behavior. Journal of Geophysical Research: Solid Earth, 120(7), 5273-5289 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[110, 507, 876, 564]]<|/det|>
+40. Khoshmanesh, M. & Shirzaei, M. Episodic creep events on the San Andreas Fault caused by pore pressure variations. Nature geoscience, 11(8), 610-614 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 576, 876, 633]]<|/det|>
+41. Noda, H. & Lapusta, N. Stable creeping fault segments can become destructive as a result of dynamic weakening. Nature, 493(7433), 518-521 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[110, 645, 880, 739]]<|/det|>
+42. Schwartz, D. P. Past and future fault rupture lengths in seismic source characterization—The long and short of it. Bulletin of the Seismological Society of America, 108(5A), 2493-2520 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 750, 907, 844]]<|/det|>
+43. Zoback, M. D., Zoback, M. L., Mount, V. S., Suppe, J., Eaton, J. P., Healy, J. H., Oppenheimer, D., Reasenberg, P., Jones, L., Raleigh, C. B., Wong, I. G., Scotti, O. & Wentworth, C. New evidence on the state of stress of the San Andreas fault system, Science, 238, 1105-1111, (1987).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 88, 905, 180]]<|/det|>
+44. Provost, A. S. & Houston, H. Stress orientations in northern and central California: Evidence for the evolution of frictional strength along the San Andreas plate boundary system. Journal of Geophysical Research: Solid Earth, 108(B3), (2003).  
+
+<|ref|>text<|/ref|><|det|>[[110, 192, 895, 249]]<|/det|>
+45. Hardebeck, J. L. & Hauksson, E. Role of fluids in faulting inferred from stress field signatures. Science, 285(5425), 236-239 (1999).  
+
+<|ref|>text<|/ref|><|det|>[[110, 263, 894, 320]]<|/det|>
+46. Rice, J. R. Fault stress states, pore pressure distributions, and the weakness of the San Andreas fault. In International geophysics (Vol. 51, pp. 475-503). Academic Press (1992).  
+
+<|ref|>text<|/ref|><|det|>[[110, 332, 875, 423]]<|/det|>
+47. Chéry, J., Zoback, M. D. & Hassani, R. An integrated mechanical model of the San Andreas fault in central and northern California. Journal of Geophysical Research: Solid Earth, 106(B10), 22051-22066 (2001).  
+
+<|ref|>text<|/ref|><|det|>[[110, 436, 875, 463]]<|/det|>
+48. Scholz, C. H. A fault in the 'weak San Andreas' theory. Nature, 406(6793), 234-234 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[110, 473, 888, 565]]<|/det|>
+49. Lynch, J. C. & Richards, M. A. Finite element models of stress orientations in well-developed strike-slip fault zones: Implications for the distribution of lower crustal strain. Journal of Geophysical Research: Solid Earth, 106(B11), 26707-26729 (2001).  
+
+<|ref|>text<|/ref|><|det|>[[110, 576, 912, 633]]<|/det|>
+50. Brune, J. N., Henyey, T. L. & Roy, R. F. Heat flow, stress, and rate of slip along the San Andreas fault, California. Journal of Geophysical Research, 74(15), 3821-3827 (1969).  
+
+<|ref|>text<|/ref|><|det|>[[110, 645, 806, 702]]<|/det|>
+51. Lachenbruch, A. H. & Sass, J. H. Heat flow and energetics of the San Andreas fault zone. Journal of Geophysical Research: Solid Earth, 85(B11), 6185-6222 (1980).  
+
+<|ref|>text<|/ref|><|det|>[[110, 714, 909, 806]]<|/det|>
+52. Wibberley, C. A., Yielding, G., & Di Toro, G. Recent advances in the understanding of fault zone internal structure: a review. Geological Society, London, Special Publications, 299(1), 5-33 (2008).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 88, 909, 178]]<|/det|>
+53. Cheng, Y. & Ben-Zion, Y. Variations of earthquake properties before, during, and after the 2019 M7.1 Ridgecrest, CA, earthquake. Geophysical Research Letters, 47(18), e2020GL089650 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[111, 193, 790, 283]]<|/det|>
+54. Xu, W., Feng, G., Meng, L., Zhang, A., Ampuero, J. P., Bürgmann, R. & Fang, L. Transpressional rupture cascade of the 2016 Mw 7.8 Kaikoura earthquake, New Zealand. Journal of Geophysical Research: Solid Earth, 123(3), 2396-2409 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[111, 297, 901, 355]]<|/det|>
+55. Kodaira, S., Iidaka, T., Kato, A., Park, J. O., Iwasaki, T. & Kaneda, Y. High pore fluid pressure may cause silent slip in the Nankai Trough. Science, 304(5675), 1295-1298 (2004).  
+
+<|ref|>text<|/ref|><|det|>[[111, 367, 900, 459]]<|/det|>
+56. Saffer, D. M. & Marone, C. Comparison of smectite-and illite-rich gouge frictional properties: application to the updip limit of the seismogenic zone along subduction megathrusts. Earth and Planetary Science Letters, 215(1-2), 219-235 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[111, 472, 900, 566]]<|/det|>
+57. d'Alessio, M. A., Williams, C. F. & Bürgmann, R. Frictional strength heterogeneity and surface heat flow: Implications for the strength of the creeping San Andreas fault. Journal of Geophysical Research: Solid Earth, 111(B5) (2006).  
+
+<|ref|>text<|/ref|><|det|>[[111, 578, 894, 667]]<|/det|>
+58. Xu, S., Fukuyama, E., Yamashita, F., Kawakata, H., Mizoguchi, K. & Takizawa, S. Fault strength and rupture process controlled by fault surface topography. Nature Geoscience, 16(1), 94-100 (2023).  
+
+<|ref|>text<|/ref|><|det|>[[111, 681, 874, 738]]<|/det|>
+59. Jiang, J., Bock, Y. & Klein, E. Coevolving early afterslip and aftershock signatures of a San Andreas fault rupture. Science Advances, 7(15), eabc1606 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[111, 751, 874, 808]]<|/det|>
+60. Okada, Y. Internal deformation due to shear and tensile faults in a half-space. Bulletin of the seismological society of America, 82(2), 1018-1040 (1992).  
+
+<|ref|>text<|/ref|><|det|>[[111, 821, 912, 878]]<|/det|>
+61. Hardebeck, J. L. & Okada, T. Temporal stress changes caused by earthquakes: A review. Journal of Geophysical Research: Solid Earth, 123(2), 1350-1365 (2018).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 88, 907, 149]]<|/det|>
+62. Hardebeck, J. L. & Shearer, P. M. Using S/P amplitude ratios to constrain the focal mechanisms of small earthquakes. Bulletin of the Seismological Society of America, 93(6), 2434-2444 (2003). 
+
+<|ref|>text<|/ref|><|det|>[[110, 160, 850, 214]]<|/det|>
+63. Vidale, J. E., Ellsworth, W. L., Cole, A. & Marone, C. Variations in rupture process with recurrence interval in a repeated small earthquake. Nature, 368(6472), 624-626 (1994). 
+
+<|ref|>text<|/ref|><|det|>[[110, 229, 904, 284]]<|/det|>
+64. Nadeau, R. M. & McEvilly, T. V. Fault slip rates at depth from recurrence intervals of repeating microearthquakes. Science, 285(5428), 718-721 (1999). 
+
+<|ref|>text<|/ref|><|det|>[[110, 299, 895, 389]]<|/det|>
+65. Wyss, M., Sammis, C. G., Nadeau, R. M. & Wiemer, S. Fractal dimension and b-value on creeping and locked patches of the San Andreas fault near Parkfield, California. Bulletin of the Seismological Society of America, 94(2), 410-421 (2004). 
+
+<|ref|>text<|/ref|><|det|>[[110, 404, 779, 458]]<|/det|>
+66. Vavryčuk, V. Iterative joint inversion for stress and fault orientations from focal mechanisms. Geophysical Journal International, 199(1), 69-77 (2014). 
+
+<|ref|>text<|/ref|><|det|>[[110, 473, 890, 564]]<|/det|>
+67. Lund, B. & Slunga, R. Stress tensor inversion using detailed microearthquake information and stability constraints: Application to Ölfus in southwest Iceland. Journal of Geophysical Research: Solid Earth, 104(B7), 14947-14964 (1999). 
+
+<|ref|>text<|/ref|><|det|>[[110, 579, 816, 667]]<|/det|>
+68. Yang, W., Hauksson, E. & Shearer, P. M. Computing a large refined catalog of focal mechanisms for southern California (1981–2010): Temporal stability of the style of faulting. Bulletin of the Seismological Society of America, 102(3), 1179-1194 (2012).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 90, 912, 325]]<|/det|>
+Acknowledgments: We thank Shiqing Xu, Taka'aki Taira, Naoki Uchida, and Saeko Kita for valuable comments and suggestions. Funding: This work was funded by the California Governor's Office of Emergency Services (Cal OES), Agreement Number 6172- 2018. Author contributions: Y.C. and R.B. designed the study. Y.C. processed all datasets and performed modeling and analyses. Y.C., R.B., and R.A. all contribute to discussions. Y.C. lead the writing of the manuscript with contributions from R.B. and R.A. Competing interests: The authors declare that they have no competing interests.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[100, 133, 912, 500]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[79, 519, 875, 540]]<|/det|>
+<center>Fig. 1 Improvements in constraining subsurface coupling using earthquake focal mechanisms. </center>  
+
+<|ref|>text<|/ref|><|det|>[[80, 552, 911, 747]]<|/det|>
+Schematic comparison of data and assumptions used in a previous geodetic inversion studies \(^{11,12,13,15}\) and b this study relying on the spatial distribution of earthquakes and slip directions from focal mechanisms. Dark red areas denote locked patches and light red areas denote creeping areas. Modeled fault c creep rate and d creep direction around a locked patch. Yellow stars and beachballs denote earthquake locations and focal mechanisms, respectively. Black and blue beachballs are on- fault and off- fault earthquakes, respectively. Red arrows show the assumed fault slip direction.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[252, 92, 737, 660]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[84, 675, 910, 907]]<|/det|>
+<center>Fig. 2 Seismicity distribution along the CSAF. a Map view and b cross-section view of relocated earthquakes from 1984 to 2015 (Waldhauser and Schaff, 2008, extended to later years) along the central San Andreas Fault. Events in the box XX' are colored by faulting style. Small beach balls denote \(M \geq 4.0\) earthquakes. Large beach balls denote the 2003 \(M_{w}6.5\) San Simeon and 2004 \(M_{w}6.0\) Parkfield earthquakes' focal mechanisms. White stars denote historic major earthquakes. Note vertical exaggeration \(\mathrm{VE} = 3.13\) in b. The local coordinate system has its origin at latitude \(35.867^{\circ}\mathrm{N}\) , longitude \(120.447^{\circ}\mathrm{W}\) and is oriented \(\mathrm{N42^{\circ}W}\) . </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[87, 87, 884, 597]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[88, 608, 905, 911]]<|/det|>
+<center>Fig. 3 Spatiotemporal variation of seismicity and its relationship with the main fault. a Rotated map view, b cross-section view, and c spatiotemporal variations of seismicity (black dots), \(M \geq 4.0\) earthquakes (blue beachballs), and repeating earthquakes (26; red dots) along the Central San Andreas Fault. The local coordinate system has its origin at latitude \(35.867^{\circ}\mathrm{N}\) , longitude \(120.447^{\circ}\mathrm{W}\) and is oriented \(N42^{\circ}\mathrm{W}\) . d Spatial variations of fault strike (red line) and dip (blue line) angles determined using PCA analysis from relocated seismicity around each 15-km-long 15-km-deep fault segment stepping at 1-km intervals along the fault. The reference fault strike and dip are \(N138^{\circ}\mathrm{E}\) and \(90^{\circ}\) (vertical), respectively. Cumulative Density Functions (CDF) of the e horizontal distance of events from main fault and f azimuthal difference from main fault strike </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[141, 88, 907, 250]]<|/det|>
+for earthquakes located within 1 km horizontal distance from the main fault based on focalmechanism nodal plane closest to the main fault. Black, blue, and red curves denote \(M \geq 1.0\) earthquakes, \(M \geq 4.0\) earthquakes, and repeating earthquake sequences, respectively. SJB: San Juan Bautista, MR: Melendy Ranch; BW: Bitterwater; SC: Slack Canyon; SAFOD: San Andreas Fault Observatory at Depth; MM: Middle Mountain; PK: Parkfield; GH: Gold Hill.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[90, 90, 884, 660]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[85, 677, 905, 909]]<|/det|>
+<center>Fig. 4 Modeled and observed on-fault displacement. a Fault model 1 setup with freely slipping zones (blue), locked sections (gray) and constant-rate creeping zones (red). The deep creeping zone driving the shallow creep extends from \(15\mathrm{km}\) to \(2000\mathrm{km}\) depth and far beyond the lateral ends of the CSAF. The shallow fault is fully coupled beyond \(166\mathrm{km}\) NW and -17km SE of the fault. The size of each fault patch is \(3\times 3\mathrm{km}\) . White stars denote \(M\geq 4.0\) earthquakes. b Modeled and c observed fault creep rate estimated from the occurrence of repeating earthquakes. d Modeled and e observed fault slip direction estimated from the rake and dip of repeating earthquake focal </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[141, 88, 904, 250]]<|/det|>
+mechanisms. Colored dots and grids in c and e represent the values from each individual repeating earthquake sequence and those averaged in each spatial bin. Positive creep directions indicate a NE- side- up dip- slip component. f Modeled (blue) and observed surface creep rate (red) estimated from InSAR data (Jolivet et al., 2014). Green circles and squares denote average surface creep rates from creepmeters and alignment arrays (circle: 30; square: 19), respectively.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[88, 90, 905, 590]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[80, 602, 904, 870]]<|/det|>
+<center>Fig. 5 Off-fault stress field and kinematics. Comparison of a the angle between the main fault and the modeled off-fault maximum horizontal stress orientation \((\theta)\) 1.5 km NE of the main fault with b the angle between the main fault and the observed maximum horizontal stress orientation calculated from \(M\geq 1.0\) focal mechanisms located within 2 km around the fault trace. White dots denote the locations of focal mechanisms used in stress inversion. c The percentage of oblique-reverse-faulting events \((P e r c_{r a k e > 0};\) red curve) for M1.0 focal mechanisms located within 2 km around the fault trace. d Point-to-point comparison of \(\theta\) in b and \(P e r c_{r a k e > 0}\) in c. e Schematic illustration showing the variation of fault zone structure, weakness, and stress field. </center>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[43, 92, 768, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 129, 425, 150]]<|/det|>
+- CSAFSupplementaryMaterials1127.pdf
+
+<--- Page Split --->

preprint/preprint__7e8d9494866549fe6fb82c521aaa52f5960e0b5ad6ca19e409ea66d4b8156b98/images_list.json

@@ -0,0 +1,77 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. Anti-SARS-CoV-2 IgG and IgM antibody responses following vaccination",
+    "footnote": [],
+    "bbox": [
+      [
+        140,
+        70,
+        884,
+        333
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Paired maternal, cord, and infant IgG and neutralization antibodies",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        70,
+        865,
+        750
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. Neutralization to IgG antibody correlation",
+    "footnote": [],
+    "bbox": [
+      [
+        118,
+        68,
+        875,
+        272
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4. Maternal delivery and Cord-to-maternal antibody transfer ratios timing",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        95,
+        875,
+        420
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5. PhIP-seq/VirScan paired maternal and cord SARS-CoV-2 Spike protein epitope",
+    "footnote": [],
+    "bbox": [
+      [
+        115,
+        70,
+        866,
+        744
+      ]
+    ],
+    "page_idx": 26
+  }
+]

preprint/preprint__7e8d9494866549fe6fb82c521aaa52f5960e0b5ad6ca19e409ea66d4b8156b98/preprint__7e8d9494866549fe6fb82c521aaa52f5960e0b5ad6ca19e409ea66d4b8156b98.mmd

@@ -0,0 +1,420 @@
+
+# Evaluation of transplacental transfer of mRNA vaccine products and functional antibodies during pregnancy and early infancy  
+
+Mary Prahl ( \(\boxed{\boxed{\pi}}\) mary.prahl@ucsf.edu) University of California, San Francisco  
+
+Yarden Golan University of California San Francisco  
+
+Arianna Cassidy University of California San Francisco  
+
+Yusuke Matsui Gladstone Institute  
+
+Lin Li University of California San Francisco  
+
+Bonny Alvarenga University of California, San Francisco https://orcid.org/0000- 0001- 7922- 9454  
+
+Hao Chen University of California San Francisco  
+
+Unurzul Jigmeddagva University of California San Francisco  
+
+Christine Lin University of California San Francisco  
+
+Veronica Gonzalez University of California San Francisco  
+
+Megan Chidboy University of California San Francisco  
+
+Lakshmi Warrier University of California San Francisco  
+
+Sirirak Buarpung University of California San Francisco  
+
+Amy Murtha UCSF  
+
+Valerie Flaherman University of California San Francisco  
+
+Warner Greene
+
+<--- Page Split --->
+
+J. David Gladstone Institutes  
+
+Alan Wu University of California  
+
+Kara Lynch UCSF  
+
+Jayant Rajan University of California, San Francisco  
+
+Stephanie Gaw University of California, San Francisco https://orcid.org/0000- 0003- 0891- 6964  
+
+## Article  
+
+Keywords: SARS- CoV- 2, COVID- 19, Pregnancy, Vaccine, Antibody, Neonatal Immunity, Neutralizing 47 Antibody, Phage Array, mRNA Vaccination, BNT- 162b2, mRNA- 1273, Placenta, Cord Blood  
+
+Posted Date: December 15th, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 1150427/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 July 30th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 32188- 1.
+
+<--- Page Split --->
+
+Evaluation of transplacental transfer of mRNA vaccine products and functional antibodies during pregnancy and early infancy  
+
+Mary Prahl, M.D., Yarden Golan, Ph.D., Arianna G. Cassidy, M.D., Yusuke Matsui M.D., Ph.D., Lin Li, M.D., PhD, Bonny Alvarenga, Hao Chen, Ph.D., Unurzul Jigmeddagva, M.D., Christine Y. Lin, Veronica J. Gonzalez, M.D, Megan A. Chidboy, Lakshmi Warrier, Sirirak Buarpung, D.V.M., Ph.D., Amy P. Murtha M.D., Valerie J. Flaherman, M.D., M.P.H., Warner C. Greene M.D., Ph.D., Alan H.B. Wu, Ph.D., Kara L. Lynch Ph.D., Jayant Rajan, M.D., Ph.D., Stephanie L. Gaw, M.D., Ph.D  
+
+10  
+
+Author AffiliationsFrom the Department of Pediatrics, University of California, San Francisco (M.P., V.J.F.), Division of Pediatric Infectious Diseases and Global Health, University of California, San Francisco (M.P.), Department of Bioengineering and Therapeutic Sciences, University of California, San Francisco (Y.G.), Department of Medicine, University of California, San Francisco (L.W., B.A., J.R.), Weill Institute for Neurosciences, Division of Neurology, University of California, San Francisco, CA (B.A.), Gladstone Center for HIV Cure Research, Gladstone Institute, San Francisco, CA (Y.M., W.C.G.) Departments of Medicine and Microbiology and Immunology, University of California, San Francisco (W.C.G.), Department of Laboratory Medicine, University of California, San Francisco (A.H.B.W., K.L.L.), Center for Reproductive Sciences, Department of Obstetrics, Gynecology, and Reproductive Sciences, University of California San Francisco (A.G.C., L.L., H.C., U.J., C.Y.L., V.J.G., M.A.C., S.B., A.P.M., S.L.G.). Co-Corresponding authors: 1) Mary Prahl- Division of Pediatric Infectious Diseases and Global Health, Department of Pediatrics, University of California San Francisco, 550 16<sup>th</sup> St. 4<sup>th</sup> Floor. San Francisco, CA 94158. Phone (415) 514-0510, Email: mary.prahl@ucsf.edu 2) Stephanie L.
+
+<--- Page Split --->
+
+Gaw- Division of Maternal- Fetal Medicine, Department of Obstetrics, Gynecology, and Reproductive Sciences, University of California San Francisco, 513 Parnassus Ave, HSE16, Box 0556, San Francisco, CA 94143. Phone: (415) 476- 0535. Email: Stephanie.Gaw@ucsf.edu  
+
+## Abstract  
+
+Studies are needed to evaluate the safety and effectiveness of mRNA SARS- CoV- 2 vaccination during pregnancy, and the levels of protection provided to their newborns through placental transfer of antibodies. We evaluated the transplacental transfer of mRNA vaccine products and functional anti- SARS- CoV- 2 antibodies during pregnancy and early infancy in a cohort of 20 individuals vaccinated during pregnancy. We found no evidence of mRNA vaccine products in maternal blood, placenta tissue, or cord blood at delivery. However, we found time- dependent efficient transfer of IgG and neutralizing antibodies to the neonate that persisted during early infancy. Additionally, using phage immunoprecipitation sequencing, we found a vaccine- specific signature of SARS- CoV- 2 Spike protein epitope binding that is transplacentally transferred during pregnancy. In conclusion, products of mRNA vaccines are not transferred to the fetus during pregnancy, however timing of vaccination during pregnancy is critical to ensure transplacental transfer of protective antibodies during early infancy.  
+
+## Keywords  
+
+SARS- CoV- 2, COVID- 19, Pregnancy, Vaccine, Antibody, Neonatal Immunity, Neutralizing Antibody, Phage Array, mRNA Vaccination, BNT- 162b2, mRNA- 1273, Placenta, Cord Blood  
+
+## Introduction  
+
+Growing evidence has shown that pregnant individuals are at higher risk for SARS- CoV- 2- related morbidity and mortality<sup>1-4</sup>. Despite this, vaccination uptake by pregnant individuals has been slower than the general population<sup>5</sup>, in part because of maternal concern of adverse
+
+<--- Page Split --->
+
+effects on the embryo or fetus, even with strong consensus recommendations for COVID- 19 vaccination prior to or during pregnancy from several medical societies. Pregnant individuals were excluded from initial vaccine trials, and complete data on safety, efficacy, optimal timing of the vaccine in pregnancy, or its impact on the fetus has been delayed, which may impact individual medical decision making. Current COVID- 19 vaccines fully approved and under emergency use in the United States include the mRNA vaccines BNT- 162b2 and mRNA- 1273, which target the SARS- CoV- 2 Spike protein and stimulate protective immune responses. In addition to protecting the mother against severe disease, vaccination during pregnancy may protect the newborn through passive transfer of maternal immunoglobulin. SARS- CoV- 2 infection and vaccination during pregnancy produces an IgG response that is transferred to the fetus. Evidence of newborn protection might help address maternal concerns about adverse effects. However, detailed studies of the transplacental transfer of vaccine products and vaccine- related antibody dynamics, functional properties, and persistence during infancy of transferred SARS- CoV- 2 antibodies are needed to provide such evidence.  
+
+We examined the transplacental transfer of mRNA vaccine products and humoral responses using samples from pregnant individuals and their infants vaccinated with either BNT- 162b2 or mRNA- 1273 mRNA vaccine during pregnancy.  
+
+## Results:  
+
+Cohort: We evaluated 20 pregnant individuals who received COVID- 19 mRNA vaccines during pregnancy and their infants. Participants were vaccinated between December 2020 and April 2021. Gestational age at first dose ranged from 13 weeks to 40 weeks (mean 31.2, SD 5.9 weeks). Nineteen participants delivered live, singleton infants between January 2021 through April 2021 at gestational ages ranging from 37.4 to 41.1 weeks (mean 39.2, SD 1.1 weeks). One participant who was vaccinated at 13 weeks had a termination of pregnancy due to a lethal skeletal dysplasia of genetic etiology at 20.4 weeks. Eight participants received BNT- 162b2
+
+<--- Page Split --->
+
+(Pfizer- BioNTech) and twelve received mRNA- 1273 (Moderna) vaccines. Eighteen participants received both vaccine doses prior to delivery, and two participants received the second dose after delivery. The time from first mRNA vaccine dose ranged from 6- 97 (mean 51, SD 24.3) days prior to delivery, time from the second dose ranged from 2- 75 (mean 32, SD 21.3) days prior to delivery, and in two participants 15 and 21 days after delivery. No participants received a \(3^{\text{rd}}\) dose prior to delivery. Infants born to vaccinated mothers were followed up at convenience time points ranging from age 3 weeks to 15 weeks of life (mean 8.3, SD 3.2). Further demographic data is detailed in Table S1.  
+
+## Vaccine mRNA products do not cross the placenta  
+
+To determine the transplacental transfer of mRNA vaccine derived products, we examined available maternal blood at delivery, placenta tissue, and cord blood for Spike protein by Western blot and vaccine mRNA by PCR. All available delivery samples (maternal blood, placental tissue, and cord blood) were negative for Spike protein by Western blot (Supp Figure 1, Supp Table 3) and did not have detectable levels of vaccine mRNA by PCR (Suppl Table 3). Together, this indicates that products of mRNA vaccination do not reach the fetus after vaccination during pregnancy at readily detectable levels.  
+
+## mRNA vaccination in pregnancy leads to a robust antibody response  
+
+Similar to prior studies \(^{14,15,17}\) , we found that mRNA vaccination during pregnancy led to an increase in anti- SARS- Cov- 2 IgG following dose 1 (n=7, mean 388.6, SD 224.8 RFU) and an even further robust increase after vaccination dose 2 (n=12, mean 3214, SD 1383 RFU). Anti- SARS- CoV- 2 IgM (n=7, mean 53.3, SD 50.2 RFU) was detected in two maternal participants following dose 1, but only 1 participant following dose 2 (n=12, mean 23.8, SD 17 RFU, Fig 1).
+
+<--- Page Split --->
+
+## Vaccine induced anti-SARS-CoV-2 IgG and neutralizing antibodies are transplacentially  
+
+## transferred  
+
+We then evaluated the transplacental transfer of maternally derived anti- SARS- CoV- 2 IgG antibodies to their infants. Maternal blood at delivery was available in 19/20 participants and cord blood was available in 17/20 participants. Anti- SARS- CoV- 2 IgG was detectable in \(94.7\%\) (18/19) of maternal blood samples at delivery (mean 3235, range [10, 7811] RFU). Anti- SARS- CoV- 2 IgG was detectable in \(88.2\%\) (15/17) cord blood samples (mean 2243, range [2, 4959] RFU). One participant received one mRNA vaccine dose 9 days prior to delivery, and both the maternal and cord blood were negative for IgG at the time of delivery. Another participant received two doses of mRNA vaccine (23 and 2 days) prior to delivery and maternal blood was positive at 55 RFU (positive cutoff \(>50\) RFU), however cord blood IgG was negative (Figure 2A). Maternal and cord blood anti- SARS- CoV- 2 IgG levels were moderately correlated, but not statistically significant ( \(p = 0.074\) , \(R_{s} = 0.446\) , Fig 2A). All cord blood samples were anti- SARS- CoV- 2 IgM negative.  
+
+We next evaluated the transplacental transfer of neutralizing antibody titers by a label- free surrogate neutralization assay (sVNT) from mother to cord blood. Maternal and cord blood at delivery had robust neutralizing responses (maternal \(n = 17\) , mean 220.2, range [0, 422]. Cord blood \(n = 16\) , mean 296.6, range [0, 485], Fig 2B). All mother- infant dyads with positive IgG serology at delivery had detectable transplacental transfer of neutralizing antibodies with the exception of one pair in which the mother was borderline IgG positive at delivery and cord blood was negative, for which both maternal and cord blood were negative for neutralizing titers (Fig 2B). However, maternal and cord blood neutralizing titers were not significantly correlated ( \(p = 0.361\) , \(R_{s} = - 0.244\) , Fig 2B). Taken together, this indicates that maternal mRNA vaccination induces functional neutralizing antibodies that are transferred to the infant.
+
+<--- Page Split --->
+
+## Maternally-derived vaccine induced anti-SARS-CoV-2 IgG and neutralizing antibodies  
+
+## persist through early infancy  
+
+A subset of infants was sampled at convenience timepoints during follow up (infant \(n = 11\) , weeks of life range [3,15] mean 8.3 weeks). Anti- SARS- CoV- 2 IgG levels were positive in \(81.8\%\) of infants at follow- up (9/11 infants, mean 1290, range [1, 3225] RFU, Fig 2A), with one infant still IgG positive at 12 weeks of age (Fig 2C). The two infants that were IgG negative at follow up were both born to mothers who received only one vaccine dose prior to delivery (6 and 9 days, respectively). One of these infants did not have paired maternal or cord blood available at the time of delivery for comparison, and the other was IgG negative in cord blood. Maternal and infant follow- up anti- SARS- CoV- 2 IgG levels were not significantly correlated; however, cord blood and infant follow- up IgG levels were significantly associated \((p = 0.492, R_{s} = 0.249\) and \(p = 0.021, R_{s} = 0.76\) , respectively, Fig 2A). All infants were IgM negative at the time of follow up. All infants with available IgG positive samples at follow up had detectable neutralizing titers \((n = 8\) , mean 154, range [41- 256], Fig 2B). Maternal and infant follow- up neutralizing titers were not significantly correlated, as well as cord and infant follow up neutralizing titers \((p = 0.665\) , \(R_{s} = - 0.191\) and \(p = 0.662, R_{s} = 0.214\) , respectively, Fig 2B).  
+
+To compare the rate of decay of IgG antibody levels in mothers and their infants, we evaluated 5 dyads with paired maternal and infant blood samples on the same day at the time of follow- up (range 3- 9 weeks post- delivery). Maternal antibody IgG levels decreased faster in mothers than infants (mean delta - 974 RFU and - 670 RFU, respectively. Fig 2E) at the follow up timepoint. Taken together this indicates, maternally- derived functional vaccine induced antibodies persist at high levels in newborns through early infancy during a critical time of immune vulnerability and may decay slower than maternal IgG antibodies.  
+
+## Vaccine induced antibody timing and transplacental facilitated transfer
+
+<--- Page Split --->
+
+We assessed the relationship of anti- SARS- CoV- 2 IgG levels to neutralizing antibody levels. We found a strong correlation between IgG and neutralizing titers in maternal plasma at delivery \((R_{s} = 0.744, p = 0.0012)\) and infant follow up \((R_{s} = 0.738, p = 0.046)\) timepoints, but no significant association between IgG and neutralizing titers in cord blood \((R_{s} = 0.121, p = 0.656,\) Figure 3).  
+
+We then evaluated the impact of timing of vaccination on maternal antibody levels at delivery. We found no statistically significant correlation between maternal IgG levels at delivery and time since dose 1 \((R_{s} = - 0.335, p = 0.160)\) and gestational age at delivery \((R_{s} = 0.270, p = 0.265,\) Fig 4A,B). This lack of correlation appeared to be driven by two participants that had low or absent levels of antibodies at delivery and received their first dose of vaccine within 30 days of delivery. We then evaluated neutralizing titers in those participants with known detectable IgG levels at delivery and found that maternal neutralizing titers at delivery trended with days since vaccine dose 1 but was not statistically significantly \((R_{s} = - 0.422, p = 0.093)\) , and maternal neutralizing titers at delivery was not associated with gestational age at dose 1 \((R_{s} = 0.074,\) \(p = 0.780)\) . One participant was borderline IgG positive at delivery (vaccinated within 30 days of delivery) and did not have detectable neutralizing titers at delivery (Fig 4C,D).  
+
+To assess facilitated antibody transfer, we evaluated cord- to- maternal antibody IgG and neutralization titer ratios by time since vaccination and gestational age. We found that IgG ratios were highly correlated with both time since first maternal vaccination dose and gestational age at first dose \((R_{s} = 0.917, p< 0.0001\) and \(R_{s} = - 0.739, p = 0.002\) , respectively. Fig 4E,F). However, neutralization titer cord- to- maternal ratios by time since first vaccination dose and gestational age at first dose were not significantly associated \((R_{s} = 0.366, p = 0.179\) and \(R_{s} = - 0.032, p = 0.913\) , respectively, Figure 4G,H). Together, this may indicate that timing of vaccination in pregnancy is critical for maternal- fetal antibody transfer, and functional neutralizing antibodies are differentially transferred to the fetus as compared to total anti- SARS- CoV- 2 IgG during gestation.
+
+<--- Page Split --->
+
+## mRNA vaccination leads to a unique SARS-CoV2 Spike protein antibody epitope binding signature  
+
+We next investigated antibody linear epitope binding and transplacental transfer using the PhIP- seq/VirScan SARS- CoV- 2 Spike protein phage display array in mother- infant dyads at the time of birth (Figure 5). We found that timing of vaccination was important for the transplacental transfer of Spike protein epitope binding antibodies. Two mother- infant dyads had minimal Spike protein specific epitope binding. The first dyad only received one dose of mRNA vaccine 9 days prior to delivery, and the other dyad received the second vaccine dose 2 days prior to delivery.  
+
+We found high levels of SARS- CoV- 2 Spike protein epitope binding in 4 major peaks we designate as regions 1- 4 (Figure 5A). Region 1 overlays the carboxy terminal of the N- terminal domain. Region 2 overlaps with key residues for the S1/S2 cleavage site. Regions 3 and 4 are within S2, flanking the fusion loop and the transmembrane portion of the Spike protein, respectively. However, we found minimal binding in the receptor binding domain (RBD) of Spike protein. Prior evaluation using the PhIP- seq/VirScan SARS- CoV- 2 epitope phage array during SARS- CoV- 2 infection demonstrated similar binding in regions 3 and 4, however in SARS- CoV- 2 infection there was minimal binding in regions 1 and 2 demonstrating that antibody epitope binding in these regions may be unique to vaccination<sup>18</sup>. Additionally, there is proportional representation of linear epitope binding across the SARS- CoV- 2 Spike protein proteome between mothers and infants (Figure 5B). Taken together, SARS- CoV- 2 antibody linear epitope binding after vaccination during pregnancy shows similar patterns, with multiple immunodominant regions found in the majority of mothers and infants. Some of these regions are unique to vaccination and not observed during natural infection<sup>18- 20</sup>.  
+
+## Discussion
+
+<--- Page Split --->
+
+Among twenty women who received the COVID- 19 mRNA vaccine during pregnancy, our study found no evidence of transplacental transfer of mRNA vaccine products but did find high levels of functional vaccine- derived antibodies that transferred to the infant at delivery and persisted during early infancy. Additionally, we identified high levels of epitope binding in two regions of Spike protein unique to SARS- CoV- 2 vaccination<sup>18</sup>. These data may address some of the many unanswered questions regarding COVID- 19 vaccination in pregnancy: including the dynamics of antibody production in the pregnant immune state, and the optimal timing of immunization in pregnancy to impart passive immunity to the newborn during the vulnerable first few weeks of infancy.  
+
+Uptake of COVID- 19 vaccination in pregnancy has been slow<sup>5</sup>, and reasons for vaccine hesitancy are likely multifactorial — but theoretical concerns that vaccine mRNA could cross the placenta have been raised. We found no evidence of mRNA vaccine products in any of our delivery samples. Additionally, no infants in our study had a fetal immune response to Spike protein as demonstrated by a negative anti- SARS- CoV- 2 IgM antibody in cord blood and infant follow up samples. This further supports the lack of transfer of vaccine products, as only IgG is transferred from the mother, and IgM production would indicate an endogenous fetal immune response which has rarely been seen in natural infection with SARS- CoV- 2 during pregnancy<sup>16,21- 23</sup>. This provides additional reassurance that mRNA vaccination is safe during pregnancy.  
+
+We found that the timing of immunization during pregnancy is important to ensure transplacental transfer of protective antibodies to the neonate, and during critical windows of immune vulnerability during early infancy. Consistent with prior studies showing robust immune responses to mRNA vaccination<sup>14,15,17</sup>, we found high levels of IgG after two doses of mRNA vaccine. However, completion of the vaccination series well before delivery was important to ensure transfer of antibodies to the infant. Two mothers only received one vaccine dose prior to delivery and did not transfer antibodies as demonstrated by the lack of antibodies in cord (in one with available cord blood) and in both infants at follow- up. Additionally, neutralizing antibodies
+
+<--- Page Split --->
+
+were not transferred in a mother who received her second dose of vaccine 2 days prior to delivery. All evaluated mothers who received both doses during pregnancy and with the second dose greater than 9 days prior to delivery transferred IgG and neutralizing antibodies to their infants. Consistent with early studies of antibody transfer after COVID- 19 vaccination in pregnancy, most of our participants were vaccinated in the third trimester of pregnancy. Larger studies of individuals vaccinated prior to pregnancy and in the first and second trimester are needed to understand persistence and waning of vaccine- induced immune responses.  
+
+Additionally, we believe we are the first to report that infants in the first few months of life continued to have maternal vaccine- derived anti- SARS- CoV- 2 antibodies that were functional as demonstrated by high levels of neutralizing antibodies presenting infants up to 12 weeks of age. This is consistent with known persistence of maternally- derived antibodies from other vaccinations including pertussis, rubella, varicella<sup>24- 26</sup>. Additionally, we have previously found persistence of anti- SARS- CoV- 2 IgG antibodies in infants after natural infection up to 6 months<sup>16</sup>. However, the functional capability of these antibodies as compared to anti- SARS- CoV- 2 vaccination- derived antibodies is unknown. Further evaluation of the longitudinal persistence of maternal vaccine- derived antibodies during infancy will be critical to determine optimal timing of COVID- 19 vaccination in infancy.  
+
+Evaluation of paired maternal and baby samples at post- partum follow up timepoints showed a faster decline in maternal IgG antibody levels than infants, suggesting that persistence of maternally- derived antibody may be prolonged for infants. Differences in renal excretion and neonatal Fc receptor (FcRn) expression, which is involved in antibody degradation<sup>27</sup> in the infant as compared to adults, could underly these differences and should be explored further.  
+
+Consistent with observations in non- pregnant adults, we found that IgG levels in mothers at delivery, and at infant follow- up were highly correlated with neutralizing titers<sup>28</sup>. However, cord blood IgG levels did not correlate with neutralizing titers. Moreover, IgG cord- to- maternal
+
+<--- Page Split --->
+
+ratios, which represent a proxy of maternal to fetal antibody transfer, were highly correlated with timing of vaccination (gestational age and days since the first dose), but cord- to- maternal neutralizing titer ratios were not significantly associated with time since vaccination nor gestational age. During gestation there is facilitated transfer of maternally derived antibodies through the binding of the neonatal Fc receptor in the synctiotrophoblast layer<sup>29</sup>. Differences in glycosylation<sup>30,31</sup>, FcR/FcRn binding affinity<sup>17,32</sup>, preferential IgG subclass transfer<sup>33,34</sup> may be different in functional neutralizing antibodies as compared to total IgG antibody transfer. However, a limitation of this study is the majority of participants were vaccinated in the third trimester. Further investigations on factors that influence the transport of functional antibodies across trimesters are needed to understand antibody dynamics and optimal transfer of protective antibodies to infants.  
+
+Using a PhIP- seq/VirScan SARS- CoV- 2 Spike protein phage array we were able to compare linear epitope antibody binding in mothers and their infants. Consistent with IgG and neutralizing antibody evaluation, timing of vaccination was critical to ensure the transplacental transfer of antibodies to the infant. Additionally, we identified unique regions of antibody epitope binding in our vaccinated cohort that were not identified using the same phage library in a prior evaluation of a cohort of SARS- CoV- 2 infected individuals<sup>18</sup>. One of these regions included the carboxy terminal of the N- terminal domain, with other work having shown that the N- terminal domain is targeted by neutralizing antibodies against Spike protein<sup>35</sup>. We did not see significant binding in the receptor binding domain (RBD), which may be attributable to the fact that the phage display library displayed short, linear peptides while antibodies targeting RBD are known to target conformational epitopes. Lastly, we found that the same immunodominant regions targeted by antibodies targeting the Spike protein in both mothers and infants.  
+
+In summary, this work provides further evidence that mRNA vaccination is safe in pregnancy and demonstrates that it generates time- dependent protective, functional antibody responses in mothers and infants that persist during early infancy.
+
+<--- Page Split --->
+
+## Methods  
+
+Cohort and Sample collection: The University of California San Francisco (UCSF) institutional review board approved the study (20- 32077). Written informed consent was obtained from all participants. We enrolled 20 pregnant individuals who were vaccinated with either BNT- 162b2 or mRNA- 1273 mRNA vaccines. Pregnant individuals were followed through delivery, and their infants were followed up to 12 weeks of life. Maternal blood was collected during pregnancy (pre- vaccine, 3- 4 weeks post- dose 1, 4- 8 weeks post- dose 2). During delivery, maternal blood, placenta tissue, and cord blood was collected. Infant follow- up blood was collected at convenience timepoints. Whole blood was immediately added to RNAlater in a 1:1.3 ratio. Plasma was isolated from whole blood by centrifugation and immediately cryopreserved. Full- thickness placental biopsy was collected within 1 hour of delivery, washed three times with phosphate buffered saline, and preserved in RNAlater.  
+
+SARS- CoV- 2 plasma serology. Anti- SARS- CoV- 2 plasma IgM and IgG antibodies were measured using the Pylon 3D automated immunoassay system (ET Healthcare, Palo Alto, CA). In brief, quartz glass probes are pre- coated with either affinity purified goat anti Human IgM (IgM capture) or Protein G (IgG capture) are dipped into diluted patient sample. Samples are washed, and then the probe is dipped into the assay reagent containing both biotinylated recombinant spike protein receptor binding domain (RBD) and nucleocapsid protein (NP). After a washing, the probe is incubated with a Cy@5- streptavidin (Cy5- SA) polysaccharide conjugate reagent, allowing for cyclic amplification of the fluorescence signal. The background corrected signal is reported as relative fluorescent units (RFU) which is proportional to the amount of specific antibodies in the sample allowing for quantification. Levels of IgM and IgG were considered positive if greater than 50 relative fluorescence units.
+
+<--- Page Split --->
+
+## SARS-CoV-2 neutralizing assay  
+
+SARS- CoV- 2 antibody neutralization titers were measured using a label- free surrogate neutralization assay (LF- sVNT) previously described28. Briefly, the method measures the binding ability of recombinant RBD (Sino Biological, Wayne, PA) coated onto sensing probes (Gator Bio, Palo Alto, CA) to recombinant ACE2 (Sino Biological, Wayne, PA) after neutralizing RBD with SARS- CoV- 2 antibodies in serum. Measurements were done using a thin- film interferometry (TFI) label- free immunoassay analyzer (Gator Bio, Palo Alto, CA). Each serum sample was diluted in a series (1:25, 1:100, 1:250, 1:500, 1:1000, 1:2000) in running buffer (PBS at pH 7.4 with 0.02% Tween 20, 0.2% BSA, and 0.05% NaN3) for analysis. The first testing cycle for each diluted sample measured the binding ability of RBD to ACE2 with neutralization, and the second cycle provided the full binding ability of RBD without neutralization. In each cycle, the recorded time course of signals, as known as the sensorgram, was recorded. The readout measured the signal increase in RBD- ACE2 complex formation, representing the quantity of RBD- ACE2 complex on the sensing probe. A neutralization rate was calculated as the ratio of the readout in the first cycle to that in the second cycle, presented as a percentage. To obtain the neutralizing antibody titer (IC50) for each serum sample, the neutralization rates were plotted against dilutions, and the points were fitted using a linear interpolation model. The reciprocal of the dilution resulting in a 50% neutralization rate was defined as the neutralizing antibody titer.  
+
+SARS- CoV- 2 Spike protein Western blot. Maternal blood and cord blood were diluted in RNAlater in 1:1.3 ratio, placenta was preserved in RNAlater. Protein lysates were obtained from samples using RIPA buffer (150 mM NaCl, 25 mM Tris- HCl (pH 7.4), 1% NP- 40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate) containing Halt™ protease inhibitor cocktail (ThermoScientific). Cell Lysates were resolved by SDS/PAGE on a Bis- Tris methane 4–12% polyacrylamide gel and transferred to a nitrocellulose membrane, blocked with 5% skimmed
+
+<--- Page Split --->
+
+milk diluted in PBS, an incubated overnight at \(4^{\circ}C\) with anti- SARS- CoV- 2 Spike mouse mAb (1A9, GeneTex) or anti- GAPDH rabbit polyclonal antibody (GTX100118, GeneTex) respectively diluted 1:1,000 or 1:5,000 in blocking buffer. The membrane was washed in PBS buffer containing Tween- 20 (0.1%) and then incubated for 1 h with horseradish peroxidase- conjugated anti- mouse and anti- rabbit secondary antibody (Jackson ImmunoResearch) diluted respectively 1:5,000 and 1:10,000. The membrane was thoroughly washed, and proteins visualized using Immobilon Forte Western HRP substrate (Millipore).  
+
+SARS- CoV- 2 Spike mRNA PCR. Maternal blood and cord blood were diluted in RNAlater in 1:1.3 ratio, placenta was preserved in RNAlater. Tissues were kept at \(- 80^{\circ}C\) until analyzed. RNA was isolated from samples using the RNeasy Micro or Mini Kit (Qiagen) according to manufacturer's protocol. RNA concentration was measured using nanodrop and all samples had \(>30 \text{ng / ul}\) total RNA. \(500 \text{ng}\) RNA was transcribed into cDNA using qScript cDNA synthesis kit (Quantabio). Primers were design to detect the vaccines mRNA (mRNA- 1273 Moderna and BNT162b2 Pfizer- BioNtech) as previously described<sup>36</sup>. Forward primer:  
+
+AACGCCACCAACGTGGTCATC. Reverse primer: GTTGTTGGCGCTGCTGTACAC. Primers were shown to detect samples containing as low as 1.5 pg of vaccine using vaccine standard curve (Table S2). QuantaStudio 6 Flex (Applied Biosystems) instrument and SsoFast EvaGreen supermix (Bio- Rad) were used for PCR reaction: 30 second \(95^{\circ}C\) followed by 40 cycles of 5 second \(95^{\circ}C\) and 20 seconds \(60^{\circ}C\) . All samples were run in triplicate as 20 μL reactions, and Ct values corresponding to \(< 1.5 \text{pg}\) of vaccine based on standard curve (Table S2) were interpreted as a negative result. For vaccines cDNA standard curves, \(10000 \text{pg / \mu L}\) vaccine mRNA (as cDNA) sample was used for serial dilution in 1:3 ratio, up to \(0.06 \text{pg / \mu L}\) . Two \(\mu \text{L}\) of these diluted samples were used in each well to create standard curves.  
+
+PhIP- Seq/VirScan Coronavirus phage display assay
+
+<--- Page Split --->
+
+Immunoprecipitation of phage- bound patient antibodies  
+
+Maternal plasma at delivery and cord plasma were evaluated by PhIP- Seq/Virscan Coronavirus phage display. Construction of the Coronavirus PhIP- Seq library and detailed methods for immunoprecipitation, sequencing and bioinformatic processing of data are identical to what has previously been described<sup>18</sup>. For the purposes of the analysis conducted in this study, analysis was restricted to sero- reactivity against the SARS- CoV- 2 Spike protein. As previously described, a total of two rounds of amplification and selection were performed for all PhIP- Seq analyses.  
+
+Next Generation Sequencing library prep  
+
+Amplicon sequencing library preps were performed using the Labcyte Echo 525 and an Integra Via Flow 96 and were identical to what has previously been described<sup>18</sup>. All libraries were pooled by equal volume, cleaned and size selected using Ampure XP beads at 1.0X per manufacturer's protocol. Libraries were quantified by High Sensitivity DNA Qubit and quality- checked by High Sensitivity DNA Bioanalyzer. Sequencing was then performed on a NovaSeq S1 (300 cycle kit with 1.3 billion clusters) aiming for sequencing depths of at least 1 million reads per sample.  
+
+Bioinformatic Analysis of PhIP- Seq Data  
+
+Sequencing reads were aligned to a reference database of the full viral peptide library using the Bowtie2 aligner. For all VirScan libraries, the null distribution of each peptide's log10(rpK) was modeled using a set of 95 pre- pandemic, healthy control sera. All counts were augmented by 1 to avoid zero counts in the healthy control sera samples. Multiple distribution fits were examined for these data, with the Normal distribution showing the best fit. These null distributions were used to calculate p values for the observed log10(rpK) of each peptide within a given sample. The calculated p values were corrected for multiple hypothesis using the Benjamini- Hochberg
+
+<--- Page Split --->
+
+method. Any peptide with a corrected p value of \(< 0.001\) was considered significantly enriched over the healthy background. To identify regions targeted by host antibodies, all library peptides were aligned to the SARS- CoV- 2 reference genome. Using the aligned position of the significantly enriched peptides which aligned full- length against the reference, we determined the proportion of individuals (mothers and infants) that were reactive at each residue of the Spike protein. All plots were generated using the R ggplot2 package.  
+
+## Statistical analysis:  
+
+Statistical analyses were performed using PRISM v9.2 (GraphPad), STATA 16 (StataCorp), and R version 3.6.3 and R Studio version 1.1.447. Descriptive statistics include mean, standard deviations, and ranges for continuous variables. The Wilcoxon rank- sum test was used for two- group comparisons of continuous variables including maternal pre- and post- vaccine antibody responses. Associations between continuous variables were assessed using Spearman's rank correlation \((R_{s})\) including comparisons between maternal, cord and infant follow- up antibody IgG and neutralizing titer responses, and timing of vaccination. Two- sided \(p\) values were calculated for all test statistics, and \(p< 0.05\) was considered significant. PhIP- Seq/VirScan bioinformatics as detailed above.  
+
+## Data Availability  
+
+The data set generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.  
+
+## Acknowledgements:  
+
+M.P. was supported by the National Institutes of Health (NIAID K23AI127886), the Marino Family Foundation, and UCSF REAC award. Y.G. was supported by the Weizmann Institute of Science - National Postdoctoral Award Program for Advancing Women in Science, and of the
+
+<--- Page Split --->
+
+International Society for Research in Human Milk and Lactation (ISRHDL) Trainee Bridge Fund. Y.M. and W.C.G. were supported by The Roddenberry Foundation. S.L.G. was supported by the National Institutes of Health (NIAID K08AI141728), and the Bill and Melinda Gates Foundation (INV- 017035), and a generous gift from the Kryzewski Family.  
+
+We thank all the mothers and infants that participated in this study. We thank Kenneth Scott, BS, RPh, (UCSF Health Pharmacy, University of California, San Francisco) and Hannah J. Jang, PhD, RN, PHN, CNL (UCSF School of Nursing, University of California, San Francisco), for voluntarily providing unused vaccine for this study, and to Dr. Margaret Feeney (University of California, San Francisco) and Dr. Nadav Ahituv (University of California, San Francisco) for support of these experiments.  
+
+## Author contributions:  
+
+M.P. Helped conceive and design the project, oversaw recruitment, designed, and performed sample collection, oversaw experiment design, oversaw data analysis, provided funding, and drafted the manuscript.  
+
+Y.G. Recruited and consented enrollees, oversaw sample collection, designed, performed, and analyzed mRNA PCR experiments, performed data analysis.  
+
+A.G.C. Recruited and consented enrollees, oversaw sample collection, performed chart review, and helped draft the manuscript.  
+
+Y.M. Performed and helped design Western blot.  
+
+L.L. Performed and analyzed mRNA PCR experiments, performed sample collection.  
+
+B.A. Performed phage immunoprecipitation assays.  
+
+H.C. and U.J. performed and helped design critical experiments, and data collection.  
+
+C.Y.L., V.J.L., M.C., L.W., S.B. Performed and coordinated sample collection, and data collection.
+
+<--- Page Split --->
+
+V.J.F. Helped conceive and coordinate the project.  
+
+A.P.M. Provide funding.  
+
+W.C.G. Helped design western blot and oversaw data analysis.  
+
+A.H.B.W Designed and oversaw all serology experiments.  
+
+K.L.L. Designed and oversaw all neutralizing antibody experiments.  
+
+J.R. Designed, analyzed, and oversaw phage immunoprecipitation sequencing assays.  
+
+S.L.G. conceived and designed the project, oversaw recruitment, oversaw experiment design,  
+
+oversaw data analysis, provided funding, and helped draft the manuscript.  
+
+M.P., Y.G., Y.U., L.L., A.H.B.W, W.C.G, K.L.L., and S.L.G verified the underlying data.  
+
+All authors reviewed and approved the manuscript.
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1. Anti-SARS-CoV-2 IgG and IgM antibody responses following vaccination </center>  
+
+A. Maternal plasma anti-SARS-CoV-2 IgG antibody relative fluorescence units (RFU) levels prior to vaccination \((n = 4)\) , 3-4 weeks post-dose 1 \((n = 7)\) , and 4-8 weeks post-dose 2 \((n = 12)\) . B. Maternal plasma anti-SARS-CoV-2 IgM (RFU) levels prior to vaccination \((n = 4)\) , 3-4 weeks post-dose 1 \((n = 7)\) , and 4-8 weeks post-dose 2 \((n = 12)\) . Wilcoxon rank-sum testing. Data represent median ± quartiles, two-sided \(p\) values were calculated for all test statistics.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2. Paired maternal, cord, and infant IgG and neutralization antibodies </center>  
+
+481 A. Paired maternal plasma at delivery \((n = 19)\) , cord plasma \((n = 17)\) , and infant follow-up \((n = 10)\) 482 by anti- SARS- CoV- 2 IgG antibody relative fluorescence units (RFU), (Spearman's rank 483 correlation, dotted line indicates positive cutoff value of 50 RFU). B. Paired maternal plasma at
+
+<--- Page Split --->
+
+delivery (n=17), cord plasma (n=16), and infant follow-up (n=8) by SARS- CoV- 2 label- free surrogate neutralization assay (sVNT), (Spearman's rank correlation, dotted line indicates positive cutoff value of 25). C. Paired cord plasma (n=9) and infant follow-up plasma (n=11) anti- SARS- CoV- 2 IgG by weeks of life. D. Paired cord plasma (n=7) and infant follow-up plasma (n=8) label- free surrogate neutralization assay (sVNT) by weeks of life. E. Paired maternal plasma at delivery (n=5), cord plasma (n=5), and paired maternal follow-up (n=5) and infant follow- up plasma (n=5) anti- SARS- CoV- 2 IgG. Two- sided \(p\) values were calculated for all test statistics.
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3. Neutralization to IgG antibody correlation </center>  
+
+A. Maternal plasma at delivery \((n = 17)\) B. Cord plasma \((n = 16)\) C. Infant follow-up plasma \((n = 8)\) SARS-CoV-2 label-free surrogate neutralization assay (sVNT) by anti-SARS-CoV-2 IgG correlation (Spearman's rank correlation). Two-sided \(p\) values were calculated for all test statistics.
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4. Maternal delivery and Cord-to-maternal antibody transfer ratios timing </center>  
+
+A. Maternal delivery anti-SARS-CoV-2 IgG antibody transfer ratio by days since vaccine dose 1 (n=19, dashed line indicates positive cutoff >50 RFU) 
+B. Maternal delivery anti-SARS-CoV-2 IgG antibody transfer ratio by gestational age at vaccine dose 1 (n=19, dashed line indicates positive cutoff >50 RFU) 
+C. Maternal delivery SARS-CoV-2 label-free surrogate neutralization assay (sVNT) antibody transfer ratio by days since vaccine dose 1 (n=17, dashed line indicates positive cutoff >25). 
+D. Maternal delivery SARS-CoV-2 label-free surrogate neutralization assay (sVNT) antibody transfer ratio by gestational age at vaccine dose 1 (n=17, dashed line indicates positive cutoff >25) 
+E. Cord-to-maternal anti-SARS-CoV-2 IgG antibody transfer ratio by days since vaccine dose 1 (n=15) 
+F. Cord-to-maternal anti-SARS-CoV-2 IgG antibody transfer ratio by gestational age at vaccine dose 1 (n=15) 
+G. Cord-to-maternal SARS-CoV-2 label-free surrogate neutralization assay (sVNT) antibody transfer ratio by days since vaccine dose 1 (n=15). 
+H. Cord-to-maternal SARS-CoV-2 label-free surrogate neutralization assay (sVNT)
+
+<--- Page Split --->
+
+536 antibody transfer ratio by gestational age at vaccine dose 1 (n=15). Two-sided \(p\) values were calculated for all test statistics.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Figure 5. PhIP-seq/VirScan paired maternal and cord SARS-CoV-2 Spike protein epitope </center>  
+
+## binding  
+
+A. Heatmap displaying results of significant enriched (p<0.001) linear SARS-CoV-2 Spike protein epitope binding from 15 paired mother-infant dyads in maternal plasma at delivery and cord plasma by vaccine type and time since vaccine dose 1. Areas of high cumulative epitope
+
+<--- Page Split --->
+
+567 binding designated by regions 1- 4. B. Cumulative fold enrichment of mothers and infants linear 
+
+568 SARS-CoV-2 Spike protein epitope binding.
+
+<--- Page Split --->
+
+Ellington, S. et al. Characteristics of Women of Reproductive Age with Laboratory- Confirmed SARS- CoV- 2 Infection by Pregnancy Status - United States, January 22- June 7, 2020. MMWR Morb Mortal Wkly Rep 69, 769- 775, doi:10.15585/mmwr.mm6925a1 (2020).  
+
+Ahlberg, M. et al. Association of SARS- CoV- 2 Test Status and Pregnancy Outcomes. Jama, doi:10.1001/jama.2020.19124 (2020).  
+
+Knight, M. et al. Characteristics and outcomes of pregnant women admitted to hospital with confirmed SARS- CoV- 2 infection in UK: national population based cohort study. Bmj 369, m2107, doi:10.1136/bmj.m2107 (2020).  
+
+Martinez- Portilla, R. J. et al. Pregnant women with SARS- CoV- 2 infection are at higher risk of death and pneumonia: propensity score matched analysis of a nationwide prospective cohort (COV19Mx). Ultrasound Obstet Gynecol 57, 224- 231, doi:10.1002/uog.23575 (2021).  
+
+CDC. COVID- 19 vaccination among pregnant people aged 18- 49 years overall, by race/ethnicity, and date reported to CDC - Vaccine Safety Datalink, \* United States,  
+
+<https://covid.cdc.gov/covid- data- tracker/#vaccinations- pregnant- women> (2021).  
+
+Gynecologists, A. C. o. O. a. Statement of Strong Medical Consensus for Vaccination of Pregnant Individuals Against COVID- 19, <https://www.acog.org/news/news-  
+
+releases/2021/08/statement- of- strong- medical- consensus- for- vaccination- of- pregnant-  
+
+individuals- against- covid- 19> (2021).
+
+<--- Page Split --->
+
+Maykin, M. M., Heuser, C., Feltovich, H. & with the Society for Maternal- Fetal Medicine Health Policy Advocacy, C. Pregnant people deserve the protection offered by SARS- CoV- 2 vaccines. Vaccine 39, 171- 172, doi:10.1016/j.vaccine.2020.12.007 (2021).  
+
+Jackson, L. A. et al. An mRNA Vaccine against SARS- CoV- 2 - Preliminary Report. N Engl J Med 383, 1920- 1931, doi:10.1056/NEJMoa2022483 (2020).  
+
+Polack, F. P. et al. Safety and Efficacy of the BNT162b2 mRNA Covid- 19 Vaccine. N Engl J Med 383, 2603- 2615, doi:10.1056/NEJMoa2034577 (2020).  
+
+Atveo, C. et al. Compromised SARS- CoV- 2- specific placental antibody transfer. Cell, doi:10.1016/j.cell.2020.12.027 (2020).  
+
+Edlow, A. G. et al. Assessment of Maternal and Neonatal SARS- CoV- 2 Viral Load, Transplacental Antibody Transfer, and Placental Pathology in Pregnancies During the COVID- 19 Pandemic. JAMA Netw Open 3, e2030455, doi:10.1001/jamanetworkopen.2020.30455 (2020).  
+
+Flannery, D. D. et al. Assessment of Maternal and Neonatal Cord Blood SARS- CoV- 2 Antibodies and Placental Transfer Ratios. JAMA Pediatr, doi:10.1001/jamapediatrics.2021.0038 (2021).  
+
+Atveo, C. et al. Compromised SARS- CoV- 2- specific placental antibody transfer. Cell 184, 628- 642. e610, doi:10.1016/j.cell.2020.12.027 (2021).  
+
+Gray, K. J. et al. Coronavirus disease 2019 vaccine response in pregnant and lactating women: a cohort study. Am J Obstet Gynecol 225, 303. e301- 303. e317, doi:10.1016/j.ajog.2021.03.023 (2021).  
+
+Nir, O. et al. Maternal- neonatal transfer of SARS- CoV- 2 immunoglobulin G antibodies among parturient women treated with BNT162b2 messenger RNA vaccine during
+
+<--- Page Split --->
+
+pregnancy. Am J Obstet Gynecol MFM 4, 100492, doi:10.1016/j.ajogmf.2021.100492 (2021).  
+
+Song, D. et al. Passive and active immunity in infants born to mothers with SARS-CoV- 2 infection during pregnancy: prospective cohort study. BMJ Open 11, e053036, doi:10.1136/bmjopen- 2021- 053036 (2021).  
+
+Atyeo, C. et al. COVID- 19 mRNA vaccines drive differential antibody Fc- functional profiles in pregnant, lactating, and non- pregnant women. Sci Transl Med, eabi8631, doi:10.1126/scitranslmed.abi8631 (2021).  
+
+Zamecnik, C. R. et al. ReScan, a Multiplex Diagnostic Pipeline, Pans Human Sera for SARS- CoV- 2 Antigens. Cell Rep Med 1, 100123, doi:10.1016/j.xcrm.2020.100123 (2020). Shrock, E. et al. Viral epitope profiling of COVID- 19 patients reveals cross- reactivity and correlates of severity. Science 370, doi:10.1126/science.abd4250 (2020).  
+
+Li, Y. et al. Linear epitopes of SARS- CoV- 2 spike protein elicit neutralizing antibodies in COVID- 19 patients. Cell Mol Immunol 17, 1095- 1097, doi:10.1038/s41423- 020- 00523- 5 (2020).  
+
+Fenizia, C. et al. Analysis of SARS- CoV- 2 vertical transmission during pregnancy. Nat Commun 11, 5128, doi:10.1038/s41467- 020- 18933- 4 (2020).  
+
+Zeng, H. et al. Antibodies in Infants Born to Mothers With COVID- 19 Pneumonia. JAMA 323, 1848- 1849, doi:10.1001/jama.2020.4861 (2020).  
+
+Zeng, L. et al. Neonatal Early- Onset Infection With SARS- CoV- 2 in 33 Neonates Born to Mothers With COVID- 19 in Wuhan, China. JAMA Pediatr 174, 722- 725,  
+
+doi:10.1001/jamapediatrics.2020.0878 (2020).
+
+<--- Page Split --->
+
+Healy, C. M. et al. Kinetics of maternal pertussis-specific antibodies in infants of mothers vaccinated with tetanus, diphtheria and acellular pertussis (Tdap) during pregnancy. Vaccine 38, 5955- 5961, doi:10.1016/j.vaccine.2020.06.050 (2020).  
+
+Sissoko, M. S. et al. Safety and efficacy of PfSPZ Vaccine against Plasmodium falciparum via direct venous inoculation in healthy malaria- exposed adults in Mali: a randomised, double- blind phase 1 trial. The Lancet Infectious Diseases 17, 498- 509, doi:10.1016/s1473- 3099(17)30104- 4 (2017).  
+
+Leuridan, E., Hens, N., Hutse, V., Aerts, M. & Van Damme, P. Kinetics of maternal antibodies against rubella and varicella in infants. Vaccine 29, 2222- 2226, doi:10.1016/j.vaccine.2010.06.004 (2011).  
+
+Pyzik, M. et al. The Neonatal Fc Receptor (FcRn): A Misnomer? Front Immunol 10, 1540, doi:10.3389/fimmu.2019.01540 (2019).  
+
+Luo, Y. R., Yun, C., Chakraborty, I., Wu, A. H. B. & Lynch, K. L. A SARS- CoV- 2 Label- Free Surrogate Virus Neutralization Test and a Longitudinal Study of Antibody Characteristics in COVID- 19 Patients. J Clin Microbiol 59, e0019321, doi:10.1128/jcm.00193- 21 (2021). Firan, M. et al. The MHC class I- related receptor, FcRn, plays an essential role in the maternofetal transfer of gamma- globulin in humans. Int Immunol 13, 993- 1002, doi:10.1093/intimm/13.8.993 (2001).  
+
+Williams, P. J. et al. Short communication: selective placental transport of maternal IgG to the fetus. Placenta 16, 749- 756, doi:10.1016/0143- 4004(95)90018- 7 (1995). Kibe, T. et al. Glycosylation and Placental Transport of Immunoglobulin G. Journal of Clinical Biochemistry and Nutrition 21, 57- 63 (1996).
+
+<--- Page Split --->
+
+681 32 Borghi, S. et al. FcRn, but not FcγRs, drives maternal-fetal transplacental transport of 682 human IgG antibodies. Proc Natl Acad Sci U S A 117, 12943- 12951, 683 doi:10.1073/pnas.2004325117 (2020). 684 33 Palmeira, P., Quinello, C., Silveira-Lessa, A. L., Zago, C. A. & Carneiro-Sampaio, M. IgG 685 placental transfer in healthy and pathological pregnancies. Clin Dev Immunol 2012, 686 985646, doi:10.1155/2012/985646 (2012). 687 34 Clements, T. et al. Update on Transplacental Transfer of IgG Subclasses: Impact of 688 Maternal and Fetal Factors. Front Immunol 11, 1920, doi:10.3389/fimmu.2020.01920 689 (2020). 690 35 Chi, X. et al. A neutralizing human antibody binds to the N-terminal domain of the Spike 691 protein of SARS-CoV-2. Science 369, 650- 655, doi:10.1126/science.abc6952 (2020). 692 36 Golan, Y. et al. Evaluation of Messenger RNA From COVID- 19 BTN162b2 and mRNA- 1273 693 Vaccines in Human Milk. JAMA Pediatr 175, 1069- 1071, 694 doi:10.1001/jamapediatrics.2021.1929 (2021).
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- SupplementalAppendixEvaluationoftransplacentalofmRNAvaccineproductsandfunctionalantibodiesduringpregnancyandearlyinfancy.pdf
+
+<--- Page Split --->

preprint/preprint__7e8d9494866549fe6fb82c521aaa52f5960e0b5ad6ca19e409ea66d4b8156b98/preprint__7e8d9494866549fe6fb82c521aaa52f5960e0b5ad6ca19e409ea66d4b8156b98_det.mmd

@@ -0,0 +1,556 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 921, 211]]<|/det|>
+# Evaluation of transplacental transfer of mRNA vaccine products and functional antibodies during pregnancy and early infancy  
+
+<|ref|>text<|/ref|><|det|>[[44, 229, 388, 270]]<|/det|>
+Mary Prahl ( \(\boxed{\boxed{\pi}}\) mary.prahl@ucsf.edu) University of California, San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 277, 387, 316]]<|/det|>
+Yarden Golan University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 324, 387, 363]]<|/det|>
+Arianna Cassidy University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 370, 223, 409]]<|/det|>
+Yusuke Matsui Gladstone Institute  
+
+<|ref|>text<|/ref|><|det|>[[44, 416, 387, 456]]<|/det|>
+Lin Li University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 463, 748, 503]]<|/det|>
+Bonny Alvarenga University of California, San Francisco https://orcid.org/0000- 0001- 7922- 9454  
+
+<|ref|>text<|/ref|><|det|>[[44, 509, 387, 549]]<|/det|>
+Hao Chen University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 556, 387, 595]]<|/det|>
+Unurzul Jigmeddagva University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 602, 387, 641]]<|/det|>
+Christine Lin University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 648, 387, 687]]<|/det|>
+Veronica Gonzalez University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 694, 387, 733]]<|/det|>
+Megan Chidboy University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 740, 387, 779]]<|/det|>
+Lakshmi Warrier University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 787, 387, 825]]<|/det|>
+Sirirak Buarpung University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 833, 153, 870]]<|/det|>
+Amy Murtha UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 878, 387, 917]]<|/det|>
+Valerie Flaherman University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 925, 175, 942]]<|/det|>
+Warner Greene
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 46, 308, 64]]<|/det|>
+J. David Gladstone Institutes  
+
+<|ref|>text<|/ref|><|det|>[[44, 71, 256, 110]]<|/det|>
+Alan Wu University of California  
+
+<|ref|>text<|/ref|><|det|>[[44, 117, 144, 155]]<|/det|>
+Kara Lynch UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 163, 388, 202]]<|/det|>
+Jayant Rajan University of California, San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 209, 744, 250]]<|/det|>
+Stephanie Gaw University of California, San Francisco https://orcid.org/0000- 0003- 0891- 6964  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 291, 102, 309]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 328, 911, 372]]<|/det|>
+Keywords: SARS- CoV- 2, COVID- 19, Pregnancy, Vaccine, Antibody, Neonatal Immunity, Neutralizing 47 Antibody, Phage Array, mRNA Vaccination, BNT- 162b2, mRNA- 1273, Placenta, Cord Blood  
+
+<|ref|>text<|/ref|><|det|>[[44, 389, 344, 408]]<|/det|>
+Posted Date: December 15th, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 427, 474, 447]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 1150427/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 465, 909, 508]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 544, 907, 587]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on July 30th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 32188- 1.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[72, 61, 884, 120]]<|/det|>
+Evaluation of transplacental transfer of mRNA vaccine products and functional antibodies during pregnancy and early infancy  
+
+<|ref|>text<|/ref|><|det|>[[70, 155, 884, 339]]<|/det|>
+Mary Prahl, M.D., Yarden Golan, Ph.D., Arianna G. Cassidy, M.D., Yusuke Matsui M.D., Ph.D., Lin Li, M.D., PhD, Bonny Alvarenga, Hao Chen, Ph.D., Unurzul Jigmeddagva, M.D., Christine Y. Lin, Veronica J. Gonzalez, M.D, Megan A. Chidboy, Lakshmi Warrier, Sirirak Buarpung, D.V.M., Ph.D., Amy P. Murtha M.D., Valerie J. Flaherman, M.D., M.P.H., Warner C. Greene M.D., Ph.D., Alan H.B. Wu, Ph.D., Kara L. Lynch Ph.D., Jayant Rajan, M.D., Ph.D., Stephanie L. Gaw, M.D., Ph.D  
+
+<|ref|>text<|/ref|><|det|>[[68, 352, 88, 368]]<|/det|>
+10  
+
+<|ref|>text<|/ref|><|det|>[[68, 382, 880, 863]]<|/det|>
+Author AffiliationsFrom the Department of Pediatrics, University of California, San Francisco (M.P., V.J.F.), Division of Pediatric Infectious Diseases and Global Health, University of California, San Francisco (M.P.), Department of Bioengineering and Therapeutic Sciences, University of California, San Francisco (Y.G.), Department of Medicine, University of California, San Francisco (L.W., B.A., J.R.), Weill Institute for Neurosciences, Division of Neurology, University of California, San Francisco, CA (B.A.), Gladstone Center for HIV Cure Research, Gladstone Institute, San Francisco, CA (Y.M., W.C.G.) Departments of Medicine and Microbiology and Immunology, University of California, San Francisco (W.C.G.), Department of Laboratory Medicine, University of California, San Francisco (A.H.B.W., K.L.L.), Center for Reproductive Sciences, Department of Obstetrics, Gynecology, and Reproductive Sciences, University of California San Francisco (A.G.C., L.L., H.C., U.J., C.Y.L., V.J.G., M.A.C., S.B., A.P.M., S.L.G.). Co-Corresponding authors: 1) Mary Prahl- Division of Pediatric Infectious Diseases and Global Health, Department of Pediatrics, University of California San Francisco, 550 16<sup>th</sup> St. 4<sup>th</sup> Floor. San Francisco, CA 94158. Phone (415) 514-0510, Email: mary.prahl@ucsf.edu 2) Stephanie L.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 61, 880, 140]]<|/det|>
+Gaw- Division of Maternal- Fetal Medicine, Department of Obstetrics, Gynecology, and Reproductive Sciences, University of California San Francisco, 513 Parnassus Ave, HSE16, Box 0556, San Francisco, CA 94143. Phone: (415) 476- 0535. Email: Stephanie.Gaw@ucsf.edu  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 194, 186, 210]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[110, 222, 880, 595]]<|/det|>
+Studies are needed to evaluate the safety and effectiveness of mRNA SARS- CoV- 2 vaccination during pregnancy, and the levels of protection provided to their newborns through placental transfer of antibodies. We evaluated the transplacental transfer of mRNA vaccine products and functional anti- SARS- CoV- 2 antibodies during pregnancy and early infancy in a cohort of 20 individuals vaccinated during pregnancy. We found no evidence of mRNA vaccine products in maternal blood, placenta tissue, or cord blood at delivery. However, we found time- dependent efficient transfer of IgG and neutralizing antibodies to the neonate that persisted during early infancy. Additionally, using phage immunoprecipitation sequencing, we found a vaccine- specific signature of SARS- CoV- 2 Spike protein epitope binding that is transplacentally transferred during pregnancy. In conclusion, products of mRNA vaccines are not transferred to the fetus during pregnancy, however timing of vaccination during pregnancy is critical to ensure transplacental transfer of protective antibodies during early infancy.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 640, 198, 656]]<|/det|>
+## Keywords  
+
+<|ref|>text<|/ref|><|det|>[[115, 668, 853, 720]]<|/det|>
+SARS- CoV- 2, COVID- 19, Pregnancy, Vaccine, Antibody, Neonatal Immunity, Neutralizing Antibody, Phage Array, mRNA Vaccination, BNT- 162b2, mRNA- 1273, Placenta, Cord Blood  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 766, 213, 783]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[115, 796, 880, 880]]<|/det|>
+Growing evidence has shown that pregnant individuals are at higher risk for SARS- CoV- 2- related morbidity and mortality<sup>1-4</sup>. Despite this, vaccination uptake by pregnant individuals has been slower than the general population<sup>5</sup>, in part because of maternal concern of adverse
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 62, 879, 500]]<|/det|>
+effects on the embryo or fetus, even with strong consensus recommendations for COVID- 19 vaccination prior to or during pregnancy from several medical societies. Pregnant individuals were excluded from initial vaccine trials, and complete data on safety, efficacy, optimal timing of the vaccine in pregnancy, or its impact on the fetus has been delayed, which may impact individual medical decision making. Current COVID- 19 vaccines fully approved and under emergency use in the United States include the mRNA vaccines BNT- 162b2 and mRNA- 1273, which target the SARS- CoV- 2 Spike protein and stimulate protective immune responses. In addition to protecting the mother against severe disease, vaccination during pregnancy may protect the newborn through passive transfer of maternal immunoglobulin. SARS- CoV- 2 infection and vaccination during pregnancy produces an IgG response that is transferred to the fetus. Evidence of newborn protection might help address maternal concerns about adverse effects. However, detailed studies of the transplacental transfer of vaccine products and vaccine- related antibody dynamics, functional properties, and persistence during infancy of transferred SARS- CoV- 2 antibodies are needed to provide such evidence.  
+
+<|ref|>text<|/ref|><|det|>[[115, 510, 840, 592]]<|/det|>
+We examined the transplacental transfer of mRNA vaccine products and humoral responses using samples from pregnant individuals and their infants vaccinated with either BNT- 162b2 or mRNA- 1273 mRNA vaccine during pregnancy.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 640, 189, 656]]<|/det|>
+## Results:  
+
+<|ref|>text<|/ref|><|det|>[[112, 668, 880, 882]]<|/det|>
+Cohort: We evaluated 20 pregnant individuals who received COVID- 19 mRNA vaccines during pregnancy and their infants. Participants were vaccinated between December 2020 and April 2021. Gestational age at first dose ranged from 13 weeks to 40 weeks (mean 31.2, SD 5.9 weeks). Nineteen participants delivered live, singleton infants between January 2021 through April 2021 at gestational ages ranging from 37.4 to 41.1 weeks (mean 39.2, SD 1.1 weeks). One participant who was vaccinated at 13 weeks had a termination of pregnancy due to a lethal skeletal dysplasia of genetic etiology at 20.4 weeks. Eight participants received BNT- 162b2
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 62, 876, 308]]<|/det|>
+(Pfizer- BioNTech) and twelve received mRNA- 1273 (Moderna) vaccines. Eighteen participants received both vaccine doses prior to delivery, and two participants received the second dose after delivery. The time from first mRNA vaccine dose ranged from 6- 97 (mean 51, SD 24.3) days prior to delivery, time from the second dose ranged from 2- 75 (mean 32, SD 21.3) days prior to delivery, and in two participants 15 and 21 days after delivery. No participants received a \(3^{\text{rd}}\) dose prior to delivery. Infants born to vaccinated mothers were followed up at convenience time points ranging from age 3 weeks to 15 weeks of life (mean 8.3, SD 3.2). Further demographic data is detailed in Table S1.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 350, 551, 370]]<|/det|>
+## Vaccine mRNA products do not cross the placenta  
+
+<|ref|>text<|/ref|><|det|>[[112, 382, 880, 594]]<|/det|>
+To determine the transplacental transfer of mRNA vaccine derived products, we examined available maternal blood at delivery, placenta tissue, and cord blood for Spike protein by Western blot and vaccine mRNA by PCR. All available delivery samples (maternal blood, placental tissue, and cord blood) were negative for Spike protein by Western blot (Supp Figure 1, Supp Table 3) and did not have detectable levels of vaccine mRNA by PCR (Suppl Table 3). Together, this indicates that products of mRNA vaccination do not reach the fetus after vaccination during pregnancy at readily detectable levels.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 638, 707, 658]]<|/det|>
+## mRNA vaccination in pregnancy leads to a robust antibody response  
+
+<|ref|>text<|/ref|><|det|>[[112, 668, 882, 819]]<|/det|>
+Similar to prior studies \(^{14,15,17}\) , we found that mRNA vaccination during pregnancy led to an increase in anti- SARS- Cov- 2 IgG following dose 1 (n=7, mean 388.6, SD 224.8 RFU) and an even further robust increase after vaccination dose 2 (n=12, mean 3214, SD 1383 RFU). Anti- SARS- CoV- 2 IgM (n=7, mean 53.3, SD 50.2 RFU) was detected in two maternal participants following dose 1, but only 1 participant following dose 2 (n=12, mean 23.8, SD 17 RFU, Fig 1).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[113, 64, 860, 84]]<|/det|>
+## Vaccine induced anti-SARS-CoV-2 IgG and neutralizing antibodies are transplacentially  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 96, 214, 112]]<|/det|>
+## transferred  
+
+<|ref|>text<|/ref|><|det|>[[111, 125, 881, 500]]<|/det|>
+We then evaluated the transplacental transfer of maternally derived anti- SARS- CoV- 2 IgG antibodies to their infants. Maternal blood at delivery was available in 19/20 participants and cord blood was available in 17/20 participants. Anti- SARS- CoV- 2 IgG was detectable in \(94.7\%\) (18/19) of maternal blood samples at delivery (mean 3235, range [10, 7811] RFU). Anti- SARS- CoV- 2 IgG was detectable in \(88.2\%\) (15/17) cord blood samples (mean 2243, range [2, 4959] RFU). One participant received one mRNA vaccine dose 9 days prior to delivery, and both the maternal and cord blood were negative for IgG at the time of delivery. Another participant received two doses of mRNA vaccine (23 and 2 days) prior to delivery and maternal blood was positive at 55 RFU (positive cutoff \(>50\) RFU), however cord blood IgG was negative (Figure 2A). Maternal and cord blood anti- SARS- CoV- 2 IgG levels were moderately correlated, but not statistically significant ( \(p = 0.074\) , \(R_{s} = 0.446\) , Fig 2A). All cord blood samples were anti- SARS- CoV- 2 IgM negative.  
+
+<|ref|>text<|/ref|><|det|>[[111, 510, 881, 818]]<|/det|>
+We next evaluated the transplacental transfer of neutralizing antibody titers by a label- free surrogate neutralization assay (sVNT) from mother to cord blood. Maternal and cord blood at delivery had robust neutralizing responses (maternal \(n = 17\) , mean 220.2, range [0, 422]. Cord blood \(n = 16\) , mean 296.6, range [0, 485], Fig 2B). All mother- infant dyads with positive IgG serology at delivery had detectable transplacental transfer of neutralizing antibodies with the exception of one pair in which the mother was borderline IgG positive at delivery and cord blood was negative, for which both maternal and cord blood were negative for neutralizing titers (Fig 2B). However, maternal and cord blood neutralizing titers were not significantly correlated ( \(p = 0.361\) , \(R_{s} = - 0.244\) , Fig 2B). Taken together, this indicates that maternal mRNA vaccination induces functional neutralizing antibodies that are transferred to the infant.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 63, 847, 84]]<|/det|>
+## Maternally-derived vaccine induced anti-SARS-CoV-2 IgG and neutralizing antibodies  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 96, 366, 115]]<|/det|>
+## persist through early infancy  
+
+<|ref|>text<|/ref|><|det|>[[112, 123, 884, 563]]<|/det|>
+A subset of infants was sampled at convenience timepoints during follow up (infant \(n = 11\) , weeks of life range [3,15] mean 8.3 weeks). Anti- SARS- CoV- 2 IgG levels were positive in \(81.8\%\) of infants at follow- up (9/11 infants, mean 1290, range [1, 3225] RFU, Fig 2A), with one infant still IgG positive at 12 weeks of age (Fig 2C). The two infants that were IgG negative at follow up were both born to mothers who received only one vaccine dose prior to delivery (6 and 9 days, respectively). One of these infants did not have paired maternal or cord blood available at the time of delivery for comparison, and the other was IgG negative in cord blood. Maternal and infant follow- up anti- SARS- CoV- 2 IgG levels were not significantly correlated; however, cord blood and infant follow- up IgG levels were significantly associated \((p = 0.492, R_{s} = 0.249\) and \(p = 0.021, R_{s} = 0.76\) , respectively, Fig 2A). All infants were IgM negative at the time of follow up. All infants with available IgG positive samples at follow up had detectable neutralizing titers \((n = 8\) , mean 154, range [41- 256], Fig 2B). Maternal and infant follow- up neutralizing titers were not significantly correlated, as well as cord and infant follow up neutralizing titers \((p = 0.665\) , \(R_{s} = - 0.191\) and \(p = 0.662, R_{s} = 0.214\) , respectively, Fig 2B).  
+
+<|ref|>text<|/ref|><|det|>[[112, 574, 886, 787]]<|/det|>
+To compare the rate of decay of IgG antibody levels in mothers and their infants, we evaluated 5 dyads with paired maternal and infant blood samples on the same day at the time of follow- up (range 3- 9 weeks post- delivery). Maternal antibody IgG levels decreased faster in mothers than infants (mean delta - 974 RFU and - 670 RFU, respectively. Fig 2E) at the follow up timepoint. Taken together this indicates, maternally- derived functional vaccine induced antibodies persist at high levels in newborns through early infancy during a critical time of immune vulnerability and may decay slower than maternal IgG antibodies.  
+
+<|ref|>sub_title<|/ref|><|det|>[[112, 830, 723, 850]]<|/det|>
+## Vaccine induced antibody timing and transplacental facilitated transfer
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 62, 876, 210]]<|/det|>
+We assessed the relationship of anti- SARS- CoV- 2 IgG levels to neutralizing antibody levels. We found a strong correlation between IgG and neutralizing titers in maternal plasma at delivery \((R_{s} = 0.744, p = 0.0012)\) and infant follow up \((R_{s} = 0.738, p = 0.046)\) timepoints, but no significant association between IgG and neutralizing titers in cord blood \((R_{s} = 0.121, p = 0.656,\) Figure 3).  
+
+<|ref|>text<|/ref|><|det|>[[112, 222, 881, 562]]<|/det|>
+We then evaluated the impact of timing of vaccination on maternal antibody levels at delivery. We found no statistically significant correlation between maternal IgG levels at delivery and time since dose 1 \((R_{s} = - 0.335, p = 0.160)\) and gestational age at delivery \((R_{s} = 0.270, p = 0.265,\) Fig 4A,B). This lack of correlation appeared to be driven by two participants that had low or absent levels of antibodies at delivery and received their first dose of vaccine within 30 days of delivery. We then evaluated neutralizing titers in those participants with known detectable IgG levels at delivery and found that maternal neutralizing titers at delivery trended with days since vaccine dose 1 but was not statistically significantly \((R_{s} = - 0.422, p = 0.093)\) , and maternal neutralizing titers at delivery was not associated with gestational age at dose 1 \((R_{s} = 0.074,\) \(p = 0.780)\) . One participant was borderline IgG positive at delivery (vaccinated within 30 days of delivery) and did not have detectable neutralizing titers at delivery (Fig 4C,D).  
+
+<|ref|>text<|/ref|><|det|>[[112, 574, 884, 880]]<|/det|>
+To assess facilitated antibody transfer, we evaluated cord- to- maternal antibody IgG and neutralization titer ratios by time since vaccination and gestational age. We found that IgG ratios were highly correlated with both time since first maternal vaccination dose and gestational age at first dose \((R_{s} = 0.917, p< 0.0001\) and \(R_{s} = - 0.739, p = 0.002\) , respectively. Fig 4E,F). However, neutralization titer cord- to- maternal ratios by time since first vaccination dose and gestational age at first dose were not significantly associated \((R_{s} = 0.366, p = 0.179\) and \(R_{s} = - 0.032, p = 0.913\) , respectively, Figure 4G,H). Together, this may indicate that timing of vaccination in pregnancy is critical for maternal- fetal antibody transfer, and functional neutralizing antibodies are differentially transferred to the fetus as compared to total anti- SARS- CoV- 2 IgG during gestation.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[113, 95, 875, 146]]<|/det|>
+## mRNA vaccination leads to a unique SARS-CoV2 Spike protein antibody epitope binding signature  
+
+<|ref|>text<|/ref|><|det|>[[112, 157, 879, 370]]<|/det|>
+We next investigated antibody linear epitope binding and transplacental transfer using the PhIP- seq/VirScan SARS- CoV- 2 Spike protein phage display array in mother- infant dyads at the time of birth (Figure 5). We found that timing of vaccination was important for the transplacental transfer of Spike protein epitope binding antibodies. Two mother- infant dyads had minimal Spike protein specific epitope binding. The first dyad only received one dose of mRNA vaccine 9 days prior to delivery, and the other dyad received the second vaccine dose 2 days prior to delivery.  
+
+<|ref|>text<|/ref|><|det|>[[111, 380, 881, 818]]<|/det|>
+We found high levels of SARS- CoV- 2 Spike protein epitope binding in 4 major peaks we designate as regions 1- 4 (Figure 5A). Region 1 overlays the carboxy terminal of the N- terminal domain. Region 2 overlaps with key residues for the S1/S2 cleavage site. Regions 3 and 4 are within S2, flanking the fusion loop and the transmembrane portion of the Spike protein, respectively. However, we found minimal binding in the receptor binding domain (RBD) of Spike protein. Prior evaluation using the PhIP- seq/VirScan SARS- CoV- 2 epitope phage array during SARS- CoV- 2 infection demonstrated similar binding in regions 3 and 4, however in SARS- CoV- 2 infection there was minimal binding in regions 1 and 2 demonstrating that antibody epitope binding in these regions may be unique to vaccination<sup>18</sup>. Additionally, there is proportional representation of linear epitope binding across the SARS- CoV- 2 Spike protein proteome between mothers and infants (Figure 5B). Taken together, SARS- CoV- 2 antibody linear epitope binding after vaccination during pregnancy shows similar patterns, with multiple immunodominant regions found in the majority of mothers and infants. Some of these regions are unique to vaccination and not observed during natural infection<sup>18- 20</sup>.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 862, 214, 879]]<|/det|>
+## Discussion
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 63, 884, 339]]<|/det|>
+Among twenty women who received the COVID- 19 mRNA vaccine during pregnancy, our study found no evidence of transplacental transfer of mRNA vaccine products but did find high levels of functional vaccine- derived antibodies that transferred to the infant at delivery and persisted during early infancy. Additionally, we identified high levels of epitope binding in two regions of Spike protein unique to SARS- CoV- 2 vaccination<sup>18</sup>. These data may address some of the many unanswered questions regarding COVID- 19 vaccination in pregnancy: including the dynamics of antibody production in the pregnant immune state, and the optimal timing of immunization in pregnancy to impart passive immunity to the newborn during the vulnerable first few weeks of infancy.  
+
+<|ref|>text<|/ref|><|det|>[[111, 350, 881, 601]]<|/det|>
+Uptake of COVID- 19 vaccination in pregnancy has been slow<sup>5</sup>, and reasons for vaccine hesitancy are likely multifactorial — but theoretical concerns that vaccine mRNA could cross the placenta have been raised. We found no evidence of mRNA vaccine products in any of our delivery samples. Additionally, no infants in our study had a fetal immune response to Spike protein as demonstrated by a negative anti- SARS- CoV- 2 IgM antibody in cord blood and infant follow up samples. This further supports the lack of transfer of vaccine products, as only IgG is transferred from the mother, and IgM production would indicate an endogenous fetal immune response which has rarely been seen in natural infection with SARS- CoV- 2 during pregnancy<sup>16,21- 23</sup>. This provides additional reassurance that mRNA vaccination is safe during pregnancy.  
+
+<|ref|>text<|/ref|><|det|>[[111, 612, 881, 882]]<|/det|>
+We found that the timing of immunization during pregnancy is important to ensure transplacental transfer of protective antibodies to the neonate, and during critical windows of immune vulnerability during early infancy. Consistent with prior studies showing robust immune responses to mRNA vaccination<sup>14,15,17</sup>, we found high levels of IgG after two doses of mRNA vaccine. However, completion of the vaccination series well before delivery was important to ensure transfer of antibodies to the infant. Two mothers only received one vaccine dose prior to delivery and did not transfer antibodies as demonstrated by the lack of antibodies in cord (in one with available cord blood) and in both infants at follow- up. Additionally, neutralizing antibodies
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 61, 879, 275]]<|/det|>
+were not transferred in a mother who received her second dose of vaccine 2 days prior to delivery. All evaluated mothers who received both doses during pregnancy and with the second dose greater than 9 days prior to delivery transferred IgG and neutralizing antibodies to their infants. Consistent with early studies of antibody transfer after COVID- 19 vaccination in pregnancy, most of our participants were vaccinated in the third trimester of pregnancy. Larger studies of individuals vaccinated prior to pregnancy and in the first and second trimester are needed to understand persistence and waning of vaccine- induced immune responses.  
+
+<|ref|>text<|/ref|><|det|>[[112, 285, 880, 595]]<|/det|>
+Additionally, we believe we are the first to report that infants in the first few months of life continued to have maternal vaccine- derived anti- SARS- CoV- 2 antibodies that were functional as demonstrated by high levels of neutralizing antibodies presenting infants up to 12 weeks of age. This is consistent with known persistence of maternally- derived antibodies from other vaccinations including pertussis, rubella, varicella<sup>24- 26</sup>. Additionally, we have previously found persistence of anti- SARS- CoV- 2 IgG antibodies in infants after natural infection up to 6 months<sup>16</sup>. However, the functional capability of these antibodies as compared to anti- SARS- CoV- 2 vaccination- derived antibodies is unknown. Further evaluation of the longitudinal persistence of maternal vaccine- derived antibodies during infancy will be critical to determine optimal timing of COVID- 19 vaccination in infancy.  
+
+<|ref|>text<|/ref|><|det|>[[113, 605, 879, 785]]<|/det|>
+Evaluation of paired maternal and baby samples at post- partum follow up timepoints showed a faster decline in maternal IgG antibody levels than infants, suggesting that persistence of maternally- derived antibody may be prolonged for infants. Differences in renal excretion and neonatal Fc receptor (FcRn) expression, which is involved in antibody degradation<sup>27</sup> in the infant as compared to adults, could underly these differences and should be explored further.  
+
+<|ref|>text<|/ref|><|det|>[[113, 797, 884, 881]]<|/det|>
+Consistent with observations in non- pregnant adults, we found that IgG levels in mothers at delivery, and at infant follow- up were highly correlated with neutralizing titers<sup>28</sup>. However, cord blood IgG levels did not correlate with neutralizing titers. Moreover, IgG cord- to- maternal
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 62, 875, 404]]<|/det|>
+ratios, which represent a proxy of maternal to fetal antibody transfer, were highly correlated with timing of vaccination (gestational age and days since the first dose), but cord- to- maternal neutralizing titer ratios were not significantly associated with time since vaccination nor gestational age. During gestation there is facilitated transfer of maternally derived antibodies through the binding of the neonatal Fc receptor in the synctiotrophoblast layer<sup>29</sup>. Differences in glycosylation<sup>30,31</sup>, FcR/FcRn binding affinity<sup>17,32</sup>, preferential IgG subclass transfer<sup>33,34</sup> may be different in functional neutralizing antibodies as compared to total IgG antibody transfer. However, a limitation of this study is the majority of participants were vaccinated in the third trimester. Further investigations on factors that influence the transport of functional antibodies across trimesters are needed to understand antibody dynamics and optimal transfer of protective antibodies to infants.  
+
+<|ref|>text<|/ref|><|det|>[[112, 414, 880, 787]]<|/det|>
+Using a PhIP- seq/VirScan SARS- CoV- 2 Spike protein phage array we were able to compare linear epitope antibody binding in mothers and their infants. Consistent with IgG and neutralizing antibody evaluation, timing of vaccination was critical to ensure the transplacental transfer of antibodies to the infant. Additionally, we identified unique regions of antibody epitope binding in our vaccinated cohort that were not identified using the same phage library in a prior evaluation of a cohort of SARS- CoV- 2 infected individuals<sup>18</sup>. One of these regions included the carboxy terminal of the N- terminal domain, with other work having shown that the N- terminal domain is targeted by neutralizing antibodies against Spike protein<sup>35</sup>. We did not see significant binding in the receptor binding domain (RBD), which may be attributable to the fact that the phage display library displayed short, linear peptides while antibodies targeting RBD are known to target conformational epitopes. Lastly, we found that the same immunodominant regions targeted by antibodies targeting the Spike protein in both mothers and infants.  
+
+<|ref|>text<|/ref|><|det|>[[113, 798, 860, 881]]<|/det|>
+In summary, this work provides further evidence that mRNA vaccination is safe in pregnancy and demonstrates that it generates time- dependent protective, functional antibody responses in mothers and infants that persist during early infancy.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 97, 193, 114]]<|/det|>
+## Methods  
+
+<|ref|>text<|/ref|><|det|>[[111, 125, 880, 469]]<|/det|>
+Cohort and Sample collection: The University of California San Francisco (UCSF) institutional review board approved the study (20- 32077). Written informed consent was obtained from all participants. We enrolled 20 pregnant individuals who were vaccinated with either BNT- 162b2 or mRNA- 1273 mRNA vaccines. Pregnant individuals were followed through delivery, and their infants were followed up to 12 weeks of life. Maternal blood was collected during pregnancy (pre- vaccine, 3- 4 weeks post- dose 1, 4- 8 weeks post- dose 2). During delivery, maternal blood, placenta tissue, and cord blood was collected. Infant follow- up blood was collected at convenience timepoints. Whole blood was immediately added to RNAlater in a 1:1.3 ratio. Plasma was isolated from whole blood by centrifugation and immediately cryopreserved. Full- thickness placental biopsy was collected within 1 hour of delivery, washed three times with phosphate buffered saline, and preserved in RNAlater.  
+
+<|ref|>text<|/ref|><|det|>[[111, 510, 884, 850]]<|/det|>
+SARS- CoV- 2 plasma serology. Anti- SARS- CoV- 2 plasma IgM and IgG antibodies were measured using the Pylon 3D automated immunoassay system (ET Healthcare, Palo Alto, CA). In brief, quartz glass probes are pre- coated with either affinity purified goat anti Human IgM (IgM capture) or Protein G (IgG capture) are dipped into diluted patient sample. Samples are washed, and then the probe is dipped into the assay reagent containing both biotinylated recombinant spike protein receptor binding domain (RBD) and nucleocapsid protein (NP). After a washing, the probe is incubated with a Cy@5- streptavidin (Cy5- SA) polysaccharide conjugate reagent, allowing for cyclic amplification of the fluorescence signal. The background corrected signal is reported as relative fluorescent units (RFU) which is proportional to the amount of specific antibodies in the sample allowing for quantification. Levels of IgM and IgG were considered positive if greater than 50 relative fluorescence units.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 64, 388, 82]]<|/det|>
+## SARS-CoV-2 neutralizing assay  
+
+<|ref|>text<|/ref|><|det|>[[111, 92, 879, 660]]<|/det|>
+SARS- CoV- 2 antibody neutralization titers were measured using a label- free surrogate neutralization assay (LF- sVNT) previously described28. Briefly, the method measures the binding ability of recombinant RBD (Sino Biological, Wayne, PA) coated onto sensing probes (Gator Bio, Palo Alto, CA) to recombinant ACE2 (Sino Biological, Wayne, PA) after neutralizing RBD with SARS- CoV- 2 antibodies in serum. Measurements were done using a thin- film interferometry (TFI) label- free immunoassay analyzer (Gator Bio, Palo Alto, CA). Each serum sample was diluted in a series (1:25, 1:100, 1:250, 1:500, 1:1000, 1:2000) in running buffer (PBS at pH 7.4 with 0.02% Tween 20, 0.2% BSA, and 0.05% NaN3) for analysis. The first testing cycle for each diluted sample measured the binding ability of RBD to ACE2 with neutralization, and the second cycle provided the full binding ability of RBD without neutralization. In each cycle, the recorded time course of signals, as known as the sensorgram, was recorded. The readout measured the signal increase in RBD- ACE2 complex formation, representing the quantity of RBD- ACE2 complex on the sensing probe. A neutralization rate was calculated as the ratio of the readout in the first cycle to that in the second cycle, presented as a percentage. To obtain the neutralizing antibody titer (IC50) for each serum sample, the neutralization rates were plotted against dilutions, and the points were fitted using a linear interpolation model. The reciprocal of the dilution resulting in a 50% neutralization rate was defined as the neutralizing antibody titer.  
+
+<|ref|>text<|/ref|><|det|>[[113, 702, 881, 880]]<|/det|>
+SARS- CoV- 2 Spike protein Western blot. Maternal blood and cord blood were diluted in RNAlater in 1:1.3 ratio, placenta was preserved in RNAlater. Protein lysates were obtained from samples using RIPA buffer (150 mM NaCl, 25 mM Tris- HCl (pH 7.4), 1% NP- 40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate) containing Halt™ protease inhibitor cocktail (ThermoScientific). Cell Lysates were resolved by SDS/PAGE on a Bis- Tris methane 4–12% polyacrylamide gel and transferred to a nitrocellulose membrane, blocked with 5% skimmed
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 63, 881, 277]]<|/det|>
+milk diluted in PBS, an incubated overnight at \(4^{\circ}C\) with anti- SARS- CoV- 2 Spike mouse mAb (1A9, GeneTex) or anti- GAPDH rabbit polyclonal antibody (GTX100118, GeneTex) respectively diluted 1:1,000 or 1:5,000 in blocking buffer. The membrane was washed in PBS buffer containing Tween- 20 (0.1%) and then incubated for 1 h with horseradish peroxidase- conjugated anti- mouse and anti- rabbit secondary antibody (Jackson ImmunoResearch) diluted respectively 1:5,000 and 1:10,000. The membrane was thoroughly washed, and proteins visualized using Immobilon Forte Western HRP substrate (Millipore).  
+
+<|ref|>text<|/ref|><|det|>[[111, 320, 884, 531]]<|/det|>
+SARS- CoV- 2 Spike mRNA PCR. Maternal blood and cord blood were diluted in RNAlater in 1:1.3 ratio, placenta was preserved in RNAlater. Tissues were kept at \(- 80^{\circ}C\) until analyzed. RNA was isolated from samples using the RNeasy Micro or Mini Kit (Qiagen) according to manufacturer's protocol. RNA concentration was measured using nanodrop and all samples had \(>30 \text{ng / ul}\) total RNA. \(500 \text{ng}\) RNA was transcribed into cDNA using qScript cDNA synthesis kit (Quantabio). Primers were design to detect the vaccines mRNA (mRNA- 1273 Moderna and BNT162b2 Pfizer- BioNtech) as previously described<sup>36</sup>. Forward primer:  
+
+<|ref|>text<|/ref|><|det|>[[111, 543, 884, 820]]<|/det|>
+AACGCCACCAACGTGGTCATC. Reverse primer: GTTGTTGGCGCTGCTGTACAC. Primers were shown to detect samples containing as low as 1.5 pg of vaccine using vaccine standard curve (Table S2). QuantaStudio 6 Flex (Applied Biosystems) instrument and SsoFast EvaGreen supermix (Bio- Rad) were used for PCR reaction: 30 second \(95^{\circ}C\) followed by 40 cycles of 5 second \(95^{\circ}C\) and 20 seconds \(60^{\circ}C\) . All samples were run in triplicate as 20 μL reactions, and Ct values corresponding to \(< 1.5 \text{pg}\) of vaccine based on standard curve (Table S2) were interpreted as a negative result. For vaccines cDNA standard curves, \(10000 \text{pg / \mu L}\) vaccine mRNA (as cDNA) sample was used for serial dilution in 1:3 ratio, up to \(0.06 \text{pg / \mu L}\) . Two \(\mu \text{L}\) of these diluted samples were used in each well to create standard curves.  
+
+<|ref|>text<|/ref|><|det|>[[113, 863, 561, 883]]<|/det|>
+PhIP- Seq/VirScan Coronavirus phage display assay
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 63, 559, 82]]<|/det|>
+Immunoprecipitation of phage- bound patient antibodies  
+
+<|ref|>text<|/ref|><|det|>[[111, 94, 877, 305]]<|/det|>
+Maternal plasma at delivery and cord plasma were evaluated by PhIP- Seq/Virscan Coronavirus phage display. Construction of the Coronavirus PhIP- Seq library and detailed methods for immunoprecipitation, sequencing and bioinformatic processing of data are identical to what has previously been described<sup>18</sup>. For the purposes of the analysis conducted in this study, analysis was restricted to sero- reactivity against the SARS- CoV- 2 Spike protein. As previously described, a total of two rounds of amplification and selection were performed for all PhIP- Seq analyses.  
+
+<|ref|>text<|/ref|><|det|>[[112, 350, 445, 369]]<|/det|>
+Next Generation Sequencing library prep  
+
+<|ref|>text<|/ref|><|det|>[[111, 381, 874, 595]]<|/det|>
+Amplicon sequencing library preps were performed using the Labcyte Echo 525 and an Integra Via Flow 96 and were identical to what has previously been described<sup>18</sup>. All libraries were pooled by equal volume, cleaned and size selected using Ampure XP beads at 1.0X per manufacturer's protocol. Libraries were quantified by High Sensitivity DNA Qubit and quality- checked by High Sensitivity DNA Bioanalyzer. Sequencing was then performed on a NovaSeq S1 (300 cycle kit with 1.3 billion clusters) aiming for sequencing depths of at least 1 million reads per sample.  
+
+<|ref|>text<|/ref|><|det|>[[112, 640, 441, 658]]<|/det|>
+Bioinformatic Analysis of PhIP- Seq Data  
+
+<|ref|>text<|/ref|><|det|>[[111, 669, 883, 882]]<|/det|>
+Sequencing reads were aligned to a reference database of the full viral peptide library using the Bowtie2 aligner. For all VirScan libraries, the null distribution of each peptide's log10(rpK) was modeled using a set of 95 pre- pandemic, healthy control sera. All counts were augmented by 1 to avoid zero counts in the healthy control sera samples. Multiple distribution fits were examined for these data, with the Normal distribution showing the best fit. These null distributions were used to calculate p values for the observed log10(rpK) of each peptide within a given sample. The calculated p values were corrected for multiple hypothesis using the Benjamini- Hochberg
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 62, 880, 243]]<|/det|>
+method. Any peptide with a corrected p value of \(< 0.001\) was considered significantly enriched over the healthy background. To identify regions targeted by host antibodies, all library peptides were aligned to the SARS- CoV- 2 reference genome. Using the aligned position of the significantly enriched peptides which aligned full- length against the reference, we determined the proportion of individuals (mothers and infants) that were reactive at each residue of the Spike protein. All plots were generated using the R ggplot2 package.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 288, 284, 306]]<|/det|>
+## Statistical analysis:  
+
+<|ref|>text<|/ref|><|det|>[[111, 318, 884, 594]]<|/det|>
+Statistical analyses were performed using PRISM v9.2 (GraphPad), STATA 16 (StataCorp), and R version 3.6.3 and R Studio version 1.1.447. Descriptive statistics include mean, standard deviations, and ranges for continuous variables. The Wilcoxon rank- sum test was used for two- group comparisons of continuous variables including maternal pre- and post- vaccine antibody responses. Associations between continuous variables were assessed using Spearman's rank correlation \((R_{s})\) including comparisons between maternal, cord and infant follow- up antibody IgG and neutralizing titer responses, and timing of vaccination. Two- sided \(p\) values were calculated for all test statistics, and \(p< 0.05\) was considered significant. PhIP- Seq/VirScan bioinformatics as detailed above.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 640, 258, 657]]<|/det|>
+## Data Availability  
+
+<|ref|>text<|/ref|><|det|>[[115, 670, 867, 721]]<|/det|>
+The data set generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 767, 293, 784]]<|/det|>
+## Acknowledgements:  
+
+<|ref|>text<|/ref|><|det|>[[112, 797, 872, 880]]<|/det|>
+M.P. was supported by the National Institutes of Health (NIAID K23AI127886), the Marino Family Foundation, and UCSF REAC award. Y.G. was supported by the Weizmann Institute of Science - National Postdoctoral Award Program for Advancing Women in Science, and of the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 62, 880, 180]]<|/det|>
+International Society for Research in Human Milk and Lactation (ISRHDL) Trainee Bridge Fund. Y.M. and W.C.G. were supported by The Roddenberry Foundation. S.L.G. was supported by the National Institutes of Health (NIAID K08AI141728), and the Bill and Melinda Gates Foundation (INV- 017035), and a generous gift from the Kryzewski Family.  
+
+<|ref|>text<|/ref|><|det|>[[111, 222, 884, 403]]<|/det|>
+We thank all the mothers and infants that participated in this study. We thank Kenneth Scott, BS, RPh, (UCSF Health Pharmacy, University of California, San Francisco) and Hannah J. Jang, PhD, RN, PHN, CNL (UCSF School of Nursing, University of California, San Francisco), for voluntarily providing unused vaccine for this study, and to Dr. Margaret Feeney (University of California, San Francisco) and Dr. Nadav Ahituv (University of California, San Francisco) for support of these experiments.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 448, 303, 466]]<|/det|>
+## Author contributions:  
+
+<|ref|>text<|/ref|><|det|>[[111, 480, 865, 572]]<|/det|>
+M.P. Helped conceive and design the project, oversaw recruitment, designed, and performed sample collection, oversaw experiment design, oversaw data analysis, provided funding, and drafted the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[111, 590, 868, 647]]<|/det|>
+Y.G. Recruited and consented enrollees, oversaw sample collection, designed, performed, and analyzed mRNA PCR experiments, performed data analysis.  
+
+<|ref|>text<|/ref|><|det|>[[111, 664, 872, 720]]<|/det|>
+A.G.C. Recruited and consented enrollees, oversaw sample collection, performed chart review, and helped draft the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[111, 737, 509, 757]]<|/det|>
+Y.M. Performed and helped design Western blot.  
+
+<|ref|>text<|/ref|><|det|>[[111, 775, 780, 796]]<|/det|>
+L.L. Performed and analyzed mRNA PCR experiments, performed sample collection.  
+
+<|ref|>text<|/ref|><|det|>[[111, 814, 528, 833]]<|/det|>
+B.A. Performed phage immunoprecipitation assays.  
+
+<|ref|>text<|/ref|><|det|>[[111, 850, 785, 870]]<|/det|>
+H.C. and U.J. performed and helped design critical experiments, and data collection.  
+
+<|ref|>text<|/ref|><|det|>[[111, 887, 875, 907]]<|/det|>
+C.Y.L., V.J.L., M.C., L.W., S.B. Performed and coordinated sample collection, and data collection.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[58, 66, 518, 85]]<|/det|>
+V.J.F. Helped conceive and coordinate the project.  
+
+<|ref|>text<|/ref|><|det|>[[115, 102, 308, 120]]<|/det|>
+A.P.M. Provide funding.  
+
+<|ref|>text<|/ref|><|det|>[[115, 140, 614, 158]]<|/det|>
+W.C.G. Helped design western blot and oversaw data analysis.  
+
+<|ref|>text<|/ref|><|det|>[[115, 177, 572, 195]]<|/det|>
+A.H.B.W Designed and oversaw all serology experiments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 214, 649, 232]]<|/det|>
+K.L.L. Designed and oversaw all neutralizing antibody experiments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 250, 795, 269]]<|/det|>
+J.R. Designed, analyzed, and oversaw phage immunoprecipitation sequencing assays.  
+
+<|ref|>text<|/ref|><|det|>[[115, 286, 861, 305]]<|/det|>
+S.L.G. conceived and designed the project, oversaw recruitment, oversaw experiment design,  
+
+<|ref|>text<|/ref|><|det|>[[115, 323, 710, 342]]<|/det|>
+oversaw data analysis, provided funding, and helped draft the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 360, 772, 379]]<|/det|>
+M.P., Y.G., Y.U., L.L., A.H.B.W, W.C.G, K.L.L., and S.L.G verified the underlying data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 398, 528, 416]]<|/det|>
+All authors reviewed and approved the manuscript.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[140, 70, 884, 333]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[84, 358, 820, 377]]<|/det|>
+<center>Figure 1. Anti-SARS-CoV-2 IgG and IgM antibody responses following vaccination </center>  
+
+<|ref|>text<|/ref|><|det|>[[85, 388, 880, 538]]<|/det|>
+A. Maternal plasma anti-SARS-CoV-2 IgG antibody relative fluorescence units (RFU) levels prior to vaccination \((n = 4)\) , 3-4 weeks post-dose 1 \((n = 7)\) , and 4-8 weeks post-dose 2 \((n = 12)\) . B. Maternal plasma anti-SARS-CoV-2 IgM (RFU) levels prior to vaccination \((n = 4)\) , 3-4 weeks post-dose 1 \((n = 7)\) , and 4-8 weeks post-dose 2 \((n = 12)\) . Wilcoxon rank-sum testing. Data represent median ± quartiles, two-sided \(p\) values were calculated for all test statistics.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 70, 865, 750]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 763, 767, 782]]<|/det|>
+<center>Figure 2. Paired maternal, cord, and infant IgG and neutralization antibodies </center>  
+
+<|ref|>text<|/ref|><|det|>[[56, 796, 870, 882]]<|/det|>
+481 A. Paired maternal plasma at delivery \((n = 19)\) , cord plasma \((n = 17)\) , and infant follow-up \((n = 10)\) 482 by anti- SARS- CoV- 2 IgG antibody relative fluorescence units (RFU), (Spearman's rank 483 correlation, dotted line indicates positive cutoff value of 50 RFU). B. Paired maternal plasma at
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 61, 881, 305]]<|/det|>
+delivery (n=17), cord plasma (n=16), and infant follow-up (n=8) by SARS- CoV- 2 label- free surrogate neutralization assay (sVNT), (Spearman's rank correlation, dotted line indicates positive cutoff value of 25). C. Paired cord plasma (n=9) and infant follow-up plasma (n=11) anti- SARS- CoV- 2 IgG by weeks of life. D. Paired cord plasma (n=7) and infant follow-up plasma (n=8) label- free surrogate neutralization assay (sVNT) by weeks of life. E. Paired maternal plasma at delivery (n=5), cord plasma (n=5), and paired maternal follow-up (n=5) and infant follow- up plasma (n=5) anti- SARS- CoV- 2 IgG. Two- sided \(p\) values were calculated for all test statistics.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[118, 68, 875, 272]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 292, 553, 311]]<|/det|>
+<center>Figure 3. Neutralization to IgG antibody correlation </center>  
+
+<|ref|>text<|/ref|><|det|>[[112, 323, 867, 437]]<|/det|>
+A. Maternal plasma at delivery \((n = 17)\) B. Cord plasma \((n = 16)\) C. Infant follow-up plasma \((n = 8)\) SARS-CoV-2 label-free surrogate neutralization assay (sVNT) by anti-SARS-CoV-2 IgG correlation (Spearman's rank correlation). Two-sided \(p\) values were calculated for all test statistics.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 95, 875, 420]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 448, 799, 467]]<|/det|>
+<center>Figure 4. Maternal delivery and Cord-to-maternal antibody transfer ratios timing </center>  
+
+<|ref|>text<|/ref|><|det|>[[111, 478, 880, 852]]<|/det|>
+A. Maternal delivery anti-SARS-CoV-2 IgG antibody transfer ratio by days since vaccine dose 1 (n=19, dashed line indicates positive cutoff >50 RFU) 
+B. Maternal delivery anti-SARS-CoV-2 IgG antibody transfer ratio by gestational age at vaccine dose 1 (n=19, dashed line indicates positive cutoff >50 RFU) 
+C. Maternal delivery SARS-CoV-2 label-free surrogate neutralization assay (sVNT) antibody transfer ratio by days since vaccine dose 1 (n=17, dashed line indicates positive cutoff >25). 
+D. Maternal delivery SARS-CoV-2 label-free surrogate neutralization assay (sVNT) antibody transfer ratio by gestational age at vaccine dose 1 (n=17, dashed line indicates positive cutoff >25) 
+E. Cord-to-maternal anti-SARS-CoV-2 IgG antibody transfer ratio by days since vaccine dose 1 (n=15) 
+F. Cord-to-maternal anti-SARS-CoV-2 IgG antibody transfer ratio by gestational age at vaccine dose 1 (n=15) 
+G. Cord-to-maternal SARS-CoV-2 label-free surrogate neutralization assay (sVNT) antibody transfer ratio by days since vaccine dose 1 (n=15). 
+H. Cord-to-maternal SARS-CoV-2 label-free surrogate neutralization assay (sVNT)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 63, 857, 115]]<|/det|>
+536 antibody transfer ratio by gestational age at vaccine dose 1 (n=15). Two-sided \(p\) values were calculated for all test statistics.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 70, 866, 744]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[111, 742, 866, 763]]<|/det|>
+<center>Figure 5. PhIP-seq/VirScan paired maternal and cord SARS-CoV-2 Spike protein epitope </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[112, 774, 185, 792]]<|/det|>
+## binding  
+
+<|ref|>text<|/ref|><|det|>[[111, 805, 866, 888]]<|/det|>
+A. Heatmap displaying results of significant enriched (p<0.001) linear SARS-CoV-2 Spike protein epitope binding from 15 paired mother-infant dyads in maternal plasma at delivery and cord plasma by vaccine type and time since vaccine dose 1. Areas of high cumulative epitope
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 64, 872, 85]]<|/det|>
+567 binding designated by regions 1- 4. B. Cumulative fold enrichment of mothers and infants linear 
+
+<|ref|>text<|/ref|><|det|>[[57, 98, 466, 116]]<|/det|>
+568 SARS-CoV-2 Spike protein epitope binding.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[171, 95, 881, 228]]<|/det|>
+Ellington, S. et al. Characteristics of Women of Reproductive Age with Laboratory- Confirmed SARS- CoV- 2 Infection by Pregnancy Status - United States, January 22- June 7, 2020. MMWR Morb Mortal Wkly Rep 69, 769- 775, doi:10.15585/mmwr.mm6925a1 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[171, 244, 835, 303]]<|/det|>
+Ahlberg, M. et al. Association of SARS- CoV- 2 Test Status and Pregnancy Outcomes. Jama, doi:10.1001/jama.2020.19124 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[171, 317, 886, 413]]<|/det|>
+Knight, M. et al. Characteristics and outcomes of pregnant women admitted to hospital with confirmed SARS- CoV- 2 infection in UK: national population based cohort study. Bmj 369, m2107, doi:10.1136/bmj.m2107 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[171, 428, 860, 560]]<|/det|>
+Martinez- Portilla, R. J. et al. Pregnant women with SARS- CoV- 2 infection are at higher risk of death and pneumonia: propensity score matched analysis of a nationwide prospective cohort (COV19Mx). Ultrasound Obstet Gynecol 57, 224- 231, doi:10.1002/uog.23575 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[171, 576, 840, 636]]<|/det|>
+CDC. COVID- 19 vaccination among pregnant people aged 18- 49 years overall, by race/ethnicity, and date reported to CDC - Vaccine Safety Datalink, \* United States,  
+
+<|ref|>text<|/ref|><|det|>[[171, 651, 840, 672]]<|/det|>
+<https://covid.cdc.gov/covid- data- tracker/#vaccinations- pregnant- women> (2021).  
+
+<|ref|>text<|/ref|><|det|>[[171, 687, 864, 747]]<|/det|>
+Gynecologists, A. C. o. O. a. Statement of Strong Medical Consensus for Vaccination of Pregnant Individuals Against COVID- 19, <https://www.acog.org/news/news-  
+
+<|ref|>text<|/ref|><|det|>[[171, 762, 876, 782]]<|/det|>
+releases/2021/08/statement- of- strong- medical- consensus- for- vaccination- of- pregnant-  
+
+<|ref|>text<|/ref|><|det|>[[171, 799, 470, 818]]<|/det|>
+individuals- against- covid- 19> (2021).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[170, 63, 875, 160]]<|/det|>
+Maykin, M. M., Heuser, C., Feltovich, H. & with the Society for Maternal- Fetal Medicine Health Policy Advocacy, C. Pregnant people deserve the protection offered by SARS- CoV- 2 vaccines. Vaccine 39, 171- 172, doi:10.1016/j.vaccine.2020.12.007 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[170, 175, 872, 234]]<|/det|>
+Jackson, L. A. et al. An mRNA Vaccine against SARS- CoV- 2 - Preliminary Report. N Engl J Med 383, 1920- 1931, doi:10.1056/NEJMoa2022483 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[170, 248, 877, 307]]<|/det|>
+Polack, F. P. et al. Safety and Efficacy of the BNT162b2 mRNA Covid- 19 Vaccine. N Engl J Med 383, 2603- 2615, doi:10.1056/NEJMoa2034577 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[170, 322, 840, 381]]<|/det|>
+Atveo, C. et al. Compromised SARS- CoV- 2- specific placental antibody transfer. Cell, doi:10.1016/j.cell.2020.12.027 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[170, 396, 860, 520]]<|/det|>
+Edlow, A. G. et al. Assessment of Maternal and Neonatal SARS- CoV- 2 Viral Load, Transplacental Antibody Transfer, and Placental Pathology in Pregnancies During the COVID- 19 Pandemic. JAMA Netw Open 3, e2030455, doi:10.1001/jamanetworkopen.2020.30455 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[170, 545, 840, 640]]<|/det|>
+Flannery, D. D. et al. Assessment of Maternal and Neonatal Cord Blood SARS- CoV- 2 Antibodies and Placental Transfer Ratios. JAMA Pediatr, doi:10.1001/jamapediatrics.2021.0038 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[170, 655, 871, 714]]<|/det|>
+Atveo, C. et al. Compromised SARS- CoV- 2- specific placental antibody transfer. Cell 184, 628- 642. e610, doi:10.1016/j.cell.2020.12.027 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[170, 728, 856, 825]]<|/det|>
+Gray, K. J. et al. Coronavirus disease 2019 vaccine response in pregnant and lactating women: a cohort study. Am J Obstet Gynecol 225, 303. e301- 303. e317, doi:10.1016/j.ajog.2021.03.023 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[170, 840, 856, 899]]<|/det|>
+Nir, O. et al. Maternal- neonatal transfer of SARS- CoV- 2 immunoglobulin G antibodies among parturient women treated with BNT162b2 messenger RNA vaccine during
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[170, 63, 850, 120]]<|/det|>
+pregnancy. Am J Obstet Gynecol MFM 4, 100492, doi:10.1016/j.ajogmf.2021.100492 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[170, 137, 870, 233]]<|/det|>
+Song, D. et al. Passive and active immunity in infants born to mothers with SARS-CoV- 2 infection during pregnancy: prospective cohort study. BMJ Open 11, e053036, doi:10.1136/bmjopen- 2021- 053036 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[170, 248, 844, 344]]<|/det|>
+Atyeo, C. et al. COVID- 19 mRNA vaccines drive differential antibody Fc- functional profiles in pregnant, lactating, and non- pregnant women. Sci Transl Med, eabi8631, doi:10.1126/scitranslmed.abi8631 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[170, 359, 872, 493]]<|/det|>
+Zamecnik, C. R. et al. ReScan, a Multiplex Diagnostic Pipeline, Pans Human Sera for SARS- CoV- 2 Antigens. Cell Rep Med 1, 100123, doi:10.1016/j.xcrm.2020.100123 (2020). Shrock, E. et al. Viral epitope profiling of COVID- 19 patients reveals cross- reactivity and correlates of severity. Science 370, doi:10.1126/science.abd4250 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[170, 507, 870, 602]]<|/det|>
+Li, Y. et al. Linear epitopes of SARS- CoV- 2 spike protein elicit neutralizing antibodies in COVID- 19 patients. Cell Mol Immunol 17, 1095- 1097, doi:10.1038/s41423- 020- 00523- 5 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[170, 618, 845, 675]]<|/det|>
+Fenizia, C. et al. Analysis of SARS- CoV- 2 vertical transmission during pregnancy. Nat Commun 11, 5128, doi:10.1038/s41467- 020- 18933- 4 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[170, 691, 868, 748]]<|/det|>
+Zeng, H. et al. Antibodies in Infants Born to Mothers With COVID- 19 Pneumonia. JAMA 323, 1848- 1849, doi:10.1001/jama.2020.4861 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[170, 764, 864, 820]]<|/det|>
+Zeng, L. et al. Neonatal Early- Onset Infection With SARS- CoV- 2 in 33 Neonates Born to Mothers With COVID- 19 in Wuhan, China. JAMA Pediatr 174, 722- 725,  
+
+<|ref|>text<|/ref|><|det|>[[170, 837, 547, 857]]<|/det|>
+doi:10.1001/jamapediatrics.2020.0878 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[170, 63, 883, 157]]<|/det|>
+Healy, C. M. et al. Kinetics of maternal pertussis-specific antibodies in infants of mothers vaccinated with tetanus, diphtheria and acellular pertussis (Tdap) during pregnancy. Vaccine 38, 5955- 5961, doi:10.1016/j.vaccine.2020.06.050 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[170, 174, 875, 303]]<|/det|>
+Sissoko, M. S. et al. Safety and efficacy of PfSPZ Vaccine against Plasmodium falciparum via direct venous inoculation in healthy malaria- exposed adults in Mali: a randomised, double- blind phase 1 trial. The Lancet Infectious Diseases 17, 498- 509, doi:10.1016/s1473- 3099(17)30104- 4 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[170, 322, 821, 417]]<|/det|>
+Leuridan, E., Hens, N., Hutse, V., Aerts, M. & Van Damme, P. Kinetics of maternal antibodies against rubella and varicella in infants. Vaccine 29, 2222- 2226, doi:10.1016/j.vaccine.2010.06.004 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[170, 433, 875, 490]]<|/det|>
+Pyzik, M. et al. The Neonatal Fc Receptor (FcRn): A Misnomer? Front Immunol 10, 1540, doi:10.3389/fimmu.2019.01540 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[170, 507, 877, 712]]<|/det|>
+Luo, Y. R., Yun, C., Chakraborty, I., Wu, A. H. B. & Lynch, K. L. A SARS- CoV- 2 Label- Free Surrogate Virus Neutralization Test and a Longitudinal Study of Antibody Characteristics in COVID- 19 Patients. J Clin Microbiol 59, e0019321, doi:10.1128/jcm.00193- 21 (2021). Firan, M. et al. The MHC class I- related receptor, FcRn, plays an essential role in the maternofetal transfer of gamma- globulin in humans. Int Immunol 13, 993- 1002, doi:10.1093/intimm/13.8.993 (2001).  
+
+<|ref|>text<|/ref|><|det|>[[170, 729, 872, 863]]<|/det|>
+Williams, P. J. et al. Short communication: selective placental transport of maternal IgG to the fetus. Placenta 16, 749- 756, doi:10.1016/0143- 4004(95)90018- 7 (1995). Kibe, T. et al. Glycosylation and Placental Transport of Immunoglobulin G. Journal of Clinical Biochemistry and Nutrition 21, 57- 63 (1996).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[55, 64, 884, 580]]<|/det|>
+681 32 Borghi, S. et al. FcRn, but not FcγRs, drives maternal-fetal transplacental transport of 682 human IgG antibodies. Proc Natl Acad Sci U S A 117, 12943- 12951, 683 doi:10.1073/pnas.2004325117 (2020). 684 33 Palmeira, P., Quinello, C., Silveira-Lessa, A. L., Zago, C. A. & Carneiro-Sampaio, M. IgG 685 placental transfer in healthy and pathological pregnancies. Clin Dev Immunol 2012, 686 985646, doi:10.1155/2012/985646 (2012). 687 34 Clements, T. et al. Update on Transplacental Transfer of IgG Subclasses: Impact of 688 Maternal and Fetal Factors. Front Immunol 11, 1920, doi:10.3389/fimmu.2020.01920 689 (2020). 690 35 Chi, X. et al. A neutralizing human antibody binds to the N-terminal domain of the Spike 691 protein of SARS-CoV-2. Science 369, 650- 655, doi:10.1126/science.abc6952 (2020). 692 36 Golan, Y. et al. Evaluation of Messenger RNA From COVID- 19 BTN162b2 and mRNA- 1273 693 Vaccines in Human Milk. JAMA Pediatr 175, 1069- 1071, 694 doi:10.1001/jamapediatrics.2021.1929 (2021).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 243, 63]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 80, 581, 95]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[55, 108, 955, 125]]<|/det|>
+- SupplementalAppendixEvaluationoftransplacentalofmRNAvaccineproductsandfunctionalantibodiesduringpregnancyandearlyinfancy.pdf
+
+<--- Page Split --->

preprint/preprint__7e97f7b9da0d9892df44202ecacbf12865c99ad875f96a7bbaa5276d0c7b668b/images_list.json

@@ -0,0 +1,168 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        100,
+        120,
+        904,
+        870
+      ]
+    ],
+    "page_idx": 14
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2: Gene expression and functional validation of Gnmr, Sardh and CG5955 in skeletal muscle in WT and obesity models.",
+    "footnote": [],
+    "bbox": [
+      [
+        84,
+        120,
+        912,
+        825
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        101,
+        140,
+        825,
+        707
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        90,
+        125,
+        896,
+        456
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6: Proposed mechanism of TRF in Drosophila skeletal muscle under obesogenic challenges.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        50,
+        115,
+        833,
+        840
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        40,
+        70,
+        884,
+        777
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        66,
+        92,
+        927,
+        777
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        50,
+        70,
+        770,
+        789
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5",
+    "footnote": [],
+    "bbox": [
+      [
+        65,
+        60,
+        702,
+        777
+      ]
+    ],
+    "page_idx": 38
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        368,
+        952,
+        703
+      ]
+    ],
+    "page_idx": 40
+  }
+]

preprint/preprint__7e982bca866a39190fd88f0488bc87bcf29e37f1fdbce365b6032e9477099024/images_list.json

@@ -0,0 +1,107 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. Outline of the experiment and identification algorithm. (a) The time evolution under a target Hamiltonian \\(h_0\\) is implemented on an part of the Google Sycamore chip (gray) using the pulse sequence depicted in the middle. (b) The expected value of canonical coordinates \\(x_m\\) and \\(p_m\\) for each qubit \\(m\\) over time is estimated from measurements using different \\(\\psi_n\\) as input states. (c) The data shown in (b) for each time \\(t_0\\) can be interpreted as a (complex-valued) matrix with entries indexed by measured and initial excited qubit, \\(m\\) and \\(n\\) . The identification algorithm proceeds in two steps: 1. From the matrix time-series, the Hamiltonian eigenfrequencies are extracted using our newly introduced algorithm coined tensorESPRIT, introduced in the SM, or an adapted version of the ESPRIT algorithm. The blue line indicates the denoised, high-resolution signal as 'seen' by the algorithm. 2. After removing the initial ramp using the data at some fixed time, the Hamiltonian eigenspaces are reconstructed using a non-convex optimization algorithm over the orthogonal group. We obtain a diagonal orthogonal estimate of the final ramp. From the extracted frequencies and reconstructed eigenspaces, we can calculate the identified Hamiltonian \\(\\hat{h}\\) that describes the measured time evolution and a tomographic estimate of the initial ramp.",
+    "footnote": [],
+    "bbox": [
+      [
+        85,
+        65,
+        914,
+        328
+      ]
+    ],
+    "page_idx": 3
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. A single Hamiltonian recovery of a 5-mode Hamiltonian and the corresponding time domain data. (a) The full experimental time-series data \\(\\langle x_{m}(t)\\rangle_{\\psi_{n}}\\) for \\(m,n = 1,\\ldots ,5\\) and the best fit of those data in terms of our model \\(\\frac{1}{2} (M\\exp (-i t h)S)_{m,n}\\) for a diagonal and orthogonal \\(M\\) and linear map \\(S\\) (solid lines). (b) The target Hamiltonian matrix \\(h_{0}\\) , the identified Hamiltonian \\(\\hat{h}\\) , and the deviation between them. The error of each diagonal entry is \\(\\pm (0.16 + 0.99)\\mathrm{MHz}\\) and of each off-diagonal entry \\(\\pm (0.12 + 0.50)\\mathrm{MHz}\\) and comprises of the statistical and the systematic error, respectively. The analog implementation error \\(\\mathcal{E}_{\\mathrm{analog}}(\\hat{h},h_{0})\\) is \\(0.73\\pm (0.07 + 0.62)\\mathrm{MHz}\\) , and \\(0.32\\pm 0.00\\mathrm{MHz}\\) for the eigenfrequencies. The analog implementation error \\(\\mathcal{E}_{\\mathrm{analog}}(\\hat{S},\\mathbb{1})\\) of the identified initial map is \\(0.61\\pm (0.00 + 0.12)\\) . (c) The real part of the initial map \\(\\hat{S}\\) and the diagonal orthogonal estimate \\(\\hat{D}_{M}\\) of the final map \\(M\\) , inferred from the data using the identified Hamiltonian \\(\\hat{h}\\) . (d) Absolute value of the time-domain deviation of the fit from the full experimental data for each time series, given by deviation \\([\\hat{h},\\hat{S},\\hat{D}_{M}](t)_{m,n}\\coloneqq \\langle a_{m}(t)\\rangle_{\\psi_{n}} - \\frac{1}{2}\\hat{D}_{M}\\exp (-i t\\hat{h})\\hat{S}\\) . The insets represent the root-mean-square deviation of the Hamiltonian fit from the experimental data per time series, averaged over the evolution time for each matrix entry \\((m,n)\\) , resulting in an entry-wise summarized quality of fit. We find a total root-mean-square deviation of the fit of 0.14. (e) Instantaneous root-mean-square deviation of the identified Hamiltonian \\(\\hat{h}\\) , initial map \\(\\hat{S}\\) and final map \\(\\hat{D}_{M}\\) and of the target Hamiltonian \\(h_{0}\\) with initial map fit \\(S_{0}\\) from the experimental data averaged over the distinct time series.",
+    "footnote": [],
+    "bbox": [
+      [
+        84,
+        60,
+        920,
+        428
+      ]
+    ],
+    "page_idx": 4
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. Comparing frequency and full identification errors. (a) In an \\(N = 6\\) subset of connected qubits, by varying \\(b\\) from 0 to 1, we implement 51 different Hamiltonians. The plot shows the Fourier transform of the time domain data. (b) The extracted eigenfrequencies (denoised peaks in panel (a)) are shown as colored dots, where the assigned color is indicative of the deviation between targeted eigenfrequencies (gray lines) and the identified ones from position of the peaks. (c) Analog implementation error \\(\\mathcal{E}_{\\mathrm{analog}}(\\hat{h},h_0)\\) of the identified Hamiltonian (dark red) compared to the implementation error \\(\\mathcal{E}_{\\mathrm{analog}}(\\mathrm{eig}(\\hat{h}),\\mathrm{eig}(h_0))\\) of the identified frequencies (golden). Colored (gray) error bars quantify the statistical (systematic) error. (d) Layout of the six qubits on the Sycamore processor and median of the entry-wise absolute-value deviation of the Hamiltonian matrix entries from their targeted values across the ensemble of 51 different values of \\(b\\in [0,1]\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        93,
+        60,
+        480,
+        675
+      ]
+    ],
+    "page_idx": 5
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4. Error map of Hamiltonian implementation across the Sycamore processor. Over the grid of 27 qubits, we randomly choose subsets of connected qubits and couplers of size \\(N = 5\\) . On each subset we implement two Hamiltonians with \\(b = 0, 0.5\\) and run the identification algorithm. Two instances are shown in panel (a). For each subset, we compute the deviation of the identified Hamiltonian and initial map from their respective target and assign it to each qubit or coupler involved. Due to overlap of subsets, each qubit or coupler has been involved in at least 5 different choices of subsets. Panels (b) and (c) show the median deviation for the Hamiltonian and initial map implementations, respectively. Panel (d) shows the mean of the sign flips in the identified (diagonal \\(\\pm 1\\) ) final map for each qubit.",
+    "footnote": [],
+    "bbox": [
+      [
+        531,
+        66,
+        904,
+        540
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5. Analog implementation error scaling and comparing different quantum processors. We measure the analog implementation error of the implemented Hamiltonians (dark red) and their eigenfrequencies (golden) as well as the deviation \\((\\textstyle \\sum_{l = 0}^{L}\\| \\mathrm{deviation}[\\hat{h},\\hat{S},\\hat{D}_{M}](t_{l})\\|_{L_{2}}^{2} / (N^{2}(L + 1)))^{1 / 2}\\) of the fit from the experimental data (dark blue) all averaged over implementations of Hamiltonians with quasi-random local potential on an increasing number of qubits on two different quantum processors—Sycamore #1 (circles) and #2 (diamonds). Each point is the mean of the respective quantity over 51 Hamiltonian implementations (21 for \\(N = 5\\) and 20 for \\(N = 14\\) on Sycamore #2). The data points at \\(N = 6\\) on Sycamore #1 summarizes Fig. 3(c). The error bars represent one standard deviation.",
+    "footnote": [],
+    "bbox": [
+      [
+        86,
+        66,
+        485,
+        273
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6. Initial ramp removal versus final ramp removal. We identify Hamiltonians of a set of 5-qubit Hamiltonians with Hofstadter butterfly potentials \\(\\mu_{q} = 20\\cos (2\\pi qb)\\) MHz for qubits \\(q = 1,\\ldots ,5\\) and flux value \\(b\\) in without regularization. (a) Deviation of the orthogonal part \\(\\hat{O}_S\\) ( \\(\\hat{O}_M\\) ) of the identified initial map \\(\\hat{S}\\) (final map \\(\\hat{M}\\) ) from the closest diagonal orthogonal matrix \\(\\hat{D}_S\\) ( \\(\\hat{D}_M\\) ). (b) Analog implementation error of the corresponding identified Hamiltonians \\(\\hat{h}_S\\) ( \\(\\hat{h}_M\\) ). (c) Total root-mean-square deviation of the time series data from the Hamiltonian fit.",
+    "footnote": [],
+    "bbox": [
+      [
+        84,
+        63,
+        485,
+        344
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Figure 7. Numerical benchmarks for larger system sizes. Recovery error of frequencies (golden) and Hamiltonians (red) from simulated time series averaged over 20 instances of Harper Hamiltonians for different system sizes. The error bars represent one standard deviation. The evolution is simulated for up to \\(0.6 \\mu s\\) and sampled at a rate of \\(250 \\mathrm{MHz}\\) . Statistical noise is simulated using \\(10^{3}\\) shots per expected value and SPAM is modeled by using randomly chosen idle qubit and coupler frequencies, linear ramping of \\(1.5 \\mathrm{GHz / s}\\) padded by \\(0.05 \\mathrm{ns}\\) . The fitting error of the time series is depicted in blue, right \\(y\\) -axis. We refer to the SM, Sec. VII A for details.",
+    "footnote": [],
+    "bbox": [
+      [
+        520,
+        61,
+        912,
+        210
+      ]
+    ],
+    "page_idx": 10
+  }
+]

preprint/preprint__7e982bca866a39190fd88f0488bc87bcf29e37f1fdbce365b6032e9477099024/preprint__7e982bca866a39190fd88f0488bc87bcf29e37f1fdbce365b6032e9477099024.mmd

@@ -0,0 +1,305 @@
+
+# Robustly learning the Hamiltonian dynamics of a superconducting quantum processor  
+
+Jens Eisert  
+
+jenseisert@gmail.com  
+
+FU Berlin https://orcid.org/0000- 0003- 3033- 1292Dominik HangleiterUniversity of Maryland College Park https://orcid.org/0000- 0002- 4766- 7967Ingo RothFreie Universität Berlin https://orcid.org/0000- 0002- 1191- 7442Jonas FuksaFU Berlin https://orcid.org/0000- 0003- 4606- 2584Pedram RoushanGoogle (United States) https://orcid.org/0000- 0003- 1917- 3879  
+
+## Article  
+
+# Keywords:  
+
+Posted Date: February 16th, 2024  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 3813225/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 November 6th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 52629- 3.
+
+<--- Page Split --->
+
+# Robustly learning the Hamiltonian dynamics of a superconducting quantum processor  
+
+Dominik Hangleiter, \(^{1,2,\ast}\) Ingo Roth, \(^{3,4,\ast}\) Jonás Fuksa, \(^{4}\) Jens Eisert, \(^{4,5}\) and Pedram Roushan \(^{6}\) \(^{1}\) Joint Center for Quantum Information and Computer Science (QuCS), University of Maryland and NIST, College Park, MD 20742, USA \(^{2}\) Joint Quantum Institute (JQI), University of Maryland and NIST, College Park, MD 20742, USA \(^{3}\) Quantum Research Center, Technology Innovation Institute (TII), Abu Dhabi \(^{4}\) Dahlem Center for Complex Quantum Systems, Freie Universität Berlin, 14195 Berlin, Germany \(^{5}\) Helmholtz-Zentrum Berlin für Materialien und Energie, 14109 Berlin, Germany \(^{6}\) Google Quantum AI, Mountain View, CA, USA  
+
+The required precision to perform quantum simulations beyond the capabilities of classical computers imposes major experimental and theoretical challenges. The key to solving these issues are highly precise ways of characterizing analog quantum simulators. Here, we robustly estimate the free Hamiltonian parameters of bosonic excitations in a superconducting- qubit analog quantum simulator from measured time- series of single- mode canonical coordinates. We achieve the required levels of precision in estimating the Hamiltonian parameters by maximally exploiting the model structure, making it robust against noise and state- preparation and measurement (SPAM) errors. Importantly, we are also able to obtain tomographic information about those SPAM errors from the same data, crucial for the experimental applicability of Hamiltonian learning in dynamical quantum- quench experiments. Our learning algorithm is highly scalable both in terms of the required amounts of data and post- processing. To achieve this, we develop a new super- resolution technique coined tensorESPRIT for frequency extraction from matrix time- series. The algorithm then combines tensorESPRIT with constrained manifold optimization for the eigenspace reconstruction with pre- and post- processing stages. For up to 14 coupled superconducting qubits on two Sycamore processors, we identify the Hamiltonian parameters—verifying the implementation on one of them up to sub- MHz precision—and construct a spatial implementation error map for a grid of 27 qubits. Our results constitute a fully characterized, highly accurate implementation of an analog dynamical quantum simulation and introduce a diagnostic toolkit for understanding, calibrating, and improving analog quantum processors.  
+
+Analog quantum simulators promise to shed light on fundamental questions of physics that have remained elusive to the standard methods of inference [1, 2]. Recently, enormous progress in controlling individual quantum degrees of freedom has been made towards making this vision a reality [3- 6]. While in digital quantum computers small errors can be corrected [7], it is intrinsically difficult to error- correct analog devices. Yet, the usefulness of analog quantum simulators as computational tools depends on the error of the implemented dynamics. Meeting this requirement hinges on devising characterization methods that not only yield a benchmark of the overall functioning of the device [e.g., 8- 10], but more importantly provide diagnostic information about the sources of errors.  
+
+Developing characterization tools for analog quantum simulators requires hardware developments as well as theoretical analysis and method development. With the advent of highly controlled quantum systems, efficient methods for identifying certain Hamiltonian parameters from dynamical data have been devised for specific classes of Hamiltonians. Key ideas are the use of Fourier analysis [11- 17] and tracking the dynamics of single excitations [18- 23]. For general Hamiltonian models, specific algebraic structures of the Hamiltonian terms can be exploited [24, 25]. Generalizing these ideas, a local Hamiltonian can be learned from a single eigenstate or its steady state [26- 31] or using quantum- quenches [32, 33], an approach dubbed 'correlation matrix method' [34]. Alterna  
+
+tively, one can apply general- purpose machine- learning methods [35- 39]. More recently, optimal theoretical guarantees have been derived for Hamiltonian learning schemes [40- 42] based on Pauli noise tomography [43, 44]. Crucially, these protocols assume perfect mid- circuit quenches, which—as we find here—can be a limiting assumption in practice.  
+
+This recent rapid theoretical development is not quite matched by concomitant experimental efforts. The effectiveness of some of these methods has been demonstrated for the estimation of a small number of coupling parameters of fixed two- and three- qubit Hamiltonians in nuclear magnetic resonance (NMR) experiments [45- 48]. While in NMR, the dominant noise process is decoherence, in tunable quantum simulators such as superconducting qubits, trapped ions or cold atoms in optical lattices, state preparation and measurement (SPAM) errors, as we also demonstrate here, play a central role. Initial steps at characterizing such errors as well as the dissipative Lindblad dynamics for up to two qubits in a superconducting qubit platform have been taken recently [49, 50]. Hamiltonian learning of thermal states has recently also been applied in many- body experiments as a means to characterize the entanglement of up to 20- qubit subsystems whose reduced states are parameterized by the so- called entanglement Hamiltonian [51- 53]. The challenge remains to develop and experimentally demonstrate the feasibility of scalable methods for a robust and precise identification of Hamiltonian dynamics of intermediate- size systems subject to both incoherent noise and systematic SPAM errors.  
+
+Here, we develop bespoke protocols to robustly and accurately identify the full Hamiltonian of a large- scale bosonic system and implement those protocols on superconducting quantum processors. Given the complexity of the learning
+
+<--- Page Split --->
+
+task, we focus on identifying the non- interacting part of a potentially interacting system. We are able to estimate the corresponding Hamiltonian parameters as well as SPAM errors pertaining to all individual components of the superconducting chip for up to 14- mode Hamiltonians tuned across a broad parameter regime, in contrast to previous experiments. Given the identified Hamiltonians, we quantify their implementation error. We demonstrate and verify that a targeted intermediate- size Hamiltonian is implemented on a large region of the superconducting processor with sub- MHz precision in a broad parameter range.  
+
+To this end, building on previous ideas for Hamiltonian identification [19, 24], we devise a simple and robust algorithm that exploits the structure of the system at hand. For the identification we make use of quadratically many experimental time- series tracking excitations via expectation values of canonical coordinates. Our structure- enforcing algorithm isolates two core tasks that need to be solved in Hamiltonian identification after suitable pre- processing of the data: frequency extraction and eigenspace reconstruction.  
+
+To solve the first task in a robust and structure- specific way, we develop a novel algorithm coined tensorESPRIT, which utilizes ideas from super- resolving, denoised Fourier analysis [54- 56] and tensor networks to extract frequencies from a matrix time- series. For the second task we use constrained manifold optimization over the orthogonal group [57]. Crucially, by explicitly exploiting all structure constraints of the identification problem, our method allows us to distinguish and obtain tomographic information about state- preparation and measurement errors. In the quench- based experiment this information renders identification and verification of the dynamics experimentally feasible in the first place. We further support our method development with numerical simulations of different noise effects and benchmark against more direct algorithmic approaches. We find that in contrast to other approaches our method is scalable to larger system sizes out of the reach of our current experimental efforts.  
+
+Our work constitutes a detailed case study that lays bare and provides solutions for the difficulties of practical Hamiltonian learning in a seemingly simple system. It thus provides a blueprint and paves the way for devising practical model- specific identification algorithms both for the interaction parameters of bosonic or fermionic systems and more complex settings.  
+
+Setup. We characterize the Hamiltonian governing analog dynamics of Google Sycamore chips which consist of a two- dimensional array of nearest- neighbour coupled superconducting qubits. Each physical qubit is a non- linear oscillator with bosonic excitations (microwave photons) [58]. Using the rotating- wave approximation the dynamics governing the excitations of the qubits in the rotating frame can be well described by the Bose- Hubbard Hamiltonian [59]  
+
+\[H_{\mathrm{BH}} = \sum_{i}\left(\mu_{i}a_{i}^{\dagger}a_{i} + \eta_{i}a_{i}^{\dagger}a_{i}^{\dagger}a_{i}a_{i}\right) - \sum_{i\neq j}J_{i,j}a_{i}^{\dagger}a_{j}, \quad (1)\]  
+
+where \(a_{i}^{\dagger}\) and \(a_{i}\) denote bosonic creation and annihilation operators at site \(i\) , respectively, \(\mu \in \mathbb{R}^{N}\) are the on- site poten  
+
+tials, \(J\in \mathbb{R}^{N\times N}\) are the hopping rates between nearest neighbour qubits, and \(\eta \in \mathbb{R}^{N}\) are the strength of on- site interactions. The qubit frequency, the nearest- neighbour coupling between them, and the non- linearity (anharmonicity) set \(\mu ,J\) and \(\eta\) . We are able to tune \(\mu\) and \(J\) on nanosecond timescales, while \(\eta\) is fixed.  
+
+Here, we focus on the specific task of identifying the values of \(\mu_{i}\) and \(J_{i,j}\) . The corresponding non- interacting part of the Hamiltonian acting on \(N\) modes can be conveniently parametrized as  
+
+\[H(h) = -\sum_{i,j = 1}^{N}h_{i,j}a_{i}^{\dagger}a_{j} \quad (2)\]  
+
+with an \(N\times N\) real symmetric parameter matrix \(h\) with entries \(h_{i,j}\) , which is composed of the on- site chemical potentials \(\mu_{i}\) on its diagonal and the hopping energies \(J_{i,j}\) for \(i\neq j\) . The identification of the non- interacting part \(H(h)\) of \(H_{\mathrm{BH}}\) can be viewed as a first step in a hierarchical procedure for characterizing dynamical quantum simulations with tunable interactions and numbers of particles.  
+
+The non- interacting part \(H(h)\) of the Hamiltonian \(H_{\mathrm{BH}}\) can be inferred when initially preparing a state where only a single qubit is excited with a single photon. For initial states with a single excitation, the interaction term vanishes, hence effectively \(\eta = 0\) . Consequently, only the two lowest energy levels of the non- linear oscillators enter the dynamics. Therefore, referring to them as qubits (two- level systems) is precise. Specifically, we identify the parameters \(h_{i,j}\) from dynamical data of the following form. We initialize the system in \(|\psi_{n}\rangle := (\mathbb{1} + a_{n}^{\dagger})|0\rangle^{\otimes N} / \sqrt{2}\) and measure the canonical coordinates \(x_{m} = (a_{m} + a_{m}^{\dagger}) / 2\) and \(p_{m} = (a_{m} - a_{m}^{\dagger}) / (2\mathrm{i})\) for all combinations of \(m,n = 1,\ldots ,N\) . In terms of the qubit architecture, this amounts to local Pauli- \(X\) and Pauli- \(Y\) basis measurements, respectively. We combine the statistical averages over multiple measurements to obtain an empirical estimator for \(\langle a_{m}(t)\rangle_{\psi_{n}} = \langle x_{m}(t)\rangle_{\psi_{n}} + \mathrm{i}\langle p_{m}(t)\rangle_{\psi_{n}}\) . For particle- number preserving dynamics, this data is of the form  
+
+\[\langle a_{m}(t)\rangle_{\psi_{n}} = \frac{1}{2}\exp (-i t h)_{m,n}. \quad (3)\]  
+
+It therefore directly provides estimates of the entries of the time- evolution unitary at time \(t\) in the single- particle sector of the bosonic Fock space.  
+
+In Fig. 1, we show an overview of the experimental procedure, and the different steps of the Hamiltonian identification algorithm. Every experiment uses a few coupled qubits, from the larger array of qubits on the device (Fig. 1(a)). On those qubits, the goal is to implement the time- evolution with targeted Hamiltonian parameters \(h_{0}\) , which are subject to connectivity constraints imposed by the couplings of the qubits. To achieve this, we perform the following pulse sequence to collect dynamical data of the form (3). Before the start of the sequence, the qubits are at frequencies (of the \(|0\rangle\) to \(|1\rangle\) transition) that could be a few hundred MHz apart from each other. In the beginning, all qubits are in their ground state \(|0\rangle\) . To prepare the initial state, a \(\pi /2\) - pulse is applied to one
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1. Outline of the experiment and identification algorithm. (a) The time evolution under a target Hamiltonian \(h_0\) is implemented on an part of the Google Sycamore chip (gray) using the pulse sequence depicted in the middle. (b) The expected value of canonical coordinates \(x_m\) and \(p_m\) for each qubit \(m\) over time is estimated from measurements using different \(\psi_n\) as input states. (c) The data shown in (b) for each time \(t_0\) can be interpreted as a (complex-valued) matrix with entries indexed by measured and initial excited qubit, \(m\) and \(n\) . The identification algorithm proceeds in two steps: 1. From the matrix time-series, the Hamiltonian eigenfrequencies are extracted using our newly introduced algorithm coined tensorESPRIT, introduced in the SM, or an adapted version of the ESPRIT algorithm. The blue line indicates the denoised, high-resolution signal as 'seen' by the algorithm. 2. After removing the initial ramp using the data at some fixed time, the Hamiltonian eigenspaces are reconstructed using a non-convex optimization algorithm over the orthogonal group. We obtain a diagonal orthogonal estimate of the final ramp. From the extracted frequencies and reconstructed eigenspaces, we can calculate the identified Hamiltonian \(\hat{h}\) that describes the measured time evolution and a tomographic estimate of the initial ramp. </center>  
+
+of the qubits, resulting in its Bloch vector moving to the equator. Then ramping pulses are applied to all qubits to bring them to the desired detuning around a common rendezvous frequency (6500 MHz in this work). At the same time, pulses are applied to the couplers to set the nearest- neighbour hopping to the desired value (20 MHz in this work). The pulses are held at the target values for time \(t\) , corresponding to the evolution time of the experiment. Subsequently, the couplers are ramped back to zero coupling and the qubits back to their initial frequency, where \(\langle x_m(t) \rangle\) and \(\langle p_m(t) \rangle\) on the desired qubit \(m\) is measured. The initial and final pulse ramping take place over a finite time of 2–3 ns, and therefore give rise to a non- trivial effect on the dynamics, which we take into account in the identification procedure. In fact, we find that the effects of the ramping phase are the dominant source of SPAM errors in the quench- based analog simulation. The experimental data (Fig. 1(b)) on \(N\) qubits are \(N \times N\) time- series estimates of \(\langle a_m(t) \rangle_{\psi_n}\) for \(t = 0, 1, \ldots , T\) ns and all pairs \(n, m = 1, \ldots , N\) . Given those data, the identification task amounts to identifying the 'best' coefficient matrix \(h\) , describing the time- sequence of snapshots of the single- particle unitary matrix \(\frac{1}{2} \exp (- \mathrm{i} t h)\) .  
+
+Identification method. We can identify the generator \(h\) of the unitary in two steps (Fig. 1(c)), making use of the eigen- decomposition of the Hamiltonian (see Methods). In the first step, the time- dependent part of the identification problem is solved, namely, identifying the Hamiltonian eigenvalues (eigenfrequencies). In the second step, given the eigenvalues, the eigenbasis for the Hamiltonian of \(h\) is determined. In order to make the identification method noise- robust, we furthermore exploit structural constraints of the model. First, the Hamiltonian has a spectrum such that the time- series data has a time- independent, sparse frequency spectrum with exactly \(N\) contributions. Second, the Fourier coefficients of the data have an explicit form as the outer product of the orthogonal eigenvectors of the Hamiltonian. Third, the Hamiltonian parameter matrix is real and has an a priori known sparse support due to the experimental connectivity constraints. These structural constraints are not respected by various sources of incoherent noise, including particle loss and finite shot noise, and coherent noise, in particular the SPAM error. Thereby, an identification protocol that takes these constraints into account is intrinsically robust against various imperfections.   
+
+To robustly identify the sparse frequencies from the experimental data, we develop a new super- resolution and denoising algorithm tensorESPRIT that is applicable to matrix- valued time series and uses tensor network techniques in conjunction with super- resolution techniques for scalar data [55]. Achieving high precision in this step is crucial for identifying the eigenvectors in the presence of noise. To robustly identify the eigenbasis, in the second step, we perform least- square optimization of the time- series data under the orthonormality constraint with a gradient descent algorithm on the manifold structure of the orthogonal group [57]. Here, we incorporate
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2. A single Hamiltonian recovery of a 5-mode Hamiltonian and the corresponding time domain data. (a) The full experimental time-series data \(\langle x_{m}(t)\rangle_{\psi_{n}}\) for \(m,n = 1,\ldots ,5\) and the best fit of those data in terms of our model \(\frac{1}{2} (M\exp (-i t h)S)_{m,n}\) for a diagonal and orthogonal \(M\) and linear map \(S\) (solid lines). (b) The target Hamiltonian matrix \(h_{0}\) , the identified Hamiltonian \(\hat{h}\) , and the deviation between them. The error of each diagonal entry is \(\pm (0.16 + 0.99)\mathrm{MHz}\) and of each off-diagonal entry \(\pm (0.12 + 0.50)\mathrm{MHz}\) and comprises of the statistical and the systematic error, respectively. The analog implementation error \(\mathcal{E}_{\mathrm{analog}}(\hat{h},h_{0})\) is \(0.73\pm (0.07 + 0.62)\mathrm{MHz}\) , and \(0.32\pm 0.00\mathrm{MHz}\) for the eigenfrequencies. The analog implementation error \(\mathcal{E}_{\mathrm{analog}}(\hat{S},\mathbb{1})\) of the identified initial map is \(0.61\pm (0.00 + 0.12)\) . (c) The real part of the initial map \(\hat{S}\) and the diagonal orthogonal estimate \(\hat{D}_{M}\) of the final map \(M\) , inferred from the data using the identified Hamiltonian \(\hat{h}\) . (d) Absolute value of the time-domain deviation of the fit from the full experimental data for each time series, given by deviation \([\hat{h},\hat{S},\hat{D}_{M}](t)_{m,n}:= \langle a_{m}(t)\rangle_{\psi_{n}} - \frac{1}{2}\hat{D}_{M}\exp (-i t\hat{h})\hat{S}\) . The insets represent the root-mean-square deviation of the Hamiltonian fit from the experimental data per time series, averaged over the evolution time for each matrix entry \((m,n)\) , resulting in an entry-wise summarized quality of fit. We find a total root-mean-square deviation of the fit of 0.14. (e) Instantaneous root-mean-square deviation of the identified Hamiltonian \(\hat{h}\) , initial map \(\hat{S}\) and final map \(\hat{D}_{M}\) and of the target Hamiltonian \(h_{0}\) with initial map fit \(S_{0}\) from the experimental data averaged over the distinct time series. </center>  
+
+the connectivity constraint on the coefficient matrix \(h\) by making use of regularization techniques [60].  
+
+Robustness against ramp errors. The initial and final ramping pulses result in a time- independent, linear transformation at the beginning and end of the time series. It is important to stress that such ramping pulses are expected to be generic in a wide range of experimental implementations of dynamical analog quantum simulations. Robustness of an Hamiltonian identification method against these imperfections is essential for accurate estimates in practice. We can model the effect of such particle number preserving state preparation and measurement (SPAM) errors via linear maps \(S\) and \(M\) , respectively, see the SM for details. This alters our model of the ideal data (3) to  
+
+\[\langle a_{m}(t)\rangle_{\psi_{n}} = \frac{1}{2} (M\cdot \exp (-i t h)\cdot S)_{m,n}. \quad (4)\]  
+
+While for the frequency identification such time  
+
+independent errors 'only' deteriorate the signal- to- noise ratio, for the identification of the eigenvectors of \(h\) it is crucial to take the effects of non- trivial \(S\) and \(M\) into account. Given the details of the ramping procedure, we expect that the deviation of the initial map \(S\) from the identity will be significantly larger than that of the final map \(M\) and provide evidence for this in the Methods. In particular, the final map will be dominated by phase accumulation on the diagonal.  
+
+By pre- processing the data, we can robustly remove an arbitrary initial map \(S\) . By post- processing, we can obtain an orthogonal diagonal estimate \(\hat{D}_{M}\) of the final map \(M\) . We give numerical evidence that the estimate \(\hat{D}_{M}\) gives good results in the particular experimental setting. From the identified Hamiltonian and an orthogonal diagonal estimate \(\hat{D}_{M}\) of \(M\) , we get an estimate \(\hat{S}\) of \(S\) .  
+
+Error sources. There are two main remaining sources of error that affect the Hamiltonian identification. First, the esti
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3. Comparing frequency and full identification errors. (a) In an \(N = 6\) subset of connected qubits, by varying \(b\) from 0 to 1, we implement 51 different Hamiltonians. The plot shows the Fourier transform of the time domain data. (b) The extracted eigenfrequencies (denoised peaks in panel (a)) are shown as colored dots, where the assigned color is indicative of the deviation between targeted eigenfrequencies (gray lines) and the identified ones from position of the peaks. (c) Analog implementation error \(\mathcal{E}_{\mathrm{analog}}(\hat{h},h_0)\) of the identified Hamiltonian (dark red) compared to the implementation error \(\mathcal{E}_{\mathrm{analog}}(\mathrm{eig}(\hat{h}),\mathrm{eig}(h_0))\) of the identified frequencies (golden). Colored (gray) error bars quantify the statistical (systematic) error. (d) Layout of the six qubits on the Sycamore processor and median of the entry-wise absolute-value deviation of the Hamiltonian matrix entries from their targeted values across the ensemble of 51 different values of \(b\in [0,1]\) . </center>  
+
+mate \(\hat{h}\) has a statistical error due to the finite number of measurements used to estimate the expectation values. Second, any non- trivial final map \(M\) will produce a systematic error in the eigenbasis reconstruction and the tomographic estimate \(\hat{S}\) . We partially remedy this effect with an orthogonal diagonal estimate \(\hat{D}_M\) of \(M\) .  
+
+Results. We implement and characterize different Hamiltonians from time- series data on two distinct quantum Sycamore processors—Sycamore #1 and #2. The Hamiltonians we implement have a fixed overall hopping strength \(J_{i,j} = 20 \mathrm{MHz}\) and site- dependent local potentials \(\mu_i\) on subsets of qubits. Specifically, we choose the local potentials quasi- randomly \(\mu_q = 20 \cos (2\pi qb) \mathrm{MHz}\) , for \(q = 1, \ldots , N\) , where \(b\) is a number between zero and one. In one dimension, this choice corresponds to implementing the Harper Hamiltonian, which exhibits characteristic 'Hofstadter butterfly' frequency spectra as a function of the dimensionless magnetic flux \(b\) [61].  
+
+We measure deviations in the identification in terms of the analog implementation error of the identified Hamiltonian \(\hat{h}\) with respect to the targeted Hamiltonian \(h_0\) as  
+
+\[\mathcal{E}_{\mathrm{analog}}(\hat{h},h_0):= \frac{1}{N}\left\| \hat{h} -h_0\right\|_{\ell_2}, \quad (5)\]  
+
+defined in terms of the \(\ell_2\) - norm, which for a matrix \(A\) is given by \(\| A\|_{\ell_2} = (\sum_{i,j}|A_{i,j}|^2)^{1 / 2}\) . We also use the analog implementation error to quantify the implementation error of the initial map \(\hat{S}\) as \(\mathcal{E}_{\mathrm{analog}}(\hat{S}, \mathbb{1})\) , and of the eigenfrequencies \(\mathrm{eig}(\hat{h})\) as \(\mathcal{E}_{\mathrm{analog}}(\mathrm{eig}(\hat{h}), \mathrm{eig}(h_0))\) . Notice that the analog implementation error of the frequencies in the data from the targeted Hamiltonian eigenfrequencies give a lower bound to the overall implementation error of the identified Hamiltonian. This is because the \(\ell_2\) - norm used in the definition (5) of \(\mathcal{E}_{\mathrm{analog}}\) is unitarily invariant and any deviation in the eigenbasis, which we identify in the second step of our algorithm, will tend to add up with the frequency deviation.  
+
+In Fig. 2, we illustrate the properties of a single Hamiltonian identification instance in terms of both how well the simulated time evolution fits the experimental data (a,d,e) and how it compares to the targeted Hamiltonian (b) and SPAM (c). We find that most entries of the identified Hamiltonian deviate from the target Hamiltonian by less than \(0.5 \mathrm{MHz}\) with a few entries deviating by around \(1 - 2 \mathrm{MHz}\) . The overall implementation error is around \(1 \mathrm{MHz}\) . The error of the identification method is dominated by the systematic error due to the final ramping phase that is around \(1 \mathrm{MHz}\) for the individual entries, see the SM for details. Small long- range couplings exceeding the statistical error are necessary to fit the data well even when penalizing those entries via regularization. These entries are rooted in the effective rotation by the final ramping before the measurement and within the estimated systematic error.  
+
+The fit deviation from the data (Fig. 2(e)) exhibits a prominent decrease within the first few nanoseconds of the time evolution. This indicates that the time evolution differs during the initial phase of the experiment as compared to the main phase of the experiment, which we can attribute to the initial pulse
+
+<--- Page Split --->
+
+ramping of the experiment. The identified initial map describing this ramping (Fig. 2(c)) is approximately band- diagonal and deviates from being unitary, indicating fluctuations of the effective ramps between different experiments.  
+
+We find a larger time- averaged real- time error (Fig. 2(d)) in all data series \(\langle a_{m}\rangle_{\psi_{n}}\) in which \(Q_{4}\) was measured, indicating a measurement error on \(Q_{4}\) . We also observe a deviation between the parameters of the target and identified Hamiltonian in qubits \(Q_{3}\) and \(Q_{4}\) and the coupler between them. Since the deviation of the eigenfrequencies is much smaller than of the full Hamiltonian, we attribute those errors also to a nontrivial final ramping phase at those qubits that leads to a rotated eigenbasis.  
+
+In Fig. 3, we summarize multiple identification data of this type to benchmark the overall performance of a fixed set of qubits. In panel (a), we show the measured Fourier domain data for 51 different values of the magnetic flux \(b \in [0, 1]\) . In panel (b), we plot the deviation of the frequencies identified from the data. Most implemented frequencies deviate by less than \(1 \mathrm{MHz}\) from their targets. Importantly, the frequency identification is robust against systematic measurement errors. When comparing the analog implementation errors of the full Hamiltonian (Fig. 3(c)) to the corresponding frequency errors, we find an up to fourfold increase in implementation error. The Hamiltonian implementation error is affected by a systematic error due to the non- trivial final ramp. We estimate this error using a linear ramping model; see the SM for details. Since the deviation lies outside of the combined systematic and statistical error bars, our results indicate that the targeted Hamiltonian has not been implemented exactly.  
+
+In Fig. 3(d), we show the median of the entry- wise deviation of the identified Hamiltonian from its target over all magnetic flux values. Thereby, the ensemble of Hamiltonians defines an overall error benchmark. This benchmark can be associated to the individual constituents of the quantum processor, namely, the qubits, corresponding to diagonal entries of the Hamiltonian deviation, and the couplers, corresponding to the first off- diagonal matrix entries of the deviation.  
+
+We use this benchmark over an ensemble of two flux values to assess a 27- qubit array of superconducting qubits. To do so, we repeat the analysis reported in Fig. 3 for 5- qubit dynamics on different subsets of qubits and extract average errors of the individual qubits and couplers involved in the dynamics, both in terms of the identified Hamiltonian and the initial and final maps. Summarized in Fig. 4, we find significant variation in the implementation error of different couplers and qubits. While for some qubits the effects of the initial and final maps are negligible, for others they indicate the potential of a significant implementation error. From a practical point of view, such diagnostic data allows to maximally exploit the chip's error for small- scale analog simulation experiments. Let us note that within the error of our method the overall benchmark for the qubits and couplers for 5- qubit dynamics agrees with that of 3- and 4- qubit dynamics.  
+
+All of the Hamiltonian identification experiments discussed so far (Figs. 2, 3, 4) were implemented on the Sycamore #1 chip. In order to compare these results to implementations on a physically distinct chip with different calibration,  
+
+![](images/Figure_4.jpg)
+
+<center>Figure 4. Error map of Hamiltonian implementation across the Sycamore processor. Over the grid of 27 qubits, we randomly choose subsets of connected qubits and couplers of size \(N = 5\) . On each subset we implement two Hamiltonians with \(b = 0, 0.5\) and run the identification algorithm. Two instances are shown in panel (a). For each subset, we compute the deviation of the identified Hamiltonian and initial map from their respective target and assign it to each qubit or coupler involved. Due to overlap of subsets, each qubit or coupler has been involved in at least 5 different choices of subsets. Panels (b) and (c) show the median deviation for the Hamiltonian and initial map implementations, respectively. Panel (d) shows the mean of the sign flips in the identified (diagonal \(\pm 1\) ) final map for each qubit. </center>  
+
+and to demonstrate the scalability of our method, we implement Hamiltonian identification experiments for an increasing number of qubits on the Sycamore #2 chip. More precisely, for a given number of qubits \(N\) , we implement many different Hamiltonians with quasi- random local potentials, as shown in Fig. 3(c) for \(N = 6\) . We then average the analog implementation errors of the Hamiltonians and frequencies for several system sizes. The results are shown in Fig. 5. Notably, comparing the two different processors, the overall quality of fit does not depend significantly on either the number of qubits or the processor used. This indicates, first, that
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Figure 5. Analog implementation error scaling and comparing different quantum processors. We measure the analog implementation error of the implemented Hamiltonians (dark red) and their eigenfrequencies (golden) as well as the deviation \((\textstyle \sum_{l = 0}^{L}\| \mathrm{deviation}[\hat{h},\hat{S},\hat{D}_{M}](t_{l})\|_{L_{2}}^{2} / (N^{2}(L + 1)))^{1 / 2}\) of the fit from the experimental data (dark blue) all averaged over implementations of Hamiltonians with quasi-random local potential on an increasing number of qubits on two different quantum processors—Sycamore #1 (circles) and #2 (diamonds). Each point is the mean of the respective quantity over 51 Hamiltonian implementations (21 for \(N = 5\) and 20 for \(N = 14\) on Sycamore #2). The data points at \(N = 6\) on Sycamore #1 summarizes Fig. 3(c). The error bars represent one standard deviation. </center>  
+
+our reconstruction method works equally well in all scenarios and, second, that both quantum processors implement Hamiltonian time evolution that closely fits our model assumption. We also notice that the overall analog implementation error does not significantly depend on the system size. This signifies that no additional non- local errors are introduced into the system as the size is increased. At the same time, the overall error of Hamiltonian implementations on Sycamore #2 is much worse compared to those on Sycamore #1, indicating that Sycamore #2 was not as well calibrated. Hamiltonian identification thus allows us to meaningfully compare Hamiltonian implementations across different physical systems and system sizes.  
+
+Conclusion. We have implemented analog simulation of the time- evolution of non- interacting bosonic Hamiltonians with tunable parameters for up to 14 qubit lattice sites. A structure- exploiting learning method allows us to robustly identify the implemented Hamiltonian that governs the time- evolution. To achieve this, we have introduced a new super- resolution algorithm, referred to as tensorESPRIT, for precise robust identification of eigenfrequencies of a Hermitian matrix from noisy snapshots of the one parameter unitary subgroup it generates. Thereby, we diagnose the deviation from the target Hamiltonian and assess the precision of the implementation. We achieve sub- MHz error of the Hamiltonian parameters compared to their targeted values in most implementations. Combining the average performance measures over ensembles of Hamiltonians we associate benchmarks to the components of the superconducting qubit chips that quantify the performance of the hardware on the time evolution and provide specific diagnostic information. Within our Hamiltonian identification framework, we are able to identify SPAM errors due to parameter ramp phases as a severe limitation of the architecture. Importantly, such ramp phases are present in any analog quantum simulation of quenched dynamics. Our results show that minimizing those is crucial for precisely implementing a Hamiltonian.   
+
+The experimental and computational effort of the identification method scales efficiently in the number of modes of the Hamiltonian. We have also numerically identified the limitations of more direct algorithmic approaches and demonstrated the scalability of our method under empirically derived noise and error models.  
+
+Generalizing our two- step approach developed here, we expect a polynomial scaling with the dimension of the diagnosed particle sector and therefore remain efficient for diagnosing two- , three- and four- body interactions, thus allowing to build trust in the correct implementation of interacting Hamiltonian dynamics as a whole. From a broader perspective, with this work, we hope to contribute to the development of a machinery for precisely characterizing and thereby improving analog quantum devices.  
+
+## METHODS  
+
+### A. Experimental details  
+
+Details on the quantum processor. We use the Sycamore quantum processor composed of quantum systems arranged in a two- dimensional array. This processor consists of gmon qubits (transmons with tunable coupling) with frequencies ranging from 5 to 7 GHz. These frequencies are chosen to mitigate a variety of error mechanisms such as two- level defects. Our coupler design allows us to quickly tune the qubit- qubit coupling from 0 to \(40+\) MHz. The chip is connected to a superconducting circuit board and cooled down to below \(20\mathrm{mK}\) in a dilution refrigerator. Each qubit has a microwave control line used to drive an excitation and a flux control line to tune the frequency. The processor is connected through filters to room- temperature electronics that synthesize the control signals. We execute single- qubit gates by driving \(25\mathrm{ns}\) microwave pulses resonant with the qubit transition frequency.  
+
+Experimental read- out and control. The qubits are connected to a resonator that is used to read out the state of the qubit. The state of all qubits can be read simultaneously by using a frequency- multiplexing. Initial device calibration is performed using 'Optimus' [62] where calibration experiments are represented as nodes in a graph.
+
+<--- Page Split --->
+
+### B. Details of the identification algorithm  
+
+Succinctly written, our data model is given by  
+
+\[y_{m,n}[l]:= \langle a_m(t_l)\rangle_{\psi_n} = \frac{1}{2} (M\cdot \exp (-it_lh)\cdot S)_{m,n}, \quad (6)\]  
+
+where \(m,n = 1,\ldots ,N\) label the distinct time series, \(l =\) \(0,\ldots ,L\) labels the time stamps of the \(L + 1\) data points per time series. The matrices \(S\) and \(M\) are arbitrary invertible linear maps that capture the state preparation and measurement stage, as affected by the ramping of the eigenfrequencies of the qubits and couplers to their target value and back (see Fig. 1). In the experiment, we empirically estimate each such expectation value with 1000 single shots.  
+
+Our mindset for solving the identification problem is based on the eigendecomposition \(\begin{array}{r}h = \sum_{k = 1}^{N}\lambda_{k} |v_{k}\rangle \langle v_{k}| \end{array}\) of the coefficient matrix \(h\) in terms of eigenvectors \(|v_{k}\rangle\) and eigenvalues \(\lambda_{k}\) . We can write the data (6) in matrix form as  
+
+\[y[l] = \frac{1}{2}\exp (-it_lh) = \frac{1}{2}\sum_{k = 1}^{N}\mathrm{e}^{-it_l\lambda_k}|v_k\rangle \langle v_k|, \quad (7)\]  
+
+where we have dropped \(S\) and \(M\) for the time being. This decomposition suggests a simple procedure to identify the Hamiltonian using Fourier data analysis. From the matrixvalued time series data \(y[l]\) (7), we identify the Hamiltonian coefficient matrix \(h\) in two steps. First, we determine the eigenfrequencies of \(h\) . Second, we identify the eigenbasis of \(h\) . To achieve those identification tasks with the largest possible robustness to error, it is key to exploit all available structure at hand.  
+
+Step 1: Frequency extraction. In order to robustly estimate the spectrum, we exploit that the signal is sparse in Fourier space. This structure allows us to substantially denoise the signal and achieve super- resolution beyond the Nyquist limit [63, 64]. A candidate algorithm for this task, suitable for scalar time- series, is the ESPRIT algorithm, which comes with rigorous recovery guarantees [55, 56]. To extract the Hamiltonian spectrum from the matrix time- series \(y[l]\) , we apply ESPRIT to the trace of the data series (for \(S = M = \mathbb{1}\) )  
+
+\[F[l]:= \mathrm{Tr}[y[l]] = \sum_{m = 1}^{N}y_{m,m}[l] = \frac{1}{2}\sum_{k = 1}^{N}\mathrm{e}^{-it_l\lambda_k}. \quad (8)\]  
+
+The drawback of this approach is that if the spectrum of the Hamiltonian is sufficiently crowded, which will happen for large \(N\) , the Fourier modes in \(F[l]\) become indistinguishable and ESPRIT fails to identify the frequencies. In particular, ESPRIT is not able to identify degeneracies in the spectrum.  
+
+To overcome this issue and obtain a truly scalable learning procedure applicable to degenerate spectra, we develop a new algorithm coined tensorESPRIT, which extends the ideas of ESPRIT to the case of a matrix time- series using tensor network techniques. Using tensorESPRIT also improves the robustness of frequency estimation to SPAM errors. For practical Hamiltonians, tensorESPRIT becomes necessary for systems with \(N \gtrsim 12\) .  
+
+tensorESPRIT (ESPRIT) comprises of a denoising step, in which the rank of the Hankel tensor (matrix) of the data is limited to its theoretical value. Subsequently, rotational invariance of the data is used to compute a matrix from the denoised Hankel tensor (matrix), the spectrum of which has a simple relation to the spectrum of \(h\) . In the case of ESPRIT, this amounts to a multiplication of the denoised Hankel matrix by a pseudoinverse of its shifted version. Contrastingly, tensorESPRIT uses a sampling procedure to contract certain sub- matrices of the denoised Hankel tensor with the pseudoinverse of other sub- matrices. Details on both algorithms can be found in the SM.  
+
+Step 2: Eigenspace identification. To identify the eigenspaces of the Hamiltonian, we use the eigenfrequencies found in Step 1 to fix the oscillating part of the dynamics in Eq. (7). What remains is the problem of finding the eigenspaces \(|v_k\rangle \langle v_k|\) from the data. This problem is a nonconvex inverse quadratic problem, subject to orthogonality of the eigenspaces, as well as the constraint that the resulting Hamiltonian matrix respects the connectivity of the superconducting architecture. Formally, we denote the a priori known support set of the Hamiltonian matrix as \(\Omega\) , so that we can write the support constraint as \(h_{\Omega} = 0\) , where \(\bar{\Omega}\) denotes the complement of \(\Omega\) and subscripting a matrix with a support set restricts the matrix to this set. We can cast this problem into the form of a least- squares optimization problem  
+
+\[\underset {\{ |v_k\rangle \}}{\mathrm{minimise}}\quad \sum_{l = 0}^{L}\left\| y[l] - \sum_{k}\mathrm{e}^{-\mathrm{i}\lambda_k t_l}|v_k\rangle \langle v_k|\right\|_{\ell_2}^2,\] \[\mathrm{subject~to}\quad \langle v_m|v_n\rangle = \delta_{m,n},\left(\sum_k\lambda_k|v_k\rangle \langle v_k|\right)_{\bar{\Omega}} = 0,\]  
+
+equipped with non- convex constraints enforcing orthogonality, and the quadratic constraint restricting the support. In order to approximately enforce the support constraint, we make use of regularization [60]. It turns out that this can be best achieved by adding a term [65, App. A]  
+
+\[\mu \left\| \left(\sum_k\lambda_k|v_k\rangle \langle v_k|\right)_{\bar{\Omega}}\right\|_{\ell_2} \quad (10)\]  
+
+to the objective function (9), where \(\mu > 0\) is a parameter weighting the violation of the support constraint. We then solve the resulting minimization problem by using a conjugate gradient descent on the manifold of the orthogonal group [57, 66], see also the recent work [67- 69] for the use of geometric optimization for quantum characterization.  
+
+Without the support constraint this gives rise to an optimization algorithm that converges well, as shown in the SM. However, the regularization term makes the optimization landscape rugged as it introduces an entry- wise constraint that is skew to the orthogonal manifold. To deal with this, we consecutively ramp up \(\mu\) until the algorithm does not converge anymore in order to find the Hamiltonian that best approximates the support constraint while being a proper solution of
+
+<--- Page Split --->
+
+the optimization problem. For example, for the data in Fig. 2 the value of \(\mu\) is 121. In order to avoid that we identify a Hamiltonian from a local minimum of the rugged landscape, we only accept Hamiltonians that achieve a total fit of the experimental data within a \(5\%\) margin of the fit quality of the unregularized recovery problem, and use the Hamiltonian recovered without the regularization otherwise.  
+
+### C. Robustness to state preparation and measurement errors  
+
+The experimental design requires a ramping phase of the qubit and coupler frequencies from their idle location to the desired target Hamiltonian and back for the measurement. In effect, the data model (6) includes time- independent linear maps \(M\) and \(S\) that are applied at the beginning and end of the Hamiltonian time- evolution. The maps affect both the frequency extraction and the eigenspace reconstruction.  
+
+For the frequency extraction using ESPRIT, the Fourier coefficients of the trace signal \(F[l]\) become \(\langle v_k|SM|v_k\rangle\) . While the frequencies remain unchanged the Fourier coefficients now deviate from unity, significantly impairing the noise- robustness of the frequency identification. This effect is still present, albeit weaker, in tensorESPRIT, in the case of non- unitary SPAM errors. The eigenspace reconstruction is affected much more severely and requires careful consideration, as detailed below and in the SM.  
+
+Ramp removal via pre- processing. We can remove either the initial map \(S\) or the final map \(M\) from the data. To remove \(S\) , we apply the pseudoinverse \((\cdot)^{+}\) of the data \(y[l_0]\) at a fixed time \(t_{l_0}\) to the entire (time- dependent) data series in matrix form. For invertible \(S\) and \(M\) this gives rise to  
+
+\[\begin{array}{l}{y^{(l_0)}[l] = y[l](y[l_0])^+}\\ {= \sum_{k = 1}^{N}\mathrm{e}^{-\mathrm{i}\lambda_k(t_l - t_{l_0})}M|v_k\rangle \langle v_k|M^{-1}.} \end{array} \quad (11)\]  
+
+The caveat of this approach is that the noise that affected \(y[l_0]\) is now present in every entry of the new data series \(y^{(l_0)}\) . We can remedy this effect by using several time points \(y[l_0]\) for the inversion. To this end, we compute the concatenation of data series for different choices of \(l_0\) , e.g., for every \(s\) data points \(0, s, 2s, \ldots , \lfloor L / s \rfloor s\) giving rise to new data \(y_{\mathrm{total}, s} = (y^{(0)}, y^{(s)}, y^{(2s)}, \ldots , y^{( \lfloor L / s \rfloor s)}) \in \mathbb{C}^{\lfloor L / s \rfloor L}\) . If the data suffers from drift errors, it is also beneficial to restrict each data series \(y^{(l_0)}\) to entries \(y^{(l_0)}[\kappa ]\) with \(\kappa \in [l_0 - w, l_0 + w]\) , i.e., the entries in a window of size \(w\) around \(l_0\) . In practice, we use \(s = 1\) and \(w = 50\) for the reconstructions on Sycamore #1, and \(s = 1, w = L\) for those on Sycamore #2.  
+
+As we argue below, the final map \(M\) is nearly diagonal here. Hence, we can use \(y_{\mathrm{total}, s}\) from Eq. (9) and it is justified to apply the support constraint in the eigenspace reconstruction step. However, the eigenspace reconstruction will suffer from systematic errors due to the final map, even in the case when it is nearly diagonal. Below, we explain a method to partially remove this error.  
+
+Tomographic estimate of \(S\) and \(M\) . The systematic error in the reconstructed Hamiltonian eigenbasis can be expressed as an orthogonal rotation \(D_M\) from the eigenbasis that is actually implemented. Due to the gauge freedom in the model (6), we cannot hope to identify \(D_M\) fully without additional assumptions. However, as elaborated on in the SM, we can find a diagonal orthogonal estimate \(\hat{D}_M\) of the true correction \(D_M\) and hence remove a sign of the systematic error. To this end, we assume that the experimental implementation of the target Hamiltonian does not flip the sign in the hopping terms and remedy the sign of systematic error due to the final map by fitting a diagonal orthogonal rotation of the Hamiltonian eigenbasis \(\hat{D}_M\) that minimizes the implementation error. We update the reconstructed Hamiltonian to  
+
+\[\hat{h} = \hat{D}_M\hat{h}\hat{D}_M, \quad (12)\]  
+
+where \(\begin{array}{r}\tilde{h} = \sum_k\lambda_k|v_k\rangle \langle v_k| \end{array}\) and \(\{\vert v_k\rangle \}\) is the eigenbasis obtained by solving the problem (9), and use \(\hat{D}_M\) as an estimate of \(M\) . We can now obtain a tomographic estimate of the initial map through  
+
+\[\hat{S}:= \frac{2}{L + 1}\sum_{l = 0}^{L}\exp [\mathrm{i}t_l\hat{h}]\hat{D}_M y[l]. \quad (13)\]  
+
+The recovered model \((\hat{h},\hat{S},\hat{D}_M)\) gives good prediction accuracy on simulated data, as demonstrated in Fig. 7 and in the SM, and fits well the experimental data, as demonstrated in Figs. 2, 5 and 6.  
+
+Imbalance between initial and final ramping phase. As explained above, the pre- processing step allow us to remove either the initial map \(S\) or final map \(M\) from the data, while we can only find a diagonal orthogonal estimate of the remaining map. A priori it is unclear which one of the two maps should be removed in order to reduce the systematic error more.  
+
+We have already treated the initial and final ramping phases on a different footing, however. The reason for this is rooted in the specifics of the ramping of the couplers compared to the qubits. The couplers need to be ramped from their idle frequencies to provide the desired target frequencies of \(20\mathrm{MHz}\) . This is why we expect the time scale of the initial ramping to be mainly determined by the couplers, namely the delay until they arrive around the target frequency and the time it takes to stabilize at the target frequency. In contrast, the final ramping map becomes effectively diagonal as soon as the couplers are again out of the MHz regime. We therefore expect that the initial map has a sizeable non- diagonal orthogonal component, whereas the final map is approximately diagonal.  
+
+We build trust in this assumption using experimental data in Fig. 6. We observe that the deviation of the orthogonal part \(\hat{O}_S\) of the identified initial map \(\hat{S}\) from its projection \(\hat{D}_S\) to diagonal orthogonal matrices is much larger than the corresponding deviation for the final map (Fig. 6(a)). Moreover, both the root- mean- square fit of the data (Fig. 6(c)) and the analog implementation error of the identified Hamiltonian with its target (Fig. 6(b)) are significantly improved when removing the initial ramp, as compared to removing the final
+
+<--- Page Split --->
+![](images/Figure_6.jpg)
+
+<center>Figure 6. Initial ramp removal versus final ramp removal. We identify Hamiltonians of a set of 5-qubit Hamiltonians with Hofstadter butterfly potentials \(\mu_{q} = 20\cos (2\pi qb)\) MHz for qubits \(q = 1,\ldots ,5\) and flux value \(b\) in without regularization. (a) Deviation of the orthogonal part \(\hat{O}_S\) ( \(\hat{O}_M\) ) of the identified initial map \(\hat{S}\) (final map \(\hat{M}\) ) from the closest diagonal orthogonal matrix \(\hat{D}_S\) ( \(\hat{D}_M\) ). (b) Analog implementation error of the corresponding identified Hamiltonians \(\hat{h}_S\) ( \(\hat{h}_M\) ). (c) Total root-mean-square deviation of the time series data from the Hamiltonian fit. </center>  
+
+ramp. This indicates that \(S\) induces a larger systematic error than \(\hat{M}\) . Correspondingly, it is indeed more advantageous to remove the initial map in the pre- processing and fit the final map with a diagonal orthogonal matrix, validating the approach taken here. In the SM, we provide further numerical evidence that this approach leads to small systematic errors and recovers a model with good predictive power.  
+
+### D. Benchmarking the algorithm  
+
+We benchmark our identification algorithm against more direct approaches in numerical simulations including models for statistical and systematic errors in the SM VI. We find that, indeed, already for small system sizes, the regularized manifold optimization algorithm developed here features an improved robustness against state preparation and measurement errors compared to (post- projected) linear inversion. For intermediate system sizes ( \(N > 10\) ), exploiting structure in the recovery algorithm then becomes an imperative. In particular, for larger system sizes the eigenspectrum of the Hamiltonian becomes unavoidably narrower spaced. We find that on instances of the Harper Hamiltonian studied here linear inversion approaches cannot be applied at all for \(N > 20\) . Regularized conjugate gradient decent in contrast yields good recovery performance even for larger systems. The same limitations apply to a direct Fourier analysis of the cumulative time series data using ESPRIT, as described above. For different families of Hamiltonians, we find that above a system size of \(N \approx 20\) tensorESPRIT still consistently recovers the frequency spectrum, while the ESPRIT algorithm fails to do so.  
+
+![](images/Figure_7.jpg)
+
+<center>Figure 7. Numerical benchmarks for larger system sizes. Recovery error of frequencies (golden) and Hamiltonians (red) from simulated time series averaged over 20 instances of Harper Hamiltonians for different system sizes. The error bars represent one standard deviation. The evolution is simulated for up to \(0.6 \mu s\) and sampled at a rate of \(250 \mathrm{MHz}\) . Statistical noise is simulated using \(10^{3}\) shots per expected value and SPAM is modeled by using randomly chosen idle qubit and coupler frequencies, linear ramping of \(1.5 \mathrm{GHz / s}\) padded by \(0.05 \mathrm{ns}\) . The fitting error of the time series is depicted in blue, right \(y\) -axis. We refer to the SM, Sec. VII A for details. </center>   
+
+Using structure not only allows our algorithm to denoise the data and achieve error robustness, it also makes precise Hamiltonian identification possible even with the number of measurements dramatically reduced in the spirit of compressed sensing. As described above, the number of measurements scales quadratically with the system size. We find that using the conjugate gradient algorithm the identification procedure reliably recovers Hamiltonians even when it has access to only about \(3\%\) of the measurements. In this regime, the linear inverse problem of finding the eigenvectors is underdetermined. Thus, the required experimental resources can be significantly reduced for large system sizes.  
+
+To demonstrate our method's scalability, Fig. 7 shows the recovery performance of the structure- exploiting algorithm on simulated data under realistic models for SPAM errors and with finite measurement statistics in the regime where the baseline approaches could not be applied anymore.  
+
+As detailed in the SM, tensorESPRIT has computational complexity in \(\mathcal{O}(L^2 N^3)\) . It is not straight- forward to bound the computational complexity of the conjugate gradient descent, as it depends on the required precision of the matrix exponential and the number of descent steps until convergence. The entire identification algorithm consumes \(\mathcal{O}(LN^2)\) memory. In practice, we find that the algorithm can be easily deployed on a consumer- grade laptop computer, e.g. reconstructing Hamiltonians of size \(N = 50\) in around 5 minutes.  
+
+### E. Error estimation  
+
+We here discuss how we estimate the systematic and statistical contributions to the error on the identified Hamiltonian \(\hat{h}\)
+
+<--- Page Split --->
+
+and initial map \(\hat{S}\) . Note that the impact of the systematic error on predicting results of experiments with the same initial and final ramps is reduced due to the gauge invariance of the model (6). Due to this freedom, some of the error in identifying \(\hat{D}_M \sim M\) gets accounted for by a corresponding error in the identification \(\hat{h} \sim h\) and \(\hat{S} \sim S\) in expressions of the type \(\hat{M} e^{- i t \hat{h}} \hat{S}\) . This prediction error can be further decreased by running the algorithm twice—removing the initial map in the first run and the final map in the second run, using the first ramp estimates to partially remove the ramps from the data before running the second iteration of the identification. This procedure is detailed and supported by numerical evidence in the SM.  
+
+Systematic error: Final ramp effect estimation. In order to estimate the magnitude of the systematic error that is induced by the non- trivial final map, we use a linear model of the final ramping phase with a constant ramping speed and constant wait time between the coupler and qubit ramping. We detail and present validation of this ramping model with a separate experiment in the SM, where we also provide empirical estimates of the model parameters.  
+
+Given a Hamiltonian matrix \(\hat{h}\) and the initial ramp \(\hat{S}\) obtained from experimental data, we recover the Hamiltonian matrix \(\hat{h}'\) from data simulated using the model \((\hat{h}, \hat{S}, M)\) , where \(M\) is the final ramp given by our ramping model. We use \(|f(\hat{h}) - f(\hat{h}')|\) as an estimate of the systematic error on quantities of the form \(f(\hat{h}) \in \mathbb{R}\) .  
+
+Statistical error: Bootstrapping. We estimate the effect of finite measurement statistics on the Hamiltonian estimate that is returned by the identification method via parametric bootstrapping. To this end, we simulate time series data with statistical noise using Haar- random unitaries \(S\) as initial ramps,  
+
+the identified Hamiltonian \(\hat{h}\) and final ramp \(M = \mathbb{1}\) , as detailed in the SM.  
+
+Acknowledgements. We acknowledge contributions from Charles Neill, Kostyantyn Kechedzhi, and Alexander Korotkov to the calibration procedure used in this analog approach. We would like to thank Christian Krumnow, Benjamin Chiaro, Alireza Seif, Markus Heinrich, and Juani Bermejo- Vega for fruitful discussions in early stages of the project. The hardware used for this experiment was developed by the Google Quantum AI hardware team, under the direction of Anthony Megrant, Julian Kelly and Yu Chen.  
+
+Funding. D. H. acknowledges funding from the U.S. Department of Defense through a QuICS Hartree fellowship. This work has been supported by the BMBF (DAQC), for which it provides benchmarking tools for analog- digital superconducting quantum devices, as well as by the DFG (specifically EI 519 20- 1 on notions of Hamiltonian learning, but also CRC 183 and GRK 2433 Deadalus). We have also received funding from the European Union's Horizon2020 research and innovation programme (PASQuanS2) on programmable quantum simulators, and the ERC (DebuQC).  
+
+Author contributions. D.H. and I.R. conceived of the Hamiltonian identification algorithm. J.F. conceived of the tensorESPRIT algorithm. D.H., I.R. and J.F. analyzed the experimental data and benchmarked the identification algorithm. P.R. took the experimental data. D.H. and I.R. wrote the initial manuscript. All authors contributed to discussions and writing the final manuscript.  
+
+Data and materials availability. The experimental data is available from the authors upon reasonable request.  
+
+Conflict of interest. The authors declare no conflict of interest.  
+
+[1] R. P. Feynman, Simulating Physics with computers, Int. J. Theor. Phys. 21, 467 (1982). [2] S. Lloyd, Universal quantum simulators, Science 273, 1073 (1996). [3] I. Bloch, J. Dalibard, and S. Nascimbene, Quantum simulations with ultracold quantum gases, Nature Phys. 8, 267 (2012). [4] R. Blatt and C. F. Roos, Quantum simulations with trapped ions, Nature Phys. 8, 277 (2012). [5] J. Smith, A. Lee, P. Richerme, B. Neyenhuis, P. W. Hess, P. Hauke, M. Heyl, D. Huse, and C. Monroe, Many- body localization in a quantum simulator with programmable random disorder, Nature Phys. 12, 907 (2016). [6] S. Ebadi, T. T. Wang, H. Levine, A. Keesling, G. Semeghini, A. Omran, D. Bluvstein, R. Samajdar, H. Pichler, W. W. Ho, S. Choi, S. Sachdev, M. Greiner, V. Vuletic, and M. D. Lukin, Quantum phases of matter on a 256- atom programmable quantum simulator, Nature 595, 227 (2021). [7] D. Aharonov and M. Ben- Or, Fault- tolerant quantum computation with constant error rate, SIAM J. Comput. 38, 1207 (2008). [8] E. Derbyshire, J. Y. Malo, A. J. Daley, E. Kashefi, and P. Wallden, Randomized benchmarking in the analogue setting, Quantum Sci. Technol. 5, 034001 (2020).  
+
+[9] R. Shaffer, E. Megidish, J. Broz, W.- T. Chen, and H. Haffner, Practical verification protocols for analog quantum simulators, npj Quant. Inf. 7, 1 (2021). [10] J. Helsen, S. Nezami, M. Reagor, and M. Walter, Matchgate benchmarking: Scalable benchmarking of a continuous family of many- qubit gates, Quantum 6, 657 (2020). [11] S. G. Schirmer, A. Kolli, and D. K. L. Oi, Experimental Hamiltonian identification for controlled two- level systems, Phys. Rev. A 69, 050306 (2004). [12] J. H. Cole, S. G. Schirmer, A. D. Greentree, C. J. Wellard, D. K. L. Oi, and L. C. L. Hollenberg, Identifying an experimental two- state Hamiltonian to arbitrary accuracy, Phys. Rev. A 71, 062312 (2005). [13] J. H. Cole, A. D. Greentree, D. K. L. Oi, S. G. Schirmer, C. J. Wellard, and L. C. L. Hollenberg, Identifying a two- state Hamiltonian in the presence of decoherence, Phys. Rev. A 73, 062333 (2006). [14] J. H. Cole, S. J. Devitt, and L. C. L. Hollenberg, Precision characterization of two- qubit Hamiltonians via entanglement mapping, J. Phys. A 39, 14649 (2006). [15] S. G. Schirmer, D. K. L. Oi, and S. J. Devitt, Physics- based mathematical models for quantum devices via experimental system identification, J. Phys. Conf. Ser. 107, 012011 (2008).
+
+<--- Page Split --->
+
+[16] S. G. Schirmer and D. K. L. Oi, Two- qubit Hamiltonian tomography by Bayesian analysis of noisy data, Phys. Rev. A 80, 022333 (2009).[17] D. K. L. Oi and S. G. Schirmer, Quantum system characterization with limited resources, Phil. Trans. R. Soc. A 370, 5386 (2012).[18] D. Burgarth, K. Maruyama, and F. Nori, Coupling strength estimation for spin chains despite restricted access, Phys. Rev. A 79, 020305 (2009).[19] D. Burgarth, K. Maruyama, and F. Nori, Indirect quantum tomography of quadratic Hamiltonians, New J. Phys. 13, 013019 (2011).[20] D. Burgarth and K. Maruyama, Indirect Hamiltonian identification through a small gateway, New J. Phys. 11, 103019 (2009).[21] C. Di Franco, M. Paternostro, and M. S. Kim, Hamiltonian Tomography in an access- limited setting without state initialization, Phys. Rev. Lett. 102, 187203 (2009).[22] M. Wiesniak and M. Markiewicz, Finding traps in non- linear spin arrays, Phys. Rev. A 81, 032340 (2010).[23] D. Burgarth and K. Yuasa, Quantum system identification, Phys. Rev. Lett. 108, 080502 (2012).[24] J. Zhang and M. Sarovar, Quantum Hamiltonian identification from measurement time traces, Phys. Rev. Lett. 113, 080401 (2014).[25] A. Sone and P. Cappellaro, Exact dimension estimation of interacting qubit systems assisted by a single quantum probe, Phys. Rev. A 96, 062334 (2017).[26] J. R. Garrison and T. Grover, Does a single eigenstate encode the full Hamiltonian? Phys. Rev. X 8, 021026 (2018).[27] X.- L. Qi and D. Ranard, Determining a local Hamiltonian from a single eigenstate, Quantum 3, 159 (2019).[28] E. Chertkov and B. K. Clark, Computational inverse method for constructing spaces of quantum models from wave functions, Phys. Rev. X 8, 031029 (2018).[29] E. Bairey, I. Arad, and N. H. Lindner, Learning a local Hamiltonian from local measurements, Phys. Rev. Lett. 122, 020504 (2019).[30] E. Bairey, C. Guo, D. Poletti, N. H. Lindner, and I. Arad, Learning the dynamics of open quantum systems from their steady states, New J. Phys. 22, 032001 (2020).[31] T. J. Evans, R. Harper, and S. T. Flammia, Scalable Bayesian Hamiltonian learning, (2019), arxiv:1912.07636. [32] Z. Li, L. Zou, and T. H. Hsieh, Hamiltonian tomography via quantum quench, Phys. Rev. Lett. 124, 160502 (2020).[33] A. Czerwinski, Hamiltonian tomography by the quantum quench protocol with random noise, Phys. Rev. A 104, 052431 (2021).[34] A. Elben, S. T. Flammia, H.- Y. Huang, R. Kueng, J. Preskill, B. Vermersch, and P. Zoller, The randomized measurement toolbox, Nature Rev. Phys. 5, 9 (2023), arxiv:2203.11374. [35] A. Valenti, E. van Nieuwenburg, S. Huber, and E. Greplova, Hamiltonian learning for quantum error correction, Phys. Rev. Res. 1, 033092 (2019).[36] P. Bienias, A. Seif, and M. Hafezi, Meta Hamiltonian learning, arXiv:2104.04453. [37] S. Krastanov, S. Zhou, S. T. Flammia, and L. Jiang, Stochastic estimation of dynamical variables, Quantum Sci. Technol. 4, 035003 (2019).[38] L. Che, C. Wei, Y. Huang, D. Zhao, S. Xue, X. Nie, J. Li, D. Lu, and T. Xin, Learning quantum Hamiltonians from single- qubit measurements, Phys. Rev. Res. 3, 023246 (2021).[39] F. Wilde, A. Kshetrimayum, I. Roth, D. Hangleiter, R. Sweke, and J. Eisert, Scalably learning quantum many- body Hamiltonians from dynamical data, (2022), arxiv:2209.14328.  
+
+[40] W. Yu, J. Sun, Z. Han, and X. Yuan, Robust and efficient Hamiltonian learning, Quantum 7, 1045 (2023), arxiv:2201.00190. [41] H.- Y. Huang, Y. Tong, D. Fang, and Y. Su, Learning many- body Hamiltonians with Heisenberg- limited scaling, Phys. Rev. Lett. 130, 200403 (2023).[42] H. Li, Y. Tong, H. Ni, T. Gefen, and L. Ying, Heisenberg- limited Hamiltonian learning for interacting bosons, (2023), arxiv:2307.04690. [43] S. T. Flammia and J. J. Wallman, Efficient Estimation of Pauli Channels, ACM Transactions on Quantum Computing 1, 3:1 (2020).[44] R. Harper, S. T. Flammia, and J. J. Wallman, Efficient Learning of Quantum Noise, Nature Physics 16, 1184 (2020).[45] E. H. Lapasar, K. Maruyama, D. Burgarth, T. Takui, Y. Kondo, and M. Nakahara, Estimation of coupling constants of a three- spin chain: Case study of Hamiltonian tomography with NMR, New J. Phys. 14, 013043 (2012).[46] S.- Y. Hou, H. Li, and G.- L. Long, Experimental quantum Hamiltonian identification from measurement time traces, Science Bulletin 62, 863 (2017).[47] X. Chen, Y. Li, Z. Wu, R. Liu, Z. Li, and H. Zhou, Experimental realization of Hamiltonian tomography by quantum quenches, Phys. Rev. A 103, 042429 (2021).[48] D. Zhao, C. Wei, S. Xue, Y. Huang, X. Nie, J. Li, D. Ruan, D. Lu, T. Xin, and G. Long, Characterizing quantum simulations with quantum tomography on a spin quantum simulator, Phys. Rev. A 103, 052403 (2021).[49] E. Flurin, L. S. Martin, S. Hacohen- Gourgy, and I. Siddiqi, Using a recurrent neural network to reconstruct quantum dynamics of a superconducting qubit from physical observations, Phys. Rev. X 10, 011006 (2020).[50] G. O. Samach, A. Greene, J. Borregaard, M. Christandl, D. K. Kim, C. M. McNally, A. Melville, B. M. Niedzielski, Y. Sung, D. Rosenberg, M. E. Schwartz, J. L. Yoder, T. P. Orlando, J. I.- J. Wang, S. Gustavsson, M. Kjaergaard, and W. D. Oliver, Lindblad tomography of a superconducting quantum processor, Phys. Rev. Appl. 18, 064056 (2022).[51] C. Kokail, R. van Bijnen, A. Elben, B. Vermersch, and P. Zoller, Entanglement Hamiltonian tomography in quantum simulation, Nature Phys. 17, 936 (2021).[52] C. Kokail, B. Sundar, T. V. Zache, A. Elben, B. Vermersch, M. Dalmonte, R. van Bijnen, and P. Zoller, Quantum variational learning of the entanglement Hamiltonian, Phys. Rev. Lett. 127, 170501 (2021).[53] M. K. Joshi, C. Kokail, R. van Bijnen, F. Kranzl, T. V. Zache, R. Blatt, C. F. Roos, and P. Zoller, Exploring large- scale entanglement in quantum simulation, Nature, 1 (2023).[54] R. Roy, A. Paulraj, and T. Kailath, Estimation of signal parameters via rotational invariance techniques- ESPRIT, in MILCOM 1986- IEEE Military Communications Conference: Communications- Computers: Teamed for the 90's, Vol. 3 (IEEE, 1986) pp. 41- 6. [55] A. Fannjiang, Compressive spectral estimation with single- snapshot esprit: Stability and resolution, arXiv:1607.01827. [56] W. Li, W. Liao, and A. Fannjiang, Super- resolution limit of the ESPRIT algorithm, IEEE Trans. Inf. Th. 66, 4593 (2020), arXiv:1905.03782. [57] T. Abrudan, J. Eriksson, and V. Koivunen, Conjugate gradient algorithm for optimization under unitary matrix constraint, Signal Processing 89, 1704 (2009).[58] I. Carusotto, A. A. Houck, A. J. Kollar, P. Roushan, D. I. Schuster, and J. Simon, Photonic materials in circuit quantum electrodynamics, Nature Phys. 16, 268- 279 (2020).
+
+<--- Page Split --->
+
+[59] F. Yan, P. Krantz, Y. Sung, M. Kjaergaard, D. Campbell, J. I. J. Wang, T. P. Orlando, S. Gustavsson, and W. D. Oliver, A tunable coupling scheme for implementing high- fidelity two- qubit gates, Phys. Rev. Appl. 10, 054062 (2018).[60] P. Bühlmann and S. V. D. Geer, Statistics for high- dimensional data, Springer Series in Statistics, Vol. 9 (Springer, Berlin, 2011).[61] D. R. Hofstadter, Energy levels and wave functions of Bloch electrons in rational and irrational magnetic fields, Phys. Rev. B 14, 2239 (1976).[62] J. Kelly, P. O'Malley, M. Neeley, H. Neven, and J. M. Martinis, Physical qubit calibration on a directed acyclic graph, arXiv:1803.03226. [63] E. J. Candès and C. Fernandez- Granda, Super- resolution from noisy data, J. Fourier An. App. 19, 1229 (2013).[64] E. J. Candès and C. Fernandez- Granda, Towards a mathematical theory of super- resolution, Comm. Pure App. Math. 67, 906 (2014).[65] D. Hangleiter, I. Roth, D. Nagaj, and J. Eisert, Easing the Monte Carlo sign problem, Science Adv. 6, eabb8341 (2020).[66] A. Edelman, T. A. Arias, and S. T. Smith, The geometry of algorithms with orthogonality constraints, SIAM J. Matr. Ana. App. 20, 303 (1998).[67] I. A. Luchnikov, A. Ryzhov, S. N. Filippov, and H. Ouerdane, QGOpt: Riemannian optimization for quantum technologies, arXiv:2011.01894. [68] I. Luchnikov, M. Krechetov, and S. Filippov, Riemannian geometry and automatic differentiation for optimization problems of quantum physics and quantum technologies, New J. Phys. 23, 073006 (2021).[69] I. Roth, J. Wilkens, D. Hangleiter, and J. Eisert, Semidevice- dependent blind quantum tomography, Quantum 7, 1053 (2023).
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- Hlearningsupplemental.pdf
+
+<--- Page Split --->

preprint/preprint__7e982bca866a39190fd88f0488bc87bcf29e37f1fdbce365b6032e9477099024/preprint__7e982bca866a39190fd88f0488bc87bcf29e37f1fdbce365b6032e9477099024_det.mmd

@@ -0,0 +1,421 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 888, 177]]<|/det|>
+# Robustly learning the Hamiltonian dynamics of a superconducting quantum processor  
+
+<|ref|>text<|/ref|><|det|>[[44, 196, 140, 214]]<|/det|>
+Jens Eisert  
+
+<|ref|>text<|/ref|><|det|>[[55, 223, 279, 240]]<|/det|>
+jenseisert@gmail.com  
+
+<|ref|>text<|/ref|><|det|>[[44, 269, 711, 476]]<|/det|>
+FU Berlin https://orcid.org/0000- 0003- 3033- 1292Dominik HangleiterUniversity of Maryland College Park https://orcid.org/0000- 0002- 4766- 7967Ingo RothFreie Universität Berlin https://orcid.org/0000- 0002- 1191- 7442Jonas FuksaFU Berlin https://orcid.org/0000- 0003- 4606- 2584Pedram RoushanGoogle (United States) https://orcid.org/0000- 0003- 1917- 3879  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 515, 103, 532]]<|/det|>
+## Article  
+
+<|ref|>title<|/ref|><|det|>[[44, 553, 135, 570]]<|/det|>
+# Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 590, 336, 609]]<|/det|>
+Posted Date: February 16th, 2024  
+
+<|ref|>text<|/ref|><|det|>[[44, 629, 475, 648]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 3813225/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 666, 912, 709]]<|/det|>
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 727, 533, 747]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 782, 951, 825]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on November 6th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 52629- 3.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[140, 63, 861, 82]]<|/det|>
+# Robustly learning the Hamiltonian dynamics of a superconducting quantum processor  
+
+<|ref|>text<|/ref|><|det|>[[92, 93, 911, 194]]<|/det|>
+Dominik Hangleiter, \(^{1,2,\ast}\) Ingo Roth, \(^{3,4,\ast}\) Jonás Fuksa, \(^{4}\) Jens Eisert, \(^{4,5}\) and Pedram Roushan \(^{6}\) \(^{1}\) Joint Center for Quantum Information and Computer Science (QuCS), University of Maryland and NIST, College Park, MD 20742, USA \(^{2}\) Joint Quantum Institute (JQI), University of Maryland and NIST, College Park, MD 20742, USA \(^{3}\) Quantum Research Center, Technology Innovation Institute (TII), Abu Dhabi \(^{4}\) Dahlem Center for Complex Quantum Systems, Freie Universität Berlin, 14195 Berlin, Germany \(^{5}\) Helmholtz-Zentrum Berlin für Materialien und Energie, 14109 Berlin, Germany \(^{6}\) Google Quantum AI, Mountain View, CA, USA  
+
+<|ref|>text<|/ref|><|det|>[[175, 201, 830, 425]]<|/det|>
+The required precision to perform quantum simulations beyond the capabilities of classical computers imposes major experimental and theoretical challenges. The key to solving these issues are highly precise ways of characterizing analog quantum simulators. Here, we robustly estimate the free Hamiltonian parameters of bosonic excitations in a superconducting- qubit analog quantum simulator from measured time- series of single- mode canonical coordinates. We achieve the required levels of precision in estimating the Hamiltonian parameters by maximally exploiting the model structure, making it robust against noise and state- preparation and measurement (SPAM) errors. Importantly, we are also able to obtain tomographic information about those SPAM errors from the same data, crucial for the experimental applicability of Hamiltonian learning in dynamical quantum- quench experiments. Our learning algorithm is highly scalable both in terms of the required amounts of data and post- processing. To achieve this, we develop a new super- resolution technique coined tensorESPRIT for frequency extraction from matrix time- series. The algorithm then combines tensorESPRIT with constrained manifold optimization for the eigenspace reconstruction with pre- and post- processing stages. For up to 14 coupled superconducting qubits on two Sycamore processors, we identify the Hamiltonian parameters—verifying the implementation on one of them up to sub- MHz precision—and construct a spatial implementation error map for a grid of 27 qubits. Our results constitute a fully characterized, highly accurate implementation of an analog dynamical quantum simulation and introduce a diagnostic toolkit for understanding, calibrating, and improving analog quantum processors.  
+
+<|ref|>text<|/ref|><|det|>[[86, 451, 486, 653]]<|/det|>
+Analog quantum simulators promise to shed light on fundamental questions of physics that have remained elusive to the standard methods of inference [1, 2]. Recently, enormous progress in controlling individual quantum degrees of freedom has been made towards making this vision a reality [3- 6]. While in digital quantum computers small errors can be corrected [7], it is intrinsically difficult to error- correct analog devices. Yet, the usefulness of analog quantum simulators as computational tools depends on the error of the implemented dynamics. Meeting this requirement hinges on devising characterization methods that not only yield a benchmark of the overall functioning of the device [e.g., 8- 10], but more importantly provide diagnostic information about the sources of errors.  
+
+<|ref|>text<|/ref|><|det|>[[86, 655, 486, 841]]<|/det|>
+Developing characterization tools for analog quantum simulators requires hardware developments as well as theoretical analysis and method development. With the advent of highly controlled quantum systems, efficient methods for identifying certain Hamiltonian parameters from dynamical data have been devised for specific classes of Hamiltonians. Key ideas are the use of Fourier analysis [11- 17] and tracking the dynamics of single excitations [18- 23]. For general Hamiltonian models, specific algebraic structures of the Hamiltonian terms can be exploited [24, 25]. Generalizing these ideas, a local Hamiltonian can be learned from a single eigenstate or its steady state [26- 31] or using quantum- quenches [32, 33], an approach dubbed 'correlation matrix method' [34]. Alterna  
+
+<|ref|>text<|/ref|><|det|>[[515, 451, 916, 537]]<|/det|>
+tively, one can apply general- purpose machine- learning methods [35- 39]. More recently, optimal theoretical guarantees have been derived for Hamiltonian learning schemes [40- 42] based on Pauli noise tomography [43, 44]. Crucially, these protocols assume perfect mid- circuit quenches, which—as we find here—can be a limiting assumption in practice.  
+
+<|ref|>text<|/ref|><|det|>[[515, 538, 916, 854]]<|/det|>
+This recent rapid theoretical development is not quite matched by concomitant experimental efforts. The effectiveness of some of these methods has been demonstrated for the estimation of a small number of coupling parameters of fixed two- and three- qubit Hamiltonians in nuclear magnetic resonance (NMR) experiments [45- 48]. While in NMR, the dominant noise process is decoherence, in tunable quantum simulators such as superconducting qubits, trapped ions or cold atoms in optical lattices, state preparation and measurement (SPAM) errors, as we also demonstrate here, play a central role. Initial steps at characterizing such errors as well as the dissipative Lindblad dynamics for up to two qubits in a superconducting qubit platform have been taken recently [49, 50]. Hamiltonian learning of thermal states has recently also been applied in many- body experiments as a means to characterize the entanglement of up to 20- qubit subsystems whose reduced states are parameterized by the so- called entanglement Hamiltonian [51- 53]. The challenge remains to develop and experimentally demonstrate the feasibility of scalable methods for a robust and precise identification of Hamiltonian dynamics of intermediate- size systems subject to both incoherent noise and systematic SPAM errors.  
+
+<|ref|>text<|/ref|><|det|>[[515, 856, 916, 913]]<|/det|>
+Here, we develop bespoke protocols to robustly and accurately identify the full Hamiltonian of a large- scale bosonic system and implement those protocols on superconducting quantum processors. Given the complexity of the learning
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[86, 65, 487, 225]]<|/det|>
+task, we focus on identifying the non- interacting part of a potentially interacting system. We are able to estimate the corresponding Hamiltonian parameters as well as SPAM errors pertaining to all individual components of the superconducting chip for up to 14- mode Hamiltonians tuned across a broad parameter regime, in contrast to previous experiments. Given the identified Hamiltonians, we quantify their implementation error. We demonstrate and verify that a targeted intermediate- size Hamiltonian is implemented on a large region of the superconducting processor with sub- MHz precision in a broad parameter range.  
+
+<|ref|>text<|/ref|><|det|>[[86, 225, 487, 356]]<|/det|>
+To this end, building on previous ideas for Hamiltonian identification [19, 24], we devise a simple and robust algorithm that exploits the structure of the system at hand. For the identification we make use of quadratically many experimental time- series tracking excitations via expectation values of canonical coordinates. Our structure- enforcing algorithm isolates two core tasks that need to be solved in Hamiltonian identification after suitable pre- processing of the data: frequency extraction and eigenspace reconstruction.  
+
+<|ref|>text<|/ref|><|det|>[[86, 356, 487, 601]]<|/det|>
+To solve the first task in a robust and structure- specific way, we develop a novel algorithm coined tensorESPRIT, which utilizes ideas from super- resolving, denoised Fourier analysis [54- 56] and tensor networks to extract frequencies from a matrix time- series. For the second task we use constrained manifold optimization over the orthogonal group [57]. Crucially, by explicitly exploiting all structure constraints of the identification problem, our method allows us to distinguish and obtain tomographic information about state- preparation and measurement errors. In the quench- based experiment this information renders identification and verification of the dynamics experimentally feasible in the first place. We further support our method development with numerical simulations of different noise effects and benchmark against more direct algorithmic approaches. We find that in contrast to other approaches our method is scalable to larger system sizes out of the reach of our current experimental efforts.  
+
+<|ref|>text<|/ref|><|det|>[[86, 602, 487, 702]]<|/det|>
+Our work constitutes a detailed case study that lays bare and provides solutions for the difficulties of practical Hamiltonian learning in a seemingly simple system. It thus provides a blueprint and paves the way for devising practical model- specific identification algorithms both for the interaction parameters of bosonic or fermionic systems and more complex settings.  
+
+<|ref|>text<|/ref|><|det|>[[86, 706, 487, 821]]<|/det|>
+Setup. We characterize the Hamiltonian governing analog dynamics of Google Sycamore chips which consist of a two- dimensional array of nearest- neighbour coupled superconducting qubits. Each physical qubit is a non- linear oscillator with bosonic excitations (microwave photons) [58]. Using the rotating- wave approximation the dynamics governing the excitations of the qubits in the rotating frame can be well described by the Bose- Hubbard Hamiltonian [59]  
+
+<|ref|>equation<|/ref|><|det|>[[104, 833, 487, 870]]<|/det|>
+\[H_{\mathrm{BH}} = \sum_{i}\left(\mu_{i}a_{i}^{\dagger}a_{i} + \eta_{i}a_{i}^{\dagger}a_{i}^{\dagger}a_{i}a_{i}\right) - \sum_{i\neq j}J_{i,j}a_{i}^{\dagger}a_{j}, \quad (1)\]  
+
+<|ref|>text<|/ref|><|det|>[[85, 882, 487, 912]]<|/det|>
+where \(a_{i}^{\dagger}\) and \(a_{i}\) denote bosonic creation and annihilation operators at site \(i\) , respectively, \(\mu \in \mathbb{R}^{N}\) are the on- site poten  
+
+<|ref|>text<|/ref|><|det|>[[515, 63, 916, 152]]<|/det|>
+tials, \(J\in \mathbb{R}^{N\times N}\) are the hopping rates between nearest neighbour qubits, and \(\eta \in \mathbb{R}^{N}\) are the strength of on- site interactions. The qubit frequency, the nearest- neighbour coupling between them, and the non- linearity (anharmonicity) set \(\mu ,J\) and \(\eta\) . We are able to tune \(\mu\) and \(J\) on nanosecond timescales, while \(\eta\) is fixed.  
+
+<|ref|>text<|/ref|><|det|>[[515, 153, 916, 211]]<|/det|>
+Here, we focus on the specific task of identifying the values of \(\mu_{i}\) and \(J_{i,j}\) . The corresponding non- interacting part of the Hamiltonian acting on \(N\) modes can be conveniently parametrized as  
+
+<|ref|>equation<|/ref|><|det|>[[631, 223, 915, 267]]<|/det|>
+\[H(h) = -\sum_{i,j = 1}^{N}h_{i,j}a_{i}^{\dagger}a_{j} \quad (2)\]  
+
+<|ref|>text<|/ref|><|det|>[[515, 279, 916, 380]]<|/det|>
+with an \(N\times N\) real symmetric parameter matrix \(h\) with entries \(h_{i,j}\) , which is composed of the on- site chemical potentials \(\mu_{i}\) on its diagonal and the hopping energies \(J_{i,j}\) for \(i\neq j\) . The identification of the non- interacting part \(H(h)\) of \(H_{\mathrm{BH}}\) can be viewed as a first step in a hierarchical procedure for characterizing dynamical quantum simulations with tunable interactions and numbers of particles.  
+
+<|ref|>text<|/ref|><|det|>[[515, 380, 916, 627]]<|/det|>
+The non- interacting part \(H(h)\) of the Hamiltonian \(H_{\mathrm{BH}}\) can be inferred when initially preparing a state where only a single qubit is excited with a single photon. For initial states with a single excitation, the interaction term vanishes, hence effectively \(\eta = 0\) . Consequently, only the two lowest energy levels of the non- linear oscillators enter the dynamics. Therefore, referring to them as qubits (two- level systems) is precise. Specifically, we identify the parameters \(h_{i,j}\) from dynamical data of the following form. We initialize the system in \(|\psi_{n}\rangle \coloneqq (\mathbb{1} + a_{n}^{\dagger})|0\rangle^{\otimes N} / \sqrt{2}\) and measure the canonical coordinates \(x_{m} = (a_{m} + a_{m}^{\dagger}) / 2\) and \(p_{m} = (a_{m} - a_{m}^{\dagger}) / (2\mathrm{i})\) for all combinations of \(m,n = 1,\ldots ,N\) . In terms of the qubit architecture, this amounts to local Pauli- \(X\) and Pauli- \(Y\) basis measurements, respectively. We combine the statistical averages over multiple measurements to obtain an empirical estimator for \(\langle a_{m}(t)\rangle_{\psi_{n}} = \langle x_{m}(t)\rangle_{\psi_{n}} + \mathrm{i}\langle p_{m}(t)\rangle_{\psi_{n}}\) . For particle- number preserving dynamics, this data is of the form  
+
+<|ref|>equation<|/ref|><|det|>[[610, 638, 915, 670]]<|/det|>
+\[\langle a_{m}(t)\rangle_{\psi_{n}} = \frac{1}{2}\exp (-i t h)_{m,n}. \quad (3)\]  
+
+<|ref|>text<|/ref|><|det|>[[515, 680, 916, 722]]<|/det|>
+It therefore directly provides estimates of the entries of the time- evolution unitary at time \(t\) in the single- particle sector of the bosonic Fock space.  
+
+<|ref|>text<|/ref|><|det|>[[515, 724, 916, 912]]<|/det|>
+In Fig. 1, we show an overview of the experimental procedure, and the different steps of the Hamiltonian identification algorithm. Every experiment uses a few coupled qubits, from the larger array of qubits on the device (Fig. 1(a)). On those qubits, the goal is to implement the time- evolution with targeted Hamiltonian parameters \(h_{0}\) , which are subject to connectivity constraints imposed by the couplings of the qubits. To achieve this, we perform the following pulse sequence to collect dynamical data of the form (3). Before the start of the sequence, the qubits are at frequencies (of the \(|0\rangle\) to \(|1\rangle\) transition) that could be a few hundred MHz apart from each other. In the beginning, all qubits are in their ground state \(|0\rangle\) . To prepare the initial state, a \(\pi /2\) - pulse is applied to one
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[85, 65, 914, 328]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[85, 346, 919, 480]]<|/det|>
+<center>Figure 1. Outline of the experiment and identification algorithm. (a) The time evolution under a target Hamiltonian \(h_0\) is implemented on an part of the Google Sycamore chip (gray) using the pulse sequence depicted in the middle. (b) The expected value of canonical coordinates \(x_m\) and \(p_m\) for each qubit \(m\) over time is estimated from measurements using different \(\psi_n\) as input states. (c) The data shown in (b) for each time \(t_0\) can be interpreted as a (complex-valued) matrix with entries indexed by measured and initial excited qubit, \(m\) and \(n\) . The identification algorithm proceeds in two steps: 1. From the matrix time-series, the Hamiltonian eigenfrequencies are extracted using our newly introduced algorithm coined tensorESPRIT, introduced in the SM, or an adapted version of the ESPRIT algorithm. The blue line indicates the denoised, high-resolution signal as 'seen' by the algorithm. 2. After removing the initial ramp using the data at some fixed time, the Hamiltonian eigenspaces are reconstructed using a non-convex optimization algorithm over the orthogonal group. We obtain a diagonal orthogonal estimate of the final ramp. From the extracted frequencies and reconstructed eigenspaces, we can calculate the identified Hamiltonian \(\hat{h}\) that describes the measured time evolution and a tomographic estimate of the initial ramp. </center>  
+
+<|ref|>text<|/ref|><|det|>[[86, 508, 488, 829]]<|/det|>
+of the qubits, resulting in its Bloch vector moving to the equator. Then ramping pulses are applied to all qubits to bring them to the desired detuning around a common rendezvous frequency (6500 MHz in this work). At the same time, pulses are applied to the couplers to set the nearest- neighbour hopping to the desired value (20 MHz in this work). The pulses are held at the target values for time \(t\) , corresponding to the evolution time of the experiment. Subsequently, the couplers are ramped back to zero coupling and the qubits back to their initial frequency, where \(\langle x_m(t) \rangle\) and \(\langle p_m(t) \rangle\) on the desired qubit \(m\) is measured. The initial and final pulse ramping take place over a finite time of 2–3 ns, and therefore give rise to a non- trivial effect on the dynamics, which we take into account in the identification procedure. In fact, we find that the effects of the ramping phase are the dominant source of SPAM errors in the quench- based analog simulation. The experimental data (Fig. 1(b)) on \(N\) qubits are \(N \times N\) time- series estimates of \(\langle a_m(t) \rangle_{\psi_n}\) for \(t = 0, 1, \ldots , T\) ns and all pairs \(n, m = 1, \ldots , N\) . Given those data, the identification task amounts to identifying the 'best' coefficient matrix \(h\) , describing the time- sequence of snapshots of the single- particle unitary matrix \(\frac{1}{2} \exp (- \mathrm{i} t h)\) .  
+
+<|ref|>text<|/ref|><|det|>[[86, 840, 488, 911], [515, 508, 917, 740]]<|/det|>
+Identification method. We can identify the generator \(h\) of the unitary in two steps (Fig. 1(c)), making use of the eigen- decomposition of the Hamiltonian (see Methods). In the first step, the time- dependent part of the identification problem is solved, namely, identifying the Hamiltonian eigenvalues (eigenfrequencies). In the second step, given the eigenvalues, the eigenbasis for the Hamiltonian of \(h\) is determined. In order to make the identification method noise- robust, we furthermore exploit structural constraints of the model. First, the Hamiltonian has a spectrum such that the time- series data has a time- independent, sparse frequency spectrum with exactly \(N\) contributions. Second, the Fourier coefficients of the data have an explicit form as the outer product of the orthogonal eigenvectors of the Hamiltonian. Third, the Hamiltonian parameter matrix is real and has an a priori known sparse support due to the experimental connectivity constraints. These structural constraints are not respected by various sources of incoherent noise, including particle loss and finite shot noise, and coherent noise, in particular the SPAM error. Thereby, an identification protocol that takes these constraints into account is intrinsically robust against various imperfections.   
+
+<|ref|>text<|/ref|><|det|>[[515, 753, 917, 911]]<|/det|>
+To robustly identify the sparse frequencies from the experimental data, we develop a new super- resolution and denoising algorithm tensorESPRIT that is applicable to matrix- valued time series and uses tensor network techniques in conjunction with super- resolution techniques for scalar data [55]. Achieving high precision in this step is crucial for identifying the eigenvectors in the presence of noise. To robustly identify the eigenbasis, in the second step, we perform least- square optimization of the time- series data under the orthonormality constraint with a gradient descent algorithm on the manifold structure of the orthogonal group [57]. Here, we incorporate
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[84, 60, 920, 428]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[84, 444, 919, 630]]<|/det|>
+<center>Figure 2. A single Hamiltonian recovery of a 5-mode Hamiltonian and the corresponding time domain data. (a) The full experimental time-series data \(\langle x_{m}(t)\rangle_{\psi_{n}}\) for \(m,n = 1,\ldots ,5\) and the best fit of those data in terms of our model \(\frac{1}{2} (M\exp (-i t h)S)_{m,n}\) for a diagonal and orthogonal \(M\) and linear map \(S\) (solid lines). (b) The target Hamiltonian matrix \(h_{0}\) , the identified Hamiltonian \(\hat{h}\) , and the deviation between them. The error of each diagonal entry is \(\pm (0.16 + 0.99)\mathrm{MHz}\) and of each off-diagonal entry \(\pm (0.12 + 0.50)\mathrm{MHz}\) and comprises of the statistical and the systematic error, respectively. The analog implementation error \(\mathcal{E}_{\mathrm{analog}}(\hat{h},h_{0})\) is \(0.73\pm (0.07 + 0.62)\mathrm{MHz}\) , and \(0.32\pm 0.00\mathrm{MHz}\) for the eigenfrequencies. The analog implementation error \(\mathcal{E}_{\mathrm{analog}}(\hat{S},\mathbb{1})\) of the identified initial map is \(0.61\pm (0.00 + 0.12)\) . (c) The real part of the initial map \(\hat{S}\) and the diagonal orthogonal estimate \(\hat{D}_{M}\) of the final map \(M\) , inferred from the data using the identified Hamiltonian \(\hat{h}\) . (d) Absolute value of the time-domain deviation of the fit from the full experimental data for each time series, given by deviation \([\hat{h},\hat{S},\hat{D}_{M}](t)_{m,n}\coloneqq \langle a_{m}(t)\rangle_{\psi_{n}} - \frac{1}{2}\hat{D}_{M}\exp (-i t\hat{h})\hat{S}\) . The insets represent the root-mean-square deviation of the Hamiltonian fit from the experimental data per time series, averaged over the evolution time for each matrix entry \((m,n)\) , resulting in an entry-wise summarized quality of fit. We find a total root-mean-square deviation of the fit of 0.14. (e) Instantaneous root-mean-square deviation of the identified Hamiltonian \(\hat{h}\) , initial map \(\hat{S}\) and final map \(\hat{D}_{M}\) and of the target Hamiltonian \(h_{0}\) with initial map fit \(S_{0}\) from the experimental data averaged over the distinct time series. </center>  
+
+<|ref|>text<|/ref|><|det|>[[85, 657, 487, 687]]<|/det|>
+the connectivity constraint on the coefficient matrix \(h\) by making use of regularization techniques [60].  
+
+<|ref|>text<|/ref|><|det|>[[86, 688, 488, 860]]<|/det|>
+Robustness against ramp errors. The initial and final ramping pulses result in a time- independent, linear transformation at the beginning and end of the time series. It is important to stress that such ramping pulses are expected to be generic in a wide range of experimental implementations of dynamical analog quantum simulations. Robustness of an Hamiltonian identification method against these imperfections is essential for accurate estimates in practice. We can model the effect of such particle number preserving state preparation and measurement (SPAM) errors via linear maps \(S\) and \(M\) , respectively, see the SM for details. This alters our model of the ideal data (3) to  
+
+<|ref|>equation<|/ref|><|det|>[[151, 865, 487, 893]]<|/det|>
+\[\langle a_{m}(t)\rangle_{\psi_{n}} = \frac{1}{2} (M\cdot \exp (-i t h)\cdot S)_{m,n}. \quad (4)\]  
+
+<|ref|>text<|/ref|><|det|>[[101, 898, 488, 913]]<|/det|>
+While for the frequency identification such time  
+
+<|ref|>text<|/ref|><|det|>[[513, 657, 916, 774]]<|/det|>
+independent errors 'only' deteriorate the signal- to- noise ratio, for the identification of the eigenvectors of \(h\) it is crucial to take the effects of non- trivial \(S\) and \(M\) into account. Given the details of the ramping procedure, we expect that the deviation of the initial map \(S\) from the identity will be significantly larger than that of the final map \(M\) and provide evidence for this in the Methods. In particular, the final map will be dominated by phase accumulation on the diagonal.  
+
+<|ref|>text<|/ref|><|det|>[[514, 775, 916, 881]]<|/det|>
+By pre- processing the data, we can robustly remove an arbitrary initial map \(S\) . By post- processing, we can obtain an orthogonal diagonal estimate \(\hat{D}_{M}\) of the final map \(M\) . We give numerical evidence that the estimate \(\hat{D}_{M}\) gives good results in the particular experimental setting. From the identified Hamiltonian and an orthogonal diagonal estimate \(\hat{D}_{M}\) of \(M\) , we get an estimate \(\hat{S}\) of \(S\) .  
+
+<|ref|>text<|/ref|><|det|>[[514, 883, 916, 911]]<|/det|>
+Error sources. There are two main remaining sources of error that affect the Hamiltonian identification. First, the esti
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[93, 60, 480, 675]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[85, 680, 488, 880]]<|/det|>
+<center>Figure 3. Comparing frequency and full identification errors. (a) In an \(N = 6\) subset of connected qubits, by varying \(b\) from 0 to 1, we implement 51 different Hamiltonians. The plot shows the Fourier transform of the time domain data. (b) The extracted eigenfrequencies (denoised peaks in panel (a)) are shown as colored dots, where the assigned color is indicative of the deviation between targeted eigenfrequencies (gray lines) and the identified ones from position of the peaks. (c) Analog implementation error \(\mathcal{E}_{\mathrm{analog}}(\hat{h},h_0)\) of the identified Hamiltonian (dark red) compared to the implementation error \(\mathcal{E}_{\mathrm{analog}}(\mathrm{eig}(\hat{h}),\mathrm{eig}(h_0))\) of the identified frequencies (golden). Colored (gray) error bars quantify the statistical (systematic) error. (d) Layout of the six qubits on the Sycamore processor and median of the entry-wise absolute-value deviation of the Hamiltonian matrix entries from their targeted values across the ensemble of 51 different values of \(b\in [0,1]\) . </center>  
+
+<|ref|>text<|/ref|><|det|>[[515, 65, 916, 156]]<|/det|>
+mate \(\hat{h}\) has a statistical error due to the finite number of measurements used to estimate the expectation values. Second, any non- trivial final map \(M\) will produce a systematic error in the eigenbasis reconstruction and the tomographic estimate \(\hat{S}\) . We partially remedy this effect with an orthogonal diagonal estimate \(\hat{D}_M\) of \(M\) .  
+
+<|ref|>text<|/ref|><|det|>[[515, 160, 916, 336]]<|/det|>
+Results. We implement and characterize different Hamiltonians from time- series data on two distinct quantum Sycamore processors—Sycamore #1 and #2. The Hamiltonians we implement have a fixed overall hopping strength \(J_{i,j} = 20 \mathrm{MHz}\) and site- dependent local potentials \(\mu_i\) on subsets of qubits. Specifically, we choose the local potentials quasi- randomly \(\mu_q = 20 \cos (2\pi qb) \mathrm{MHz}\) , for \(q = 1, \ldots , N\) , where \(b\) is a number between zero and one. In one dimension, this choice corresponds to implementing the Harper Hamiltonian, which exhibits characteristic 'Hofstadter butterfly' frequency spectra as a function of the dimensionless magnetic flux \(b\) [61].  
+
+<|ref|>text<|/ref|><|det|>[[515, 336, 916, 379]]<|/det|>
+We measure deviations in the identification in terms of the analog implementation error of the identified Hamiltonian \(\hat{h}\) with respect to the targeted Hamiltonian \(h_0\) as  
+
+<|ref|>equation<|/ref|><|det|>[[580, 387, 915, 421]]<|/det|>
+\[\mathcal{E}_{\mathrm{analog}}(\hat{h},h_0)\coloneqq \frac{1}{N}\left\| \hat{h} -h_0\right\|_{\ell_2}, \quad (5)\]  
+
+<|ref|>text<|/ref|><|det|>[[515, 430, 916, 607]]<|/det|>
+defined in terms of the \(\ell_2\) - norm, which for a matrix \(A\) is given by \(\| A\|_{\ell_2} = (\sum_{i,j}|A_{i,j}|^2)^{1 / 2}\) . We also use the analog implementation error to quantify the implementation error of the initial map \(\hat{S}\) as \(\mathcal{E}_{\mathrm{analog}}(\hat{S}, \mathbb{1})\) , and of the eigenfrequencies \(\mathrm{eig}(\hat{h})\) as \(\mathcal{E}_{\mathrm{analog}}(\mathrm{eig}(\hat{h}), \mathrm{eig}(h_0))\) . Notice that the analog implementation error of the frequencies in the data from the targeted Hamiltonian eigenfrequencies give a lower bound to the overall implementation error of the identified Hamiltonian. This is because the \(\ell_2\) - norm used in the definition (5) of \(\mathcal{E}_{\mathrm{analog}}\) is unitarily invariant and any deviation in the eigenbasis, which we identify in the second step of our algorithm, will tend to add up with the frequency deviation.  
+
+<|ref|>text<|/ref|><|det|>[[515, 608, 916, 838]]<|/det|>
+In Fig. 2, we illustrate the properties of a single Hamiltonian identification instance in terms of both how well the simulated time evolution fits the experimental data (a,d,e) and how it compares to the targeted Hamiltonian (b) and SPAM (c). We find that most entries of the identified Hamiltonian deviate from the target Hamiltonian by less than \(0.5 \mathrm{MHz}\) with a few entries deviating by around \(1 - 2 \mathrm{MHz}\) . The overall implementation error is around \(1 \mathrm{MHz}\) . The error of the identification method is dominated by the systematic error due to the final ramping phase that is around \(1 \mathrm{MHz}\) for the individual entries, see the SM for details. Small long- range couplings exceeding the statistical error are necessary to fit the data well even when penalizing those entries via regularization. These entries are rooted in the effective rotation by the final ramping before the measurement and within the estimated systematic error.  
+
+<|ref|>text<|/ref|><|det|>[[515, 839, 916, 911]]<|/det|>
+The fit deviation from the data (Fig. 2(e)) exhibits a prominent decrease within the first few nanoseconds of the time evolution. This indicates that the time evolution differs during the initial phase of the experiment as compared to the main phase of the experiment, which we can attribute to the initial pulse
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[85, 66, 487, 124]]<|/det|>
+ramping of the experiment. The identified initial map describing this ramping (Fig. 2(c)) is approximately band- diagonal and deviates from being unitary, indicating fluctuations of the effective ramps between different experiments.  
+
+<|ref|>text<|/ref|><|det|>[[86, 125, 488, 255]]<|/det|>
+We find a larger time- averaged real- time error (Fig. 2(d)) in all data series \(\langle a_{m}\rangle_{\psi_{n}}\) in which \(Q_{4}\) was measured, indicating a measurement error on \(Q_{4}\) . We also observe a deviation between the parameters of the target and identified Hamiltonian in qubits \(Q_{3}\) and \(Q_{4}\) and the coupler between them. Since the deviation of the eigenfrequencies is much smaller than of the full Hamiltonian, we attribute those errors also to a nontrivial final ramping phase at those qubits that leads to a rotated eigenbasis.  
+
+<|ref|>text<|/ref|><|det|>[[86, 256, 488, 504]]<|/det|>
+In Fig. 3, we summarize multiple identification data of this type to benchmark the overall performance of a fixed set of qubits. In panel (a), we show the measured Fourier domain data for 51 different values of the magnetic flux \(b \in [0, 1]\) . In panel (b), we plot the deviation of the frequencies identified from the data. Most implemented frequencies deviate by less than \(1 \mathrm{MHz}\) from their targets. Importantly, the frequency identification is robust against systematic measurement errors. When comparing the analog implementation errors of the full Hamiltonian (Fig. 3(c)) to the corresponding frequency errors, we find an up to fourfold increase in implementation error. The Hamiltonian implementation error is affected by a systematic error due to the non- trivial final ramp. We estimate this error using a linear ramping model; see the SM for details. Since the deviation lies outside of the combined systematic and statistical error bars, our results indicate that the targeted Hamiltonian has not been implemented exactly.  
+
+<|ref|>text<|/ref|><|det|>[[86, 506, 488, 620]]<|/det|>
+In Fig. 3(d), we show the median of the entry- wise deviation of the identified Hamiltonian from its target over all magnetic flux values. Thereby, the ensemble of Hamiltonians defines an overall error benchmark. This benchmark can be associated to the individual constituents of the quantum processor, namely, the qubits, corresponding to diagonal entries of the Hamiltonian deviation, and the couplers, corresponding to the first off- diagonal matrix entries of the deviation.  
+
+<|ref|>text<|/ref|><|det|>[[86, 621, 488, 852]]<|/det|>
+We use this benchmark over an ensemble of two flux values to assess a 27- qubit array of superconducting qubits. To do so, we repeat the analysis reported in Fig. 3 for 5- qubit dynamics on different subsets of qubits and extract average errors of the individual qubits and couplers involved in the dynamics, both in terms of the identified Hamiltonian and the initial and final maps. Summarized in Fig. 4, we find significant variation in the implementation error of different couplers and qubits. While for some qubits the effects of the initial and final maps are negligible, for others they indicate the potential of a significant implementation error. From a practical point of view, such diagnostic data allows to maximally exploit the chip's error for small- scale analog simulation experiments. Let us note that within the error of our method the overall benchmark for the qubits and couplers for 5- qubit dynamics agrees with that of 3- and 4- qubit dynamics.  
+
+<|ref|>text<|/ref|><|det|>[[85, 855, 488, 911]]<|/det|>
+All of the Hamiltonian identification experiments discussed so far (Figs. 2, 3, 4) were implemented on the Sycamore #1 chip. In order to compare these results to implementations on a physically distinct chip with different calibration,  
+
+<|ref|>image<|/ref|><|det|>[[531, 66, 904, 540]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[515, 553, 917, 725]]<|/det|>
+<center>Figure 4. Error map of Hamiltonian implementation across the Sycamore processor. Over the grid of 27 qubits, we randomly choose subsets of connected qubits and couplers of size \(N = 5\) . On each subset we implement two Hamiltonians with \(b = 0, 0.5\) and run the identification algorithm. Two instances are shown in panel (a). For each subset, we compute the deviation of the identified Hamiltonian and initial map from their respective target and assign it to each qubit or coupler involved. Due to overlap of subsets, each qubit or coupler has been involved in at least 5 different choices of subsets. Panels (b) and (c) show the median deviation for the Hamiltonian and initial map implementations, respectively. Panel (d) shows the mean of the sign flips in the identified (diagonal \(\pm 1\) ) final map for each qubit. </center>  
+
+<|ref|>text<|/ref|><|det|>[[516, 753, 916, 911]]<|/det|>
+and to demonstrate the scalability of our method, we implement Hamiltonian identification experiments for an increasing number of qubits on the Sycamore #2 chip. More precisely, for a given number of qubits \(N\) , we implement many different Hamiltonians with quasi- random local potentials, as shown in Fig. 3(c) for \(N = 6\) . We then average the analog implementation errors of the Hamiltonians and frequencies for several system sizes. The results are shown in Fig. 5. Notably, comparing the two different processors, the overall quality of fit does not depend significantly on either the number of qubits or the processor used. This indicates, first, that
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[86, 66, 485, 273]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[85, 287, 488, 460]]<|/det|>
+<center>Figure 5. Analog implementation error scaling and comparing different quantum processors. We measure the analog implementation error of the implemented Hamiltonians (dark red) and their eigenfrequencies (golden) as well as the deviation \((\textstyle \sum_{l = 0}^{L}\| \mathrm{deviation}[\hat{h},\hat{S},\hat{D}_{M}](t_{l})\|_{L_{2}}^{2} / (N^{2}(L + 1)))^{1 / 2}\) of the fit from the experimental data (dark blue) all averaged over implementations of Hamiltonians with quasi-random local potential on an increasing number of qubits on two different quantum processors—Sycamore #1 (circles) and #2 (diamonds). Each point is the mean of the respective quantity over 51 Hamiltonian implementations (21 for \(N = 5\) and 20 for \(N = 14\) on Sycamore #2). The data points at \(N = 6\) on Sycamore #1 summarizes Fig. 3(c). The error bars represent one standard deviation. </center>  
+
+<|ref|>text<|/ref|><|det|>[[86, 488, 488, 676]]<|/det|>
+our reconstruction method works equally well in all scenarios and, second, that both quantum processors implement Hamiltonian time evolution that closely fits our model assumption. We also notice that the overall analog implementation error does not significantly depend on the system size. This signifies that no additional non- local errors are introduced into the system as the size is increased. At the same time, the overall error of Hamiltonian implementations on Sycamore #2 is much worse compared to those on Sycamore #1, indicating that Sycamore #2 was not as well calibrated. Hamiltonian identification thus allows us to meaningfully compare Hamiltonian implementations across different physical systems and system sizes.  
+
+<|ref|>text<|/ref|><|det|>[[87, 682, 488, 912], [515, 66, 916, 181]]<|/det|>
+Conclusion. We have implemented analog simulation of the time- evolution of non- interacting bosonic Hamiltonians with tunable parameters for up to 14 qubit lattice sites. A structure- exploiting learning method allows us to robustly identify the implemented Hamiltonian that governs the time- evolution. To achieve this, we have introduced a new super- resolution algorithm, referred to as tensorESPRIT, for precise robust identification of eigenfrequencies of a Hermitian matrix from noisy snapshots of the one parameter unitary subgroup it generates. Thereby, we diagnose the deviation from the target Hamiltonian and assess the precision of the implementation. We achieve sub- MHz error of the Hamiltonian parameters compared to their targeted values in most implementations. Combining the average performance measures over ensembles of Hamiltonians we associate benchmarks to the components of the superconducting qubit chips that quantify the performance of the hardware on the time evolution and provide specific diagnostic information. Within our Hamiltonian identification framework, we are able to identify SPAM errors due to parameter ramp phases as a severe limitation of the architecture. Importantly, such ramp phases are present in any analog quantum simulation of quenched dynamics. Our results show that minimizing those is crucial for precisely implementing a Hamiltonian.   
+
+<|ref|>text<|/ref|><|det|>[[515, 191, 916, 276]]<|/det|>
+The experimental and computational effort of the identification method scales efficiently in the number of modes of the Hamiltonian. We have also numerically identified the limitations of more direct algorithmic approaches and demonstrated the scalability of our method under empirically derived noise and error models.  
+
+<|ref|>text<|/ref|><|det|>[[515, 286, 916, 415]]<|/det|>
+Generalizing our two- step approach developed here, we expect a polynomial scaling with the dimension of the diagnosed particle sector and therefore remain efficient for diagnosing two- , three- and four- body interactions, thus allowing to build trust in the correct implementation of interacting Hamiltonian dynamics as a whole. From a broader perspective, with this work, we hope to contribute to the development of a machinery for precisely characterizing and thereby improving analog quantum devices.  
+
+<|ref|>sub_title<|/ref|><|det|>[[679, 494, 753, 507]]<|/det|>
+## METHODS  
+
+<|ref|>sub_title<|/ref|><|det|>[[636, 533, 796, 547]]<|/det|>
+### A. Experimental details  
+
+<|ref|>text<|/ref|><|det|>[[515, 573, 916, 789]]<|/det|>
+Details on the quantum processor. We use the Sycamore quantum processor composed of quantum systems arranged in a two- dimensional array. This processor consists of gmon qubits (transmons with tunable coupling) with frequencies ranging from 5 to 7 GHz. These frequencies are chosen to mitigate a variety of error mechanisms such as two- level defects. Our coupler design allows us to quickly tune the qubit- qubit coupling from 0 to \(40+\) MHz. The chip is connected to a superconducting circuit board and cooled down to below \(20\mathrm{mK}\) in a dilution refrigerator. Each qubit has a microwave control line used to drive an excitation and a flux control line to tune the frequency. The processor is connected through filters to room- temperature electronics that synthesize the control signals. We execute single- qubit gates by driving \(25\mathrm{ns}\) microwave pulses resonant with the qubit transition frequency.  
+
+<|ref|>text<|/ref|><|det|>[[515, 826, 916, 912]]<|/det|>
+Experimental read- out and control. The qubits are connected to a resonator that is used to read out the state of the qubit. The state of all qubits can be read simultaneously by using a frequency- multiplexing. Initial device calibration is performed using 'Optimus' [62] where calibration experiments are represented as nodes in a graph.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[155, 66, 417, 80]]<|/det|>
+### B. Details of the identification algorithm  
+
+<|ref|>text<|/ref|><|det|>[[102, 97, 404, 112]]<|/det|>
+Succinctly written, our data model is given by  
+
+<|ref|>equation<|/ref|><|det|>[[100, 121, 485, 152]]<|/det|>
+\[y_{m,n}[l]\coloneqq \langle a_m(t_l)\rangle_{\psi_n} = \frac{1}{2} (M\cdot \exp (-it_lh)\cdot S)_{m,n}, \quad (6)\]  
+
+<|ref|>text<|/ref|><|det|>[[86, 160, 488, 276]]<|/det|>
+where \(m,n = 1,\ldots ,N\) label the distinct time series, \(l =\) \(0,\ldots ,L\) labels the time stamps of the \(L + 1\) data points per time series. The matrices \(S\) and \(M\) are arbitrary invertible linear maps that capture the state preparation and measurement stage, as affected by the ramping of the eigenfrequencies of the qubits and couplers to their target value and back (see Fig. 1). In the experiment, we empirically estimate each such expectation value with 1000 single shots.  
+
+<|ref|>text<|/ref|><|det|>[[86, 276, 488, 336]]<|/det|>
+Our mindset for solving the identification problem is based on the eigendecomposition \(\begin{array}{r}h = \sum_{k = 1}^{N}\lambda_{k} |v_{k}\rangle \langle v_{k}| \end{array}\) of the coefficient matrix \(h\) in terms of eigenvectors \(|v_{k}\rangle\) and eigenvalues \(\lambda_{k}\) . We can write the data (6) in matrix form as  
+
+<|ref|>equation<|/ref|><|det|>[[127, 344, 487, 387]]<|/det|>
+\[y[l] = \frac{1}{2}\exp (-it_lh) = \frac{1}{2}\sum_{k = 1}^{N}\mathrm{e}^{-it_l\lambda_k}|v_k\rangle \langle v_k|, \quad (7)\]  
+
+<|ref|>text<|/ref|><|det|>[[86, 396, 488, 527]]<|/det|>
+where we have dropped \(S\) and \(M\) for the time being. This decomposition suggests a simple procedure to identify the Hamiltonian using Fourier data analysis. From the matrixvalued time series data \(y[l]\) (7), we identify the Hamiltonian coefficient matrix \(h\) in two steps. First, we determine the eigenfrequencies of \(h\) . Second, we identify the eigenbasis of \(h\) . To achieve those identification tasks with the largest possible robustness to error, it is key to exploit all available structure at hand.  
+
+<|ref|>text<|/ref|><|det|>[[86, 531, 488, 662]]<|/det|>
+Step 1: Frequency extraction. In order to robustly estimate the spectrum, we exploit that the signal is sparse in Fourier space. This structure allows us to substantially denoise the signal and achieve super- resolution beyond the Nyquist limit [63, 64]. A candidate algorithm for this task, suitable for scalar time- series, is the ESPRIT algorithm, which comes with rigorous recovery guarantees [55, 56]. To extract the Hamiltonian spectrum from the matrix time- series \(y[l]\) , we apply ESPRIT to the trace of the data series (for \(S = M = \mathbb{1}\) )  
+
+<|ref|>equation<|/ref|><|det|>[[125, 669, 488, 713]]<|/det|>
+\[F[l]\coloneqq \mathrm{Tr}[y[l]] = \sum_{m = 1}^{N}y_{m,m}[l] = \frac{1}{2}\sum_{k = 1}^{N}\mathrm{e}^{-it_l\lambda_k}. \quad (8)\]  
+
+<|ref|>text<|/ref|><|det|>[[86, 723, 488, 797]]<|/det|>
+The drawback of this approach is that if the spectrum of the Hamiltonian is sufficiently crowded, which will happen for large \(N\) , the Fourier modes in \(F[l]\) become indistinguishable and ESPRIT fails to identify the frequencies. In particular, ESPRIT is not able to identify degeneracies in the spectrum.  
+
+<|ref|>text<|/ref|><|det|>[[86, 798, 488, 912]]<|/det|>
+To overcome this issue and obtain a truly scalable learning procedure applicable to degenerate spectra, we develop a new algorithm coined tensorESPRIT, which extends the ideas of ESPRIT to the case of a matrix time- series using tensor network techniques. Using tensorESPRIT also improves the robustness of frequency estimation to SPAM errors. For practical Hamiltonians, tensorESPRIT becomes necessary for systems with \(N \gtrsim 12\) .  
+
+<|ref|>text<|/ref|><|det|>[[515, 66, 917, 240]]<|/det|>
+tensorESPRIT (ESPRIT) comprises of a denoising step, in which the rank of the Hankel tensor (matrix) of the data is limited to its theoretical value. Subsequently, rotational invariance of the data is used to compute a matrix from the denoised Hankel tensor (matrix), the spectrum of which has a simple relation to the spectrum of \(h\) . In the case of ESPRIT, this amounts to a multiplication of the denoised Hankel matrix by a pseudoinverse of its shifted version. Contrastingly, tensorESPRIT uses a sampling procedure to contract certain sub- matrices of the denoised Hankel tensor with the pseudoinverse of other sub- matrices. Details on both algorithms can be found in the SM.  
+
+<|ref|>text<|/ref|><|det|>[[515, 243, 917, 444]]<|/det|>
+Step 2: Eigenspace identification. To identify the eigenspaces of the Hamiltonian, we use the eigenfrequencies found in Step 1 to fix the oscillating part of the dynamics in Eq. (7). What remains is the problem of finding the eigenspaces \(|v_k\rangle \langle v_k|\) from the data. This problem is a nonconvex inverse quadratic problem, subject to orthogonality of the eigenspaces, as well as the constraint that the resulting Hamiltonian matrix respects the connectivity of the superconducting architecture. Formally, we denote the a priori known support set of the Hamiltonian matrix as \(\Omega\) , so that we can write the support constraint as \(h_{\Omega} = 0\) , where \(\bar{\Omega}\) denotes the complement of \(\Omega\) and subscripting a matrix with a support set restricts the matrix to this set. We can cast this problem into the form of a least- squares optimization problem  
+
+<|ref|>equation<|/ref|><|det|>[[521, 454, 912, 560]]<|/det|>
+\[\underset {\{ |v_k\rangle \}}{\mathrm{minimise}}\quad \sum_{l = 0}^{L}\left\| y[l] - \sum_{k}\mathrm{e}^{-\mathrm{i}\lambda_k t_l}|v_k\rangle \langle v_k|\right\|_{\ell_2}^2,\] \[\mathrm{subject~to}\quad \langle v_m|v_n\rangle = \delta_{m,n},\left(\sum_k\lambda_k|v_k\rangle \langle v_k|\right)_{\bar{\Omega}} = 0,\]  
+
+<|ref|>text<|/ref|><|det|>[[515, 572, 917, 645]]<|/det|>
+equipped with non- convex constraints enforcing orthogonality, and the quadratic constraint restricting the support. In order to approximately enforce the support constraint, we make use of regularization [60]. It turns out that this can be best achieved by adding a term [65, App. A]  
+
+<|ref|>equation<|/ref|><|det|>[[620, 653, 916, 700]]<|/det|>
+\[\mu \left\| \left(\sum_k\lambda_k|v_k\rangle \langle v_k|\right)_{\bar{\Omega}}\right\|_{\ell_2} \quad (10)\]  
+
+<|ref|>text<|/ref|><|det|>[[515, 709, 917, 796]]<|/det|>
+to the objective function (9), where \(\mu > 0\) is a parameter weighting the violation of the support constraint. We then solve the resulting minimization problem by using a conjugate gradient descent on the manifold of the orthogonal group [57, 66], see also the recent work [67- 69] for the use of geometric optimization for quantum characterization.  
+
+<|ref|>text<|/ref|><|det|>[[515, 797, 917, 912]]<|/det|>
+Without the support constraint this gives rise to an optimization algorithm that converges well, as shown in the SM. However, the regularization term makes the optimization landscape rugged as it introduces an entry- wise constraint that is skew to the orthogonal manifold. To deal with this, we consecutively ramp up \(\mu\) until the algorithm does not converge anymore in order to find the Hamiltonian that best approximates the support constraint while being a proper solution of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[86, 66, 488, 167]]<|/det|>
+the optimization problem. For example, for the data in Fig. 2 the value of \(\mu\) is 121. In order to avoid that we identify a Hamiltonian from a local minimum of the rugged landscape, we only accept Hamiltonians that achieve a total fit of the experimental data within a \(5\%\) margin of the fit quality of the unregularized recovery problem, and use the Hamiltonian recovered without the regularization otherwise.  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 201, 480, 215]]<|/det|>
+### C. Robustness to state preparation and measurement errors  
+
+<|ref|>text<|/ref|><|det|>[[86, 235, 488, 336]]<|/det|>
+The experimental design requires a ramping phase of the qubit and coupler frequencies from their idle location to the desired target Hamiltonian and back for the measurement. In effect, the data model (6) includes time- independent linear maps \(M\) and \(S\) that are applied at the beginning and end of the Hamiltonian time- evolution. The maps affect both the frequency extraction and the eigenspace reconstruction.  
+
+<|ref|>text<|/ref|><|det|>[[86, 337, 488, 467]]<|/det|>
+For the frequency extraction using ESPRIT, the Fourier coefficients of the trace signal \(F[l]\) become \(\langle v_k|SM|v_k\rangle\) . While the frequencies remain unchanged the Fourier coefficients now deviate from unity, significantly impairing the noise- robustness of the frequency identification. This effect is still present, albeit weaker, in tensorESPRIT, in the case of non- unitary SPAM errors. The eigenspace reconstruction is affected much more severely and requires careful consideration, as detailed below and in the SM.  
+
+<|ref|>text<|/ref|><|det|>[[86, 468, 488, 541]]<|/det|>
+Ramp removal via pre- processing. We can remove either the initial map \(S\) or the final map \(M\) from the data. To remove \(S\) , we apply the pseudoinverse \((\cdot)^{+}\) of the data \(y[l_0]\) at a fixed time \(t_{l_0}\) to the entire (time- dependent) data series in matrix form. For invertible \(S\) and \(M\) this gives rise to  
+
+<|ref|>equation<|/ref|><|det|>[[123, 551, 487, 618]]<|/det|>
+\[\begin{array}{l}{y^{(l_0)}[l] = y[l](y[l_0])^+}\\ {= \sum_{k = 1}^{N}\mathrm{e}^{-\mathrm{i}\lambda_k(t_l - t_{l_0})}M|v_k\rangle \langle v_k|M^{-1}.} \end{array} \quad (11)\]  
+
+<|ref|>text<|/ref|><|det|>[[86, 631, 488, 805]]<|/det|>
+The caveat of this approach is that the noise that affected \(y[l_0]\) is now present in every entry of the new data series \(y^{(l_0)}\) . We can remedy this effect by using several time points \(y[l_0]\) for the inversion. To this end, we compute the concatenation of data series for different choices of \(l_0\) , e.g., for every \(s\) data points \(0, s, 2s, \ldots , \lfloor L / s \rfloor s\) giving rise to new data \(y_{\mathrm{total}, s} = (y^{(0)}, y^{(s)}, y^{(2s)}, \ldots , y^{( \lfloor L / s \rfloor s)}) \in \mathbb{C}^{\lfloor L / s \rfloor L}\) . If the data suffers from drift errors, it is also beneficial to restrict each data series \(y^{(l_0)}\) to entries \(y^{(l_0)}[\kappa ]\) with \(\kappa \in [l_0 - w, l_0 + w]\) , i.e., the entries in a window of size \(w\) around \(l_0\) . In practice, we use \(s = 1\) and \(w = 50\) for the reconstructions on Sycamore #1, and \(s = 1, w = L\) for those on Sycamore #2.  
+
+<|ref|>text<|/ref|><|det|>[[86, 807, 488, 907]]<|/det|>
+As we argue below, the final map \(M\) is nearly diagonal here. Hence, we can use \(y_{\mathrm{total}, s}\) from Eq. (9) and it is justified to apply the support constraint in the eigenspace reconstruction step. However, the eigenspace reconstruction will suffer from systematic errors due to the final map, even in the case when it is nearly diagonal. Below, we explain a method to partially remove this error.  
+
+<|ref|>text<|/ref|><|det|>[[515, 66, 917, 271]]<|/det|>
+Tomographic estimate of \(S\) and \(M\) . The systematic error in the reconstructed Hamiltonian eigenbasis can be expressed as an orthogonal rotation \(D_M\) from the eigenbasis that is actually implemented. Due to the gauge freedom in the model (6), we cannot hope to identify \(D_M\) fully without additional assumptions. However, as elaborated on in the SM, we can find a diagonal orthogonal estimate \(\hat{D}_M\) of the true correction \(D_M\) and hence remove a sign of the systematic error. To this end, we assume that the experimental implementation of the target Hamiltonian does not flip the sign in the hopping terms and remedy the sign of systematic error due to the final map by fitting a diagonal orthogonal rotation of the Hamiltonian eigenbasis \(\hat{D}_M\) that minimizes the implementation error. We update the reconstructed Hamiltonian to  
+
+<|ref|>equation<|/ref|><|det|>[[662, 280, 916, 300]]<|/det|>
+\[\hat{h} = \hat{D}_M\hat{h}\hat{D}_M, \quad (12)\]  
+
+<|ref|>text<|/ref|><|det|>[[515, 311, 917, 374]]<|/det|>
+where \(\begin{array}{r}\tilde{h} = \sum_k\lambda_k|v_k\rangle \langle v_k| \end{array}\) and \(\{\vert v_k\rangle \}\) is the eigenbasis obtained by solving the problem (9), and use \(\hat{D}_M\) as an estimate of \(M\) . We can now obtain a tomographic estimate of the initial map through  
+
+<|ref|>equation<|/ref|><|det|>[[600, 384, 916, 430]]<|/det|>
+\[\hat{S}\coloneqq \frac{2}{L + 1}\sum_{l = 0}^{L}\exp [\mathrm{i}t_l\hat{h}]\hat{D}_M y[l]. \quad (13)\]  
+
+<|ref|>text<|/ref|><|det|>[[515, 441, 916, 502]]<|/det|>
+The recovered model \((\hat{h},\hat{S},\hat{D}_M)\) gives good prediction accuracy on simulated data, as demonstrated in Fig. 7 and in the SM, and fits well the experimental data, as demonstrated in Figs. 2, 5 and 6.  
+
+<|ref|>text<|/ref|><|det|>[[515, 504, 917, 589]]<|/det|>
+Imbalance between initial and final ramping phase. As explained above, the pre- processing step allow us to remove either the initial map \(S\) or final map \(M\) from the data, while we can only find a diagonal orthogonal estimate of the remaining map. A priori it is unclear which one of the two maps should be removed in order to reduce the systematic error more.  
+
+<|ref|>text<|/ref|><|det|>[[515, 590, 917, 780]]<|/det|>
+We have already treated the initial and final ramping phases on a different footing, however. The reason for this is rooted in the specifics of the ramping of the couplers compared to the qubits. The couplers need to be ramped from their idle frequencies to provide the desired target frequencies of \(20\mathrm{MHz}\) . This is why we expect the time scale of the initial ramping to be mainly determined by the couplers, namely the delay until they arrive around the target frequency and the time it takes to stabilize at the target frequency. In contrast, the final ramping map becomes effectively diagonal as soon as the couplers are again out of the MHz regime. We therefore expect that the initial map has a sizeable non- diagonal orthogonal component, whereas the final map is approximately diagonal.  
+
+<|ref|>text<|/ref|><|det|>[[515, 781, 917, 913]]<|/det|>
+We build trust in this assumption using experimental data in Fig. 6. We observe that the deviation of the orthogonal part \(\hat{O}_S\) of the identified initial map \(\hat{S}\) from its projection \(\hat{D}_S\) to diagonal orthogonal matrices is much larger than the corresponding deviation for the final map (Fig. 6(a)). Moreover, both the root- mean- square fit of the data (Fig. 6(c)) and the analog implementation error of the identified Hamiltonian with its target (Fig. 6(b)) are significantly improved when removing the initial ramp, as compared to removing the final
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[84, 63, 485, 344]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[85, 357, 488, 479]]<|/det|>
+<center>Figure 6. Initial ramp removal versus final ramp removal. We identify Hamiltonians of a set of 5-qubit Hamiltonians with Hofstadter butterfly potentials \(\mu_{q} = 20\cos (2\pi qb)\) MHz for qubits \(q = 1,\ldots ,5\) and flux value \(b\) in without regularization. (a) Deviation of the orthogonal part \(\hat{O}_S\) ( \(\hat{O}_M\) ) of the identified initial map \(\hat{S}\) (final map \(\hat{M}\) ) from the closest diagonal orthogonal matrix \(\hat{D}_S\) ( \(\hat{D}_M\) ). (b) Analog implementation error of the corresponding identified Hamiltonians \(\hat{h}_S\) ( \(\hat{h}_M\) ). (c) Total root-mean-square deviation of the time series data from the Hamiltonian fit. </center>  
+
+<|ref|>text<|/ref|><|det|>[[86, 507, 487, 607]]<|/det|>
+ramp. This indicates that \(S\) induces a larger systematic error than \(\hat{M}\) . Correspondingly, it is indeed more advantageous to remove the initial map in the pre- processing and fit the final map with a diagonal orthogonal matrix, validating the approach taken here. In the SM, we provide further numerical evidence that this approach leads to small systematic errors and recovers a model with good predictive power.  
+
+<|ref|>sub_title<|/ref|><|det|>[[182, 648, 393, 662]]<|/det|>
+### D. Benchmarking the algorithm  
+
+<|ref|>text<|/ref|><|det|>[[86, 680, 488, 911], [515, 388, 916, 445]]<|/det|>
+We benchmark our identification algorithm against more direct approaches in numerical simulations including models for statistical and systematic errors in the SM VI. We find that, indeed, already for small system sizes, the regularized manifold optimization algorithm developed here features an improved robustness against state preparation and measurement errors compared to (post- projected) linear inversion. For intermediate system sizes ( \(N > 10\) ), exploiting structure in the recovery algorithm then becomes an imperative. In particular, for larger system sizes the eigenspectrum of the Hamiltonian becomes unavoidably narrower spaced. We find that on instances of the Harper Hamiltonian studied here linear inversion approaches cannot be applied at all for \(N > 20\) . Regularized conjugate gradient decent in contrast yields good recovery performance even for larger systems. The same limitations apply to a direct Fourier analysis of the cumulative time series data using ESPRIT, as described above. For different families of Hamiltonians, we find that above a system size of \(N \approx 20\) tensorESPRIT still consistently recovers the frequency spectrum, while the ESPRIT algorithm fails to do so.  
+
+<|ref|>image<|/ref|><|det|>[[520, 61, 912, 210]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[515, 226, 916, 360]]<|/det|>
+<center>Figure 7. Numerical benchmarks for larger system sizes. Recovery error of frequencies (golden) and Hamiltonians (red) from simulated time series averaged over 20 instances of Harper Hamiltonians for different system sizes. The error bars represent one standard deviation. The evolution is simulated for up to \(0.6 \mu s\) and sampled at a rate of \(250 \mathrm{MHz}\) . Statistical noise is simulated using \(10^{3}\) shots per expected value and SPAM is modeled by using randomly chosen idle qubit and coupler frequencies, linear ramping of \(1.5 \mathrm{GHz / s}\) padded by \(0.05 \mathrm{ns}\) . The fitting error of the time series is depicted in blue, right \(y\) -axis. We refer to the SM, Sec. VII A for details. </center>   
+
+<|ref|>text<|/ref|><|det|>[[515, 446, 916, 617]]<|/det|>
+Using structure not only allows our algorithm to denoise the data and achieve error robustness, it also makes precise Hamiltonian identification possible even with the number of measurements dramatically reduced in the spirit of compressed sensing. As described above, the number of measurements scales quadratically with the system size. We find that using the conjugate gradient algorithm the identification procedure reliably recovers Hamiltonians even when it has access to only about \(3\%\) of the measurements. In this regime, the linear inverse problem of finding the eigenvectors is underdetermined. Thus, the required experimental resources can be significantly reduced for large system sizes.  
+
+<|ref|>text<|/ref|><|det|>[[515, 618, 916, 692]]<|/det|>
+To demonstrate our method's scalability, Fig. 7 shows the recovery performance of the structure- exploiting algorithm on simulated data under realistic models for SPAM errors and with finite measurement statistics in the regime where the baseline approaches could not be applied anymore.  
+
+<|ref|>text<|/ref|><|det|>[[515, 693, 916, 820]]<|/det|>
+As detailed in the SM, tensorESPRIT has computational complexity in \(\mathcal{O}(L^2 N^3)\) . It is not straight- forward to bound the computational complexity of the conjugate gradient descent, as it depends on the required precision of the matrix exponential and the number of descent steps until convergence. The entire identification algorithm consumes \(\mathcal{O}(LN^2)\) memory. In practice, we find that the algorithm can be easily deployed on a consumer- grade laptop computer, e.g. reconstructing Hamiltonians of size \(N = 50\) in around 5 minutes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[649, 850, 783, 863]]<|/det|>
+### E. Error estimation  
+
+<|ref|>text<|/ref|><|det|>[[515, 881, 916, 911]]<|/det|>
+We here discuss how we estimate the systematic and statistical contributions to the error on the identified Hamiltonian \(\hat{h}\)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[86, 65, 487, 258]]<|/det|>
+and initial map \(\hat{S}\) . Note that the impact of the systematic error on predicting results of experiments with the same initial and final ramps is reduced due to the gauge invariance of the model (6). Due to this freedom, some of the error in identifying \(\hat{D}_M \sim M\) gets accounted for by a corresponding error in the identification \(\hat{h} \sim h\) and \(\hat{S} \sim S\) in expressions of the type \(\hat{M} e^{- i t \hat{h}} \hat{S}\) . This prediction error can be further decreased by running the algorithm twice—removing the initial map in the first run and the final map in the second run, using the first ramp estimates to partially remove the ramps from the data before running the second iteration of the identification. This procedure is detailed and supported by numerical evidence in the SM.  
+
+<|ref|>text<|/ref|><|det|>[[86, 259, 487, 374]]<|/det|>
+Systematic error: Final ramp effect estimation. In order to estimate the magnitude of the systematic error that is induced by the non- trivial final map, we use a linear model of the final ramping phase with a constant ramping speed and constant wait time between the coupler and qubit ramping. We detail and present validation of this ramping model with a separate experiment in the SM, where we also provide empirical estimates of the model parameters.  
+
+<|ref|>text<|/ref|><|det|>[[86, 375, 487, 466]]<|/det|>
+Given a Hamiltonian matrix \(\hat{h}\) and the initial ramp \(\hat{S}\) obtained from experimental data, we recover the Hamiltonian matrix \(\hat{h}'\) from data simulated using the model \((\hat{h}, \hat{S}, M)\) , where \(M\) is the final ramp given by our ramping model. We use \(|f(\hat{h}) - f(\hat{h}')|\) as an estimate of the systematic error on quantities of the form \(f(\hat{h}) \in \mathbb{R}\) .  
+
+<|ref|>text<|/ref|><|det|>[[86, 471, 487, 544]]<|/det|>
+Statistical error: Bootstrapping. We estimate the effect of finite measurement statistics on the Hamiltonian estimate that is returned by the identification method via parametric bootstrapping. To this end, we simulate time series data with statistical noise using Haar- random unitaries \(S\) as initial ramps,  
+
+<|ref|>text<|/ref|><|det|>[[515, 65, 916, 95]]<|/det|>
+the identified Hamiltonian \(\hat{h}\) and final ramp \(M = \mathbb{1}\) , as detailed in the SM.  
+
+<|ref|>text<|/ref|><|det|>[[515, 99, 916, 229]]<|/det|>
+Acknowledgements. We acknowledge contributions from Charles Neill, Kostyantyn Kechedzhi, and Alexander Korotkov to the calibration procedure used in this analog approach. We would like to thank Christian Krumnow, Benjamin Chiaro, Alireza Seif, Markus Heinrich, and Juani Bermejo- Vega for fruitful discussions in early stages of the project. The hardware used for this experiment was developed by the Google Quantum AI hardware team, under the direction of Anthony Megrant, Julian Kelly and Yu Chen.  
+
+<|ref|>text<|/ref|><|det|>[[515, 230, 916, 374]]<|/det|>
+Funding. D. H. acknowledges funding from the U.S. Department of Defense through a QuICS Hartree fellowship. This work has been supported by the BMBF (DAQC), for which it provides benchmarking tools for analog- digital superconducting quantum devices, as well as by the DFG (specifically EI 519 20- 1 on notions of Hamiltonian learning, but also CRC 183 and GRK 2433 Deadalus). We have also received funding from the European Union's Horizon2020 research and innovation programme (PASQuanS2) on programmable quantum simulators, and the ERC (DebuQC).  
+
+<|ref|>text<|/ref|><|det|>[[515, 377, 916, 479]]<|/det|>
+Author contributions. D.H. and I.R. conceived of the Hamiltonian identification algorithm. J.F. conceived of the tensorESPRIT algorithm. D.H., I.R. and J.F. analyzed the experimental data and benchmarked the identification algorithm. P.R. took the experimental data. D.H. and I.R. wrote the initial manuscript. All authors contributed to discussions and writing the final manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[515, 482, 916, 510]]<|/det|>
+Data and materials availability. The experimental data is available from the authors upon reasonable request.  
+
+<|ref|>text<|/ref|><|det|>[[515, 514, 916, 541]]<|/det|>
+Conflict of interest. The authors declare no conflict of interest.  
+
+<|ref|>text<|/ref|><|det|>[[90, 600, 490, 904]]<|/det|>
+[1] R. P. Feynman, Simulating Physics with computers, Int. J. Theor. Phys. 21, 467 (1982). [2] S. Lloyd, Universal quantum simulators, Science 273, 1073 (1996). [3] I. Bloch, J. Dalibard, and S. Nascimbene, Quantum simulations with ultracold quantum gases, Nature Phys. 8, 267 (2012). [4] R. Blatt and C. F. Roos, Quantum simulations with trapped ions, Nature Phys. 8, 277 (2012). [5] J. Smith, A. Lee, P. Richerme, B. Neyenhuis, P. W. Hess, P. Hauke, M. Heyl, D. Huse, and C. Monroe, Many- body localization in a quantum simulator with programmable random disorder, Nature Phys. 12, 907 (2016). [6] S. Ebadi, T. T. Wang, H. Levine, A. Keesling, G. Semeghini, A. Omran, D. Bluvstein, R. Samajdar, H. Pichler, W. W. Ho, S. Choi, S. Sachdev, M. Greiner, V. Vuletic, and M. D. Lukin, Quantum phases of matter on a 256- atom programmable quantum simulator, Nature 595, 227 (2021). [7] D. Aharonov and M. Ben- Or, Fault- tolerant quantum computation with constant error rate, SIAM J. Comput. 38, 1207 (2008). [8] E. Derbyshire, J. Y. Malo, A. J. Daley, E. Kashefi, and P. Wallden, Randomized benchmarking in the analogue setting, Quantum Sci. Technol. 5, 034001 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[516, 600, 916, 904]]<|/det|>
+[9] R. Shaffer, E. Megidish, J. Broz, W.- T. Chen, and H. Haffner, Practical verification protocols for analog quantum simulators, npj Quant. Inf. 7, 1 (2021). [10] J. Helsen, S. Nezami, M. Reagor, and M. Walter, Matchgate benchmarking: Scalable benchmarking of a continuous family of many- qubit gates, Quantum 6, 657 (2020). [11] S. G. Schirmer, A. Kolli, and D. K. L. Oi, Experimental Hamiltonian identification for controlled two- level systems, Phys. Rev. A 69, 050306 (2004). [12] J. H. Cole, S. G. Schirmer, A. D. Greentree, C. J. Wellard, D. K. L. Oi, and L. C. L. Hollenberg, Identifying an experimental two- state Hamiltonian to arbitrary accuracy, Phys. Rev. A 71, 062312 (2005). [13] J. H. Cole, A. D. Greentree, D. K. L. Oi, S. G. Schirmer, C. J. Wellard, and L. C. L. Hollenberg, Identifying a two- state Hamiltonian in the presence of decoherence, Phys. Rev. A 73, 062333 (2006). [14] J. H. Cole, S. J. Devitt, and L. C. L. Hollenberg, Precision characterization of two- qubit Hamiltonians via entanglement mapping, J. Phys. A 39, 14649 (2006). [15] S. G. Schirmer, D. K. L. Oi, and S. J. Devitt, Physics- based mathematical models for quantum devices via experimental system identification, J. Phys. Conf. Ser. 107, 012011 (2008).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[82, 68, 490, 910]]<|/det|>
+[16] S. G. Schirmer and D. K. L. Oi, Two- qubit Hamiltonian tomography by Bayesian analysis of noisy data, Phys. Rev. A 80, 022333 (2009).[17] D. K. L. Oi and S. G. Schirmer, Quantum system characterization with limited resources, Phil. Trans. R. Soc. A 370, 5386 (2012).[18] D. Burgarth, K. Maruyama, and F. Nori, Coupling strength estimation for spin chains despite restricted access, Phys. Rev. A 79, 020305 (2009).[19] D. Burgarth, K. Maruyama, and F. Nori, Indirect quantum tomography of quadratic Hamiltonians, New J. Phys. 13, 013019 (2011).[20] D. Burgarth and K. Maruyama, Indirect Hamiltonian identification through a small gateway, New J. Phys. 11, 103019 (2009).[21] C. Di Franco, M. Paternostro, and M. S. Kim, Hamiltonian Tomography in an access- limited setting without state initialization, Phys. Rev. Lett. 102, 187203 (2009).[22] M. Wiesniak and M. Markiewicz, Finding traps in non- linear spin arrays, Phys. Rev. A 81, 032340 (2010).[23] D. Burgarth and K. Yuasa, Quantum system identification, Phys. Rev. Lett. 108, 080502 (2012).[24] J. Zhang and M. Sarovar, Quantum Hamiltonian identification from measurement time traces, Phys. Rev. Lett. 113, 080401 (2014).[25] A. Sone and P. Cappellaro, Exact dimension estimation of interacting qubit systems assisted by a single quantum probe, Phys. Rev. A 96, 062334 (2017).[26] J. R. Garrison and T. Grover, Does a single eigenstate encode the full Hamiltonian? Phys. Rev. X 8, 021026 (2018).[27] X.- L. Qi and D. Ranard, Determining a local Hamiltonian from a single eigenstate, Quantum 3, 159 (2019).[28] E. Chertkov and B. K. Clark, Computational inverse method for constructing spaces of quantum models from wave functions, Phys. Rev. X 8, 031029 (2018).[29] E. Bairey, I. Arad, and N. H. Lindner, Learning a local Hamiltonian from local measurements, Phys. Rev. Lett. 122, 020504 (2019).[30] E. Bairey, C. Guo, D. Poletti, N. H. Lindner, and I. Arad, Learning the dynamics of open quantum systems from their steady states, New J. Phys. 22, 032001 (2020).[31] T. J. Evans, R. Harper, and S. T. Flammia, Scalable Bayesian Hamiltonian learning, (2019), arxiv:1912.07636. [32] Z. Li, L. Zou, and T. H. Hsieh, Hamiltonian tomography via quantum quench, Phys. Rev. Lett. 124, 160502 (2020).[33] A. Czerwinski, Hamiltonian tomography by the quantum quench protocol with random noise, Phys. Rev. A 104, 052431 (2021).[34] A. Elben, S. T. Flammia, H.- Y. Huang, R. Kueng, J. Preskill, B. Vermersch, and P. Zoller, The randomized measurement toolbox, Nature Rev. Phys. 5, 9 (2023), arxiv:2203.11374. [35] A. Valenti, E. van Nieuwenburg, S. Huber, and E. Greplova, Hamiltonian learning for quantum error correction, Phys. Rev. Res. 1, 033092 (2019).[36] P. Bienias, A. Seif, and M. Hafezi, Meta Hamiltonian learning, arXiv:2104.04453. [37] S. Krastanov, S. Zhou, S. T. Flammia, and L. Jiang, Stochastic estimation of dynamical variables, Quantum Sci. Technol. 4, 035003 (2019).[38] L. Che, C. Wei, Y. Huang, D. Zhao, S. Xue, X. Nie, J. Li, D. Lu, and T. Xin, Learning quantum Hamiltonians from single- qubit measurements, Phys. Rev. Res. 3, 023246 (2021).[39] F. Wilde, A. Kshetrimayum, I. Roth, D. Hangleiter, R. Sweke, and J. Eisert, Scalably learning quantum many- body Hamiltonians from dynamical data, (2022), arxiv:2209.14328.  
+
+<|ref|>text<|/ref|><|det|>[[513, 70, 920, 900]]<|/det|>
+[40] W. Yu, J. Sun, Z. Han, and X. Yuan, Robust and efficient Hamiltonian learning, Quantum 7, 1045 (2023), arxiv:2201.00190. [41] H.- Y. Huang, Y. Tong, D. Fang, and Y. Su, Learning many- body Hamiltonians with Heisenberg- limited scaling, Phys. Rev. Lett. 130, 200403 (2023).[42] H. Li, Y. Tong, H. Ni, T. Gefen, and L. Ying, Heisenberg- limited Hamiltonian learning for interacting bosons, (2023), arxiv:2307.04690. [43] S. T. Flammia and J. J. Wallman, Efficient Estimation of Pauli Channels, ACM Transactions on Quantum Computing 1, 3:1 (2020).[44] R. Harper, S. T. Flammia, and J. J. Wallman, Efficient Learning of Quantum Noise, Nature Physics 16, 1184 (2020).[45] E. H. Lapasar, K. Maruyama, D. Burgarth, T. Takui, Y. Kondo, and M. Nakahara, Estimation of coupling constants of a three- spin chain: Case study of Hamiltonian tomography with NMR, New J. Phys. 14, 013043 (2012).[46] S.- Y. Hou, H. Li, and G.- L. Long, Experimental quantum Hamiltonian identification from measurement time traces, Science Bulletin 62, 863 (2017).[47] X. Chen, Y. Li, Z. Wu, R. Liu, Z. Li, and H. Zhou, Experimental realization of Hamiltonian tomography by quantum quenches, Phys. Rev. A 103, 042429 (2021).[48] D. Zhao, C. Wei, S. Xue, Y. Huang, X. Nie, J. Li, D. Ruan, D. Lu, T. Xin, and G. Long, Characterizing quantum simulations with quantum tomography on a spin quantum simulator, Phys. Rev. A 103, 052403 (2021).[49] E. Flurin, L. S. Martin, S. Hacohen- Gourgy, and I. Siddiqi, Using a recurrent neural network to reconstruct quantum dynamics of a superconducting qubit from physical observations, Phys. Rev. X 10, 011006 (2020).[50] G. O. Samach, A. Greene, J. Borregaard, M. Christandl, D. K. Kim, C. M. McNally, A. Melville, B. M. Niedzielski, Y. Sung, D. Rosenberg, M. E. Schwartz, J. L. Yoder, T. P. Orlando, J. I.- J. Wang, S. Gustavsson, M. Kjaergaard, and W. D. Oliver, Lindblad tomography of a superconducting quantum processor, Phys. Rev. Appl. 18, 064056 (2022).[51] C. Kokail, R. van Bijnen, A. Elben, B. Vermersch, and P. Zoller, Entanglement Hamiltonian tomography in quantum simulation, Nature Phys. 17, 936 (2021).[52] C. Kokail, B. Sundar, T. V. Zache, A. Elben, B. Vermersch, M. Dalmonte, R. van Bijnen, and P. Zoller, Quantum variational learning of the entanglement Hamiltonian, Phys. Rev. Lett. 127, 170501 (2021).[53] M. K. Joshi, C. Kokail, R. van Bijnen, F. Kranzl, T. V. Zache, R. Blatt, C. F. Roos, and P. Zoller, Exploring large- scale entanglement in quantum simulation, Nature, 1 (2023).[54] R. Roy, A. Paulraj, and T. Kailath, Estimation of signal parameters via rotational invariance techniques- ESPRIT, in MILCOM 1986- IEEE Military Communications Conference: Communications- Computers: Teamed for the 90's, Vol. 3 (IEEE, 1986) pp. 41- 6. [55] A. Fannjiang, Compressive spectral estimation with single- snapshot esprit: Stability and resolution, arXiv:1607.01827. [56] W. Li, W. Liao, and A. Fannjiang, Super- resolution limit of the ESPRIT algorithm, IEEE Trans. Inf. Th. 66, 4593 (2020), arXiv:1905.03782. [57] T. Abrudan, J. Eriksson, and V. Koivunen, Conjugate gradient algorithm for optimization under unitary matrix constraint, Signal Processing 89, 1704 (2009).[58] I. Carusotto, A. A. Houck, A. J. Kollar, P. Roushan, D. I. Schuster, and J. Simon, Photonic materials in circuit quantum electrodynamics, Nature Phys. 16, 268- 279 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[85, 66, 490, 290]]<|/det|>
+[59] F. Yan, P. Krantz, Y. Sung, M. Kjaergaard, D. Campbell, J. I. J. Wang, T. P. Orlando, S. Gustavsson, and W. D. Oliver, A tunable coupling scheme for implementing high- fidelity two- qubit gates, Phys. Rev. Appl. 10, 054062 (2018).[60] P. Bühlmann and S. V. D. Geer, Statistics for high- dimensional data, Springer Series in Statistics, Vol. 9 (Springer, Berlin, 2011).[61] D. R. Hofstadter, Energy levels and wave functions of Bloch electrons in rational and irrational magnetic fields, Phys. Rev. B 14, 2239 (1976).[62] J. Kelly, P. O'Malley, M. Neeley, H. Neven, and J. M. Martinis, Physical qubit calibration on a directed acyclic graph, arXiv:1803.03226. [63] E. J. Candès and C. Fernandez- Granda, Super- resolution from noisy data, J. Fourier An. App. 19, 1229 (2013).[64] E. J. Candès and C. Fernandez- Granda, Towards a mathematical theory of super- resolution, Comm. Pure App. Math. 67, 906 (2014).[65] D. Hangleiter, I. Roth, D. Nagaj, and J. Eisert, Easing the Monte Carlo sign problem, Science Adv. 6, eabb8341 (2020).[66] A. Edelman, T. A. Arias, and S. T. Smith, The geometry of algorithms with orthogonality constraints, SIAM J. Matr. Ana. App. 20, 303 (1998).[67] I. A. Luchnikov, A. Ryzhov, S. N. Filippov, and H. Ouerdane, QGOpt: Riemannian optimization for quantum technologies, arXiv:2011.01894. [68] I. Luchnikov, M. Krechetov, and S. Filippov, Riemannian geometry and automatic differentiation for optimization problems of quantum physics and quantum technologies, New J. Phys. 23, 073006 (2021).[69] I. Roth, J. Wilkens, D. Hangleiter, and J. Eisert, Semidevice- dependent blind quantum tomography, Quantum 7, 1053 (2023).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[43, 92, 768, 112]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 323, 150]]<|/det|>
+- Hlearningsupplemental.pdf
+
+<--- Page Split --->

preprint/preprint__7ea08d3a6b99d16e7f9a32724c54deb41c2325e4e80d5feab1f63b193d958db8/images_list.json

@@ -0,0 +1,130 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        50,
+        60,
+        808,
+        750
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        45,
+        53,
+        800,
+        789
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        50,
+        55,
+        680,
+        789
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5",
+    "footnote": [],
+    "bbox": [
+      [
+        45,
+        50,
+        617,
+        790
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "B",
+    "footnote": [],
+    "bbox": [
+      [
+        56,
+        45,
+        641,
+        310
+      ]
+    ],
+    "page_idx": 25
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "C",
+    "footnote": [],
+    "bbox": [
+      [
+        55,
+        335,
+        641,
+        508
+      ]
+    ],
+    "page_idx": 27
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6",
+    "footnote": [],
+    "bbox": [
+      [
+        55,
+        520,
+        641,
+        784
+      ]
+    ],
+    "page_idx": 27
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Figure 7",
+    "footnote": [],
+    "bbox": [
+      [
+        45,
+        280,
+        930,
+        688
+      ]
+    ],
+    "page_idx": 27
+  }
+]

preprint/preprint__7ea08d3a6b99d16e7f9a32724c54deb41c2325e4e80d5feab1f63b193d958db8/preprint__7ea08d3a6b99d16e7f9a32724c54deb41c2325e4e80d5feab1f63b193d958db8.mmd

@@ -0,0 +1,513 @@
+
+# Tyrosine phosphorylation of CARM1 promotes its enzymatic activity and alters its target specificity  
+
+Hidehiro Itonaga  University of Miami  
+
+Adnan Mookhtiar  Harvard University  
+
+Sarah Greenblatt  University of Miami  
+
+Fan Liu  https://orcid.org/0000- 0003- 4142- 5139  
+
+Concepcion Martinez  University of Miami  
+
+Renata Grozovsky  University of Miami  
+
+Daniel Bilbao  University of Miami  
+
+Masai Rains  University of Miami  
+
+Pierre-Jacques Hamard  University of Miami  
+
+Jun Sun  
+
+Afoma Umeano  University of Miami  
+
+Stephanie Duffort  Miller School of Medicine, University of Miami  
+
+Chuan Chen  University of Miami  https://orcid.org/0000- 0001- 5701- 0857  
+
+Na Man  University of Miami  
+
+Gloria Mas  University of Miami  
+
+Stephan Schurer  University of Miami  https://orcid.org/0000- 0001- 7180- 0978
+
+<--- Page Split --->
+
+Stephen Nimer snimer@med.miami.edu  
+
+University of Miami https://orcid.org/0000- 0003- 2439- 7586  
+
+## Article  
+
+Keywords: JAK2- V617F, CARM1, protein arginine methyltransferases, chromatin, RUNX1  
+
+Posted Date: July 12th, 2022  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 1807575/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 22nd, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 47689- 4.
+
+<--- Page Split --->
+
+## Abstract  
+
+Coactivator- associated arginine methyltransferase 1 (CARM1) is overexpressed in cancer, and it has emerged as an important target in acute myeloid leukemia and other hematologic malignancies. Janus kinase 2 (JAK2), that is activated by mutation in a variety of myeloid malignancies, can dictate chromatin structure via multiple effects. Here, we find that the hyperactivated JAK2- V617F mutant kinase phosphorylates CARM1, increasing its methyltransferase activity and altering its target specificity. Phospho- CARM1 binds and methylates the RUNX1 transcription factor, and the asymmetric dimethylation of R223 and R319 in RUNX1 is lost in engineered to express only non- phosphorylatable CARM1 mutant proteins in JAK2- V617F+ cell lines. The decreased CARM1 activity found in these cell lines impairs cycle progression and induces apoptosis. We have established a link between activated JAK2 and CARM1 activity, and demonstrate that dual targeting of JAK2 and CARM1 is more effective than monotherapy in phospho- CARM1+ cell lines.  
+
+## Introduction  
+
+Protein arginine methylation is an essential protein post- translational modification, with \(\sim 7\%\) of arginine residues being methylated, which is comparable to the \(9\%\) of serine residues that are phosphorylated and the \(7\%\) of lysine residues that are ubiquitinated [1]. Protein arginine methyltransferases (PRMTs) catalyze monomethylation, asymmetric dimethylation, or symmetric dimethylation reactions on arginine residues [2], and are classified as class I (asymmetric dimethyl arginine; ADMA), class II (symmetric dimethyl arginine; SDMA) and class III methyltransferases (monomethyl arginine; MMA) [2- 4]. PRMTs are ubiquitously expressed and they regulate multiple cellular processes, including transcription, RNA splicing, DNA replication, DNA repair, protein translation, and cellular metabolism, thereby affecting cell growth, proliferation and differentiation [5- 11].  
+
+Coactivator- associated arginine methyltransferase 1 (CARM1), also known as PRMT4, was originally identified as a coactivator for steroid hormone receptors [12]. CARM1 is a type I protein arginine methyltransferase (PRMT), that catalyzes the asymmetric dimethylation of arginine residues in histones, such as H3R17 and H3R26, which are thought to promote transcription [13, 14]. Furthermore, CARM1 also catalyzes the ADMA of non- histone substrates, including transcription factors like RUNX1 [15], histone acetyltransferases (e.g. p300 and CBP) [16- 18], the steroid receptor co- activator AIB1 [19, 20], RNA binding proteins (e.g. PABP1) [21], RNA splicing factors (e.g. SAP49, SmB, and U1C) [22], and components of the SWI/SNF chromatin remodeling complex (e.g. BAF155) [16, 23].  
+
+We previously demonstrated that CARM1 blocks the myeloid differentiation of normal hematopoietic stem/progenitor cells (HSPCs) by promoting the assembly of a repressive RUNX1 complex [15], and that Carm1 knockout in adult mouse HSPCs prevents the development of acute myeloid leukemia (AML), driven by either the AML1- ETO or MLL- AF9 oncogenes, but only modestly decreases long- term hematopoietic stem cell (HSC) numbers [24]. Carm1 knockout also abrogated the maintenance of AML, suggesting that CARM1 inhibition could have therapeutic efficacy in AML [24]. CARM1 is overexpressed
+
+<--- Page Split --->
+
+in a variety of cancers, including breast, lung, colorectal, liver, and prostate cancer [25–33] and because it drives key oncogenic processes, it could be a therapeutic target in these diseases as well.  
+
+There have been several reports of post- translational modifications (PTMs) of CARM1, including its serine phosphorylation and arginine auto- methylation [34–37]. Yet, little is known about how or whether post- translational modifications of CARM1 affect its role in promoting cancer development. We have been examining how oncogenic signaling pathways affect the activity of epigenetic modifiers, including tyrosine kinases such as Janus- activated kinases (JAK) family (JAK1, JAK2, JAK3, and TYK2) which are activated by cytokine receptors and other cell surface receptors [38–42]. Activated JAK2 signals through downstream effectors including the signal transducers and activators of transcription (STAT) transcription factors (TFs), and the Ras- mitogen- activated protein kinase (MAPK) and phosphoinositide 3- kinase (PI3K) pathways, which regulate hematopoietic cell differentiation, cell proliferation, and apoptosis [39, 43, 44]. A single somatic mutation, V617F (exon 14) in the JH2 domain of JAK2, which disrupts the JH1- JH2 autoinhibitory interaction, leading to JAK2 hyperactivation [45–47], is a common hallmark of the BCR/ABL1- negative myeloproliferative neoplasms. It is found in > 95% of polycythemia vera patients, with homozygous mutations found in patients with longstanding disease. It is also found in \(\sim 30–50\%\) of patients with essential thrombocythemia and primary myelofibrosis, and in \(\sim 1–4\%\) of adult AML and myelodysplastic syndrome patients [48–51].  
+
+We now report that when it is phosphorylated and activated, the JAK2- V617F kinase phosphorylates CARM1 on tyrosine- 149 and – 334 (which are located within the CARM1 catalytic domain), promoting its methyltransferase activity and its nuclear and chromatin localization. Based on multiple in vitro and in vivo studies, we find that CARM1 phosphorylation alters its substrate specificity and its effects on CARM1 target gene selection. We also see important biological differences in the ability of phosphorylatable vs. non- phosphorylatable CARM1 to support leukemia cell proliferation and demonstrate the therapeutic relevance of targeting both JAK2- V617F and CARM1. These results provide novel insights into the regulation of chromatin structure by tyrosine kinases and into the pathogenesis of myeloid neoplasms.  
+
+## Results  
+
+## Tyrosine residues in CARM1 are phosphorylated by JAK2 in vitro  
+
+To determine whether CARM1 is a direct substrate of JAK2, and identify potential CARM1 tyrosine phosphorylation sites, we performed cell- free in vitro kinase assays using GST- tagged CARM1 protein as the substrate for a recombinant active form of JAK2 kinase, using recombinant PAK1 (a known JAK2 substrate) as a positive control (Fig. 1A). The JAK2- dependent tyrosine phosphorylation of GST- tagged CARM1 was readily identified (lanes 4 and 5), as incorporation of a JAK2 inhibitor (ruxolitinib, RUX) in the kinase assay completely abrogated CARM1 phosphorylation (lane 7). We next identified tyrosine- 149 (Y149) and tyrosine- 334 (Y334) as the sites of JAK2 phosphorylation by subjecting the in vitro phosphorylated, recombinant GST- tagged CARM1 protein to mass spectrometry analysis (Fig. 1B- E and
+
+<--- Page Split --->
+
+S1A). These tyrosine residues are located within the core catalytic domain of CARM1 (Fig. 1F), and highly conserved in CARM1, from African clawed frog to human (Figure S1B and S1C).  
+
+## CARM1 phosphorylation is mediated by JAK2 in myeloid leukemia cells  
+
+We generated two phospho- tyrosine specific rabbit polyclonal antibodies, against either Y149 phosphorylated or Y334 phosphorylated CARM1 (Figure S2A); these antibodies recognize in vitro phosphorylated recombinant GST- tagged CARM1 protein, but not the unphosphorylated protein, in a dose- dependent manner (Figure S2B and S2C). Furthermore, these antibodies recognize MYC- tagged wild- type (WT) CARM1, but not the non- phosphorylatable MYC- tagged CARM1 mutant proteins (that contain Y149F, Y334F, or Y149F/Y334F amino acid substitutions), based on the in vitro assays (with or without active JAK2 kinase) using the immunoprecipitants from K562 cell lysates (Figure S2D).  
+
+We compared the level of CARM1 protein expression and CARM1- Y149 and - Y334 phosphorylation in 14 myeloid leukemia cell lines, using normal human \(\mathrm{CD34^{+}}\) cord blood (CB) cells as control. As previously described [15, 24], we found that CARM1 expression levels are higher in nearly all of these myeloid leukemia cell lines, than in the \(\mathrm{CD34^{+}}\) CB cells (Figure 2A). HEL cells, that express JAK2- V617F, showed the highest level of phosphorylated CARM1- Y149 and - Y334 (lane 2), while SET2 cells, which also express JAK2- V617F, showed abundant CARM1 protein but a lower relative amount of phosphorylated CARM1 protein (lane 3) than the HEL cells. K562 cells (lane 8) had abundant CARM1, but less phosphorylated CARM1. Treating HEL cells with the JAK2 inhibitor (RUX, 500 nM) decreased the phosphorylation of CARM1- Y149 and - Y334, and the phosphorylation of STAT5, confirming that JAK2 is the relevant kinase phosphorylating CARM1 (Figure 2B and S2E).  
+
+## Activated JAK2 binds to and phosphorylates CARM1  
+
+To determine whether JAK2 protein interacts with CARM1 in vivo, we transduced HEL, SET2, and K562 cells with an HA- tagged CARM1 construct. HA- tagged CARM1 was then immunoprecipitated, using an anti- HA antibody, followed by immunoblotting with an anti- JAK2 antibody. We were able to detect the direct binding of JAK2 to CARM1 in HEL cells and to a lesser extent in SET2 cells, but weakly in K562 cells (Figure 2C). To determine what accounts for these differences, we focused on the phosphorylation status of JAK2 in these cells, as HEL cells have a bi- allelic JAK2- V617F mutation, while SET2 cells have a mono- allelic JAK2- V617F mutation (Figure S3A). Biallelic mutations lead to a higher level of JAK2 autophosphorylation, leading us to hypothesize that the bi- allelic JAK2- V617F mutation in HEL cells may trigger the higher level of CARM1 phosphorylation. First, we confirmed that HEL cells have the highest level of JAK2 phosphorylation among the various myeloid leukemia cell lines tested, including SET2 cells (Figure S3B). UKE- 1 cells, which carry a bi- allelic JAK2- V617F mutation, also showed a high level of JAK2 auto- phosphorylation, and of CARM1- Y149/Y334 phosphorylation (Figure 2D). We then found that none of the other JAK family members (JAK1, JAK3 or TYK2) bound to CARM1, even when CARM1 was overexpressed in HEL cells (data not shown).
+
+<--- Page Split --->
+
+We next examined whether JAK2 auto- phosphorylation affected the binding of JAK2 to CARM1 or its kinase activity over CARM1, using an anti- HA antibody for the immunoprecipitation and an antibody against Y1007/Y1008 phosphorylated JAK2 for immunoblotting. Phosphorylated JAK2 bound CARM1 in HEL cells that overexpress HA- tagged CARM1 (Figure 2E). Given studies showing that JAK1 and TYK2 activity can transphosphorylate JAK2 [52], we generated JAK1 and TYK2 knockout (KO) HEL cells using the clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR- associated protein- 9 (Cas9) nuclease system. KO of either JAK1 or TYK2 decreased the phosphorylation of JAK2 (by \(18 - 28\%\) ) and CARM1 (by \(38 - 55\%\) ) (Figure 2F), but neither KO reduced the binding of JAK2 to CARM1 (Figure S4A). To confirm that the binding of JAK2 to CARM1 is JAK2- phosphorylation independent, we used a type II JAK2 inhibitor, CHZ868 [53]. CHZ868 treatment abrogated JAK1- mediated JAK2- Y1007/Y1008 phosphorylation (Figure S4B), but it did not reduce the binding of JAK2 to CARM1. Taken together, these results indicate that JAK2 activation, through transphosphorylation by JAK1 and TYK2, enhances JAK2 tyrosine kinase activity against CARM1, without affecting the binding of JAK2 to CARM1.  
+
+## Tyrosine phosphorylation of CARM1 increases its methyltransferase activity  
+
+To better understand how phosphorylation of CARM1 affects its substrate binding and methyltransferase activity, we analyzed published crystal structures of CARM1 and PABP1. CARM1- Y149 and - Y334 phosphorylation are predicted to increase the binding of CARM1 to its unmethylated substrates (Figure 3A and 3B). Y149 and Y334 phosphorylation could also affect the methyltransferase activity of CARM1, so we purified MYC- tagged CARM1 protein from 293T cells transfected with WT or mutant CARM1 (using the non- phosphorylatable Y149F, Y334F, and Y149F/Y334F CARM1 proteins) and examined the ability of these CARM1 proteins to methylate histone H3.1 arginine- 17 (R17) in vitro and in vivo. Upon incubation of purified CARM1 with S- [methyl- \(^{14}\mathrm{C}]\) - adenosyl- methionine and recombinant histone H3.1 in an in vitro methylation assay, while WT- CARM1 had significant methyltransferase activity against H3.1, none of the non- phosphorylatable CARM1 proteins had methyltransferase activity against histone H3.1 (Figure 3C). Next, we examined whether incubating CARM1 with active JAK2 kinase affected its ability to methylate histone H3.1 in vitro. We confirmed the phosphorylation of CARM1, and then incubated histone H3.1 with S- [methyl- \(^{14}\mathrm{C}]\) - adenosyl- methionine and phosphorylated or non- phosphorylated CARM1. The addition of JAK2 kinase increased CARM1 methyltransferase activity on histone H3.1 (Figure 3D, lane 9 vs. lane 7). The mass spectrometry analysis further confirmed the increased methylation level of histone H3R17 by phosphorylated CARM1, compared to non- phosphorylated CARM1 (Figure S5).  
+
+## The Y149F mutation in CARM1 impairs it's in vivo dimerization  
+
+CARM1 dimer formation involves interactions between the so- called dimerization arm (residues 300- 338) and helices \(\alpha \mathrm{X}\) , \(\alpha \mathrm{Y}\) , \(\alpha \mathrm{Z}\) , \(\alpha \mathrm{A}\) , and \(\alpha \mathrm{B}\) (residues 144- 232) [54]. Given that the sites of CARM1 phosphorylation (Y149 and Y334) are located within regions involved in CARM1 dimerization (Figure 3E), we demonstrated that the JAK2 inhibitor (RUX, \(10 \mu \mathrm{M}\) ) decreased the dimerization of CARM1 in 293T cells co- transfected with HA- tagged WT- CARM1 and MYC- tagged CARM1 constructs (Figure S6A). To investigate whether each tyrosine phosphorylation site in CARM1 affects dimerization, we co- transfected
+
+<--- Page Split --->
+
+HA- tagged WT- CARM1 and MYC- tagged CARM1 (WT and Y- to- F mutated) constructs into 293T cells, immunoprecipitated with anti- HA antibodies, and probed with anti- MYC antibodies. MYC- tagged CARM1- WT and - Y334F mutant CARM1 were readily detectable in the HA immunoprecipitates (Figure 3F, lane 7 and 8). However, the Y149F mutant and the Y149F/Y334F double mutant CARM1 proteins reduced the binding ability to HA- tagged WT, compared to MYC- tagged WT CARM1 (Figure 3F, lane 9 and 10 vs. lane 7). Thus, it appears that phosphorylation of Y149 in CARM1 can promotes its dimerization.  
+
+## CARM1 tyrosine phosphorylation promotes its nuclear localization  
+
+We investigated whether phosphorylation affects the subcellular localization of CARM1 using an anti- CARM1 antibody. CARM1 localizes mainly in the cytoplasm of HEL, SET2, and K562 cells, with some CARM1 found in the nuclear soluble fraction (Figure S6B). Then, using CARM1 phospho- specific antibodies, and HEL and UKE- 1 cells, we found a significantly higher proportion of Y149 and Y334 phosphorylated CARM1 in the nucleus, and in the chromatin fraction of both HEL cells and UKE- 1 cells (Figure 3G). These results indicated that the tyrosine phosphorylation of CARM1 contributes to its nuclear and chromatin localization.  
+
+## Identification of CARM1-interacting proteins using BiolD  
+
+To capture the proteins which are strongly, weakly, or transiently associated with phosphorylated CARM1, we engineered HEL cells (with abundant phosphorylated CARM1) and K562 cells (with primarily non- phosphorylated CARM1) to express the proximity- dependent biotin identification (BiolD) system (Figure 4A and S7A). Using shotgun mass spectrometry, we identified 128 and 60 proteins that significantly interact with CARM1 in HEL and K562 cells, respectively (supplemental data and Figure S7B). CARM1- associated proteins clustered into groups of proteins identified as being involved in purine nucleotide metabolism, E2F targets, MYC targets, histone and chromatin binding proteins, and ribonucleoprotein complex biogenesis, based on enriched pathway analysis of Metascape (metascape.org) for both cell lines (Figure 4B and S7C). We identified many previously known CARM1- associated proteins or substrates, including NUDT4, ADAR, RUNX1, SON, and CARM1 in HEL cells [14, 15] (Figure 4C), and confirmed the high intensity of RUNX1 fragments interacting with CARM1- BirA\* fusion in HEL cells but not in K562 cells, using target mass spectrometry (Figure S7D), which suggests that CARM1 phosphorylation enhances the binding of CARM1 to RUNX1. To test this hypothesis, we generated HEL cells overexpressing MYC- tagged CARM1 (WT, Y334F, Y149F, and Y149F/Y334F mutants), and immunoprecipitated the WT or non- phosphorylatable mutant proteins, using an antibody against the MYC- tag. WT- CARM1, but none of the non- phosphorylatable CARM1 mutants were able to pull down RUNX1 (Figure 4D), confirming the role of CARM1 phosphorylation in regulating this interaction with RUNX1.  
+
+## RUNX1 R223 and R319 are strongly methylated by phosphorylated CARM1  
+
+Having identified arginine- 223 (R223) in RUNX1 as a site of CARM1 asymmetric dimethylation [15], we examined whether the tyrosine phosphorylation of CARM1 affects its ability to methylate RUNX1. Using
+
+<--- Page Split --->
+
+mass spectrometry and an in vitro methylation assay, we confirmed R223 as a CARM1 target site and also identified arginine- 319 (R319) in RUNX1 as another potential CARM1 methylation site (Figure S8A). We next generated an asymmetric dimethylation specific anti- RUNX1- R319 antibody, and used this antibody and a previously published asymmetric dimethylation specific anti- RUNX1- R223 antibody [15] (Figure S8B) to show that RUNX1- R223 and - R319 are indeed methylated by CARM1 in vivo. Both residues, R223 which is located C- terminal to the RUNX1b DNA binding domain and R319 which is located within the activation domain of RUNX1b, are evolutionally conserved among mammals (Figure S8C- E).  
+
+To prove that CARM1 is the relevant methyltransferase, we created doxycycline- inducible CARM1 knockdown (KD) HEL cells, using three different small hairpin RNAs (shRNAs) that significantly decrease CARM1 RNA and protein expression (Figure S9). KD of CARM1 by all three shRNAs significantly decreased the level of RUNX1- R223me2a and RUNX1- R319me2a; they also decreased the ADMA levels of BAF155 and PABP1 (Figure 4E).  
+
+To define the role of CARM1 phosphorylation on the asymmetric dimethylation of RUNX1 (and its other substrates), we generated isogenic HEL cells lines carrying homozygous CARM1 non- phosphorylatable mutation (Y149F or Y334F single mutations, or Y149F/Y334F double mutation), using the CRISPR/Cas9 nuclease system (Figure S10). We then examined the HEL cells harboring non- phosphorylatable CARM1 mutations (Y149F or Y334F single mutations, or the Y149F/Y334F double mutation), and found that mutation of either site abrogates the asymmetric dimethylation of RUNX1- R223 and RUNX1- R319 (Figure 4F). HEL cells containing either of the non- phosphorylatable single mutations also showed decreased levels of asymmetrically dimethylated BAF155 and PABP1, while HEL cells with the CARM1 double mutation showed much lower levels of ADMA BAF155 and PABP1 than those with single mutations.  
+
+To confirm that JAK2 dependent tyrosine phosphorylation of CARM1 is responsible for the increased ADMA of RUNX1 and other CARM1 substrates in HEL cells, we treated HEL cells with RUX (Figure 4G). RUNX1- R223 and - R319 dimethylation was reduced after 5 days of RUX exposure. A modest decrease in BAF155 dimethylation and a greater decrease in PABP1 dimethylation were also seen.  
+
+To better understand the kinetics of methylation and re- methylation of CARM1 substrates, we treated HEL cells with a potent and selective CARM1 inhibitor (EPZ025654) for 5 days, and monitored cell growth and substrate methylation over the subsequent five days. While BAF155 and PABP1 methylation was restored after a 1- day EPZ025654- free period, RUNX1 re- methylation was first observed three days after EPZ025654 removal (Figure S11). We next assessed the re- methylation levels of BAF155, PABP1, and RUNX1 in HEL cells treated with RUX (or DMSO) after the removal of EPZ025654. RUX significantly impaired the re- methylation of RUNX1, but not that of BAF155 or PABP1 (Figure 4H). These results indicate the different requirements and time course of the effect of JAK2 on the ADMA of various CARM1 substrates; RUNX1 dimethylation is most sensitive to the JAK2 kinase inhibitor, while BAF155 dimethylation is only sensitive to the CARM1 inhibitor.  
+
+## Biological consequence of CARM1 phosphorylation
+
+<--- Page Split --->
+
+We used the non- phosphorylatable CARM1 mutant knock- in HEL cells to examine the biological effects of CARM- Y149 and - Y334 phosphorylation on cell behavior. While the single mutant cells grew normally, the double CARM1 mutant (Y149F/Y334F)- expressing HEL cells showed reduced proliferation (Figure 5A). Cell cycle analysis revealed that all three cell lines harboring homozygous single or double mutant CARM1 had a decreased S- phase fraction and an increased G2/M fraction (consistent with significant G2/M arrest) (Figure 5B). We observed an increase in apoptosis (indicated by an increased sub- G1 fraction and annexin V- positive cells) particularly in cells harboring the CARM1- Y149F/Y334F double mutation (Figure 5B and S12). These results demonstrate that phosphorylation of Y149 and Y334 in CARM1 regulates the proliferation of HEL cells, affecting both cell cycle and apoptosis to varying degrees.  
+
+## Distinct transcriptional profiles regulated by phosphorylated CARM1  
+
+To understand the molecular consequences of CARM1 phosphorylation, we examined the gene expression profiles of the non- phosphorylatable CARM1 mutation knock- in cells, by RNA sequencing. We identified 1,505 differentially expressed genes ( \(\geq 1.5\) - fold change and adjusted p- value \(< 0.05\) ) between CARM1 WT and - Y149F/Y334F double mutations (Figure 5C), while the single CARM1- Y149F or Y334F knock- in cells showed fewer differentially expressed genes, when compared to CARM1 WT cells (Figure S13A and S13B). Gene ontology (GO) analysis identified that genes involved in G2/M cell cycle progression and apoptosis were negatively enriched in Y149F/Y334F double mutation knock- in HEL cells (Figure 5D). We also used gene set enrichment analysis (GSEA) to identify pathways differentially regulated by the lack of CARM1 phosphorylation and found that gene sets associated with G2/M cell cycle progression and anti- apoptosis were significantly downregulated in Y149F/Y334F mutation knock- in HEL cells (Figure 5E). Heatmaps of FDR (q < 0.25) values in three non- phosphorylatable CARM1 mutation knock- in cell lines are shown in Figure 5F, which show that G2/M checkpoints- associated gene sets were downregulated in Y149F/Y334F double and Y149 single mutations, while anti- apoptosis- associated gene sets were downregulated in Y149F/Y334F double mutation cells. Among the genes associated with G2/M cell cycle progression and anti- apoptosis that were differentially expressed (adjusted p- value \(< 0.05\) and FC>1.5) and relevant in myeloid malignancies [55- 60], we found SMAD3, CCND2, KIF5B, BCL2, BCL2A1, and SATB1 (Figure 5G). We confirmed the downregulation of both G2/M checkpoint (SMAD3, CCND2, and KIF5B) and anti- apoptosis (BCL2, BCL2A1, and SATB1) genes in the double CARM1 mutation knock- in HEL cells, using qRT- PCR (Figure S13C and S13D).  
+
+Consistent with our previous reports that CARM1 promotes a DPF2- containing repressor complex that repress miR- 223 expression [3, 15], we found increased expression of MIR223 in CARM1- Y149F cells and in the CARM1- Y149F/Y334F double mutation cells (Figure S13E). As MIR223 promotes myeloid differentiation, the decrease of its expression led us to evaluate the expression of HSPC stemness- related genes by RNA sequencing. Gene sets associated with HSPC stemness were decreased in CARM1- Y149F/Y334F mutation knock- in HEL cells (normalized enrichment score of - 1.36 with FDR q- value of 0.216) (Figure 5H), and two HSPC- associated genes, CD34 and BMI- 1, were downregulated in Y149F single and Y149F/Y334F double mutations knock- in, based on qRT- PCR (Figure 5I). RUNX (RUNX1, 2, and
+
+<--- Page Split --->
+
+3)- target gene sets were not significantly altered in CARM1- Y149F/Y334F mutation knock- in HEL cells (FDR q- value of 0.717) (Figure S13F); however, three RUNX- target genes (ID2, MIR144, and RNF144A) were identified as a subset of core- enrichment genes with \(\geq 1.5\) - fold change and adjusted p- value \(< 0.05\) . Given that RUNX1 regulates the transcription of ID2 and MIR144 [61, 62], we independently evaluated their gene expression by qRT- PCR, and confirmed that the Y149F/Y334F double mutation knock- in induced the upregulation of ID2 and MIR144 mRNA (Figure 5J).  
+
+To investigate the global localization of RUNX1 on chromatin, we performed ChIP- seq analyses using antibodies against asymmetrically dimethylated R319- RUNX1 and total RUNX1. As we expected, knock- in of CARM1- Y149F or - Y334F single, or - Y149F/Y334F double mutation decreased overall signals of dimethylated RUNX1- R319 and, to a lesser degree, those of total RUNX1 (Figure 5K). We found that dimethylated R319- RUNX1 shared occupancy for ID2, MIR144, and MIR223 with less effect on total RUNX1 in HEL cells expressing CARM1- WT (Figure 5L and S14). In addition, CARM1 non- phosphorylatable mutation decreased the signals of dimethylated RUNX1- R319 within 5 kb of the transcription start sites for ID2, MIR144, and MIR223 with less effect on total R319- RUNX1. These results suggested that phosphorylated CARM1 regulates the chromatin binding of asymmetrically dimethylated R319- RUNX1 and to a lesser extent unmethylated RUNX1, leading to changes in RUNX1 target- gene expression.  
+
+## Targeting the JAK2-CARM1 axis  
+
+We have previously shown that CARM1 KD or inhibition reduced the cell proliferation of AML (Figure S15A), inducing G0/G1 cell cycle arrest (Figure S15B) and differentiation [24], and to a lesser degree apoptosis (Figure S15C).  
+
+We assessed the efficacy of EPZ025654 on a variety of cell lines and found a dose- dependent reduction in the proliferation of UKE- 1 cells, but not HEL cells. The half maximal inhibitory concentration (IC50) values of EPZ025654 were 23.7 μM, 1.8 μM, and 186.5 nM in HEL, UKE- 1, and SET2 cells, respectively (Figure S16A), while the IC50 values for RUX were 492 nM, 269 nM, and 56 nM in HEL, UKE- 1, and SET2 cells, respectively (Figure S16B). We assessed the level of asymmetric dimethyl RUNX1, BAF155, and PABP1 in three cell lines and found that EPZ025654 significantly reduced the levels of ADMA RUNX1 (at R223 and R319) (Figure S16C); it also reduced ADMA BAF155 and PABP1 levels, in a time- dependent manner, without affecting the phosphorylation of JAK2, CARM1, STAT5, ERK, or AKT (Figure 6A), which suggests that CARM1 inhibitors inhibit cell growth via distinct signaling pathways. To determine if the combination of EPZ025654 and RUX could be synergistic on certain cell lines, we examined cell proliferation after 6 days of EPZ025654 treatment and 2 days of RUX treatment. We observed a significant synergistic inhibition effect on HEL and UKE- 1 cells (based on a positive Bliss score), but only an addictive effect on SET2 cells (Figure 6B and S16D), which suggests that a synergistic effect of JAK2 inhibition, combined with EPZ025654, may occur primarily in cells that contain phosphorylated CARM1.  
+
+We checked efficacy of single vs. combination therapy on the colony- forming capacity of these cell lines, and found a significant reduction in the colony- forming potential of HEL and UKE- 1 cells using the
+
+<--- Page Split --->
+
+combination therapy for 14 days, compared with either drug alone, which was not seen in SET2 cells (Figure 6C). This further suggests that JAK2 inhibitors have the potential ability to sensitize cells with phosphorylated CARM1 to CARM1 inhibition.  
+
+## Discussion  
+
+Having identified the phosphorylation of Y149 and Y334 in CARM1 as novel PTMs mediated by JAK2- V617F mutants, we show that these PTMs increase the enzymatic activity and alter the cellular localization and target specificity of CARM1. CARM1 phosphorylation enhances its ability to block differentiation, and regulate apoptosis and cell cycling by controlling G2/M checkpoints (Fig. 7). Our work highlights the importance of these regulatory effects on the phenotypes driven by CARM1 in hematologic cells that express the JAK2- V617F oncogene.  
+
+We also demonstrated auto- phosphorylation and cross- phosphorylation of Y1007/Y1008 in the JAK2 activation loop by JAK2, JAK1, or TYK2 which more strongly promotes CARM1 tyrosine phosphorylation, especially in bi- allelic JAK2- V617F mutant cells. Thus, KD of either JAK1 or TYK2 decreased the phosphorylation levels of CARM1 by altering the ability of JAK2 to phosphorylate CARM1. However, phosphorylation of JAK2- V617F on Y1007/Y1108 did not affect the binding of JAK2 to CARM1. CARM1- Y149 and - Y334 phosphorylation is promoted by the active conformation of the mutant JAK2 protein, which is stimulated in the case of JAK2- V617F by the expression of type I cytokine receptors (e.g. EpoR, MPL, or G- CSFR), and inhibited by prolonged exposure to type I JAK2- inhibitors [54, 55]. Our data demonstrate that increased CARM1 tyrosine phosphorylation is a biological marker of cells with hyperactivated JAK2 (e.g. the JAK2- V617F mutant protein) [63].  
+
+We observed increased methyltransferase activity of phosphorylated CARM1 on histone 3.1, with reduced methyltransferase activity for the Y149F and Y334F mutations, which render CARM1 non- phosphorylatable. Our crystal structural model analysis shows that Y149 and Y334 phosphorylation does increase CARM1 binding to its substrates. The phosphorylation of Y149 and Y334 in CARM1 promotes its nuclear localization, allowing enhanced binding to nuclear substrates, including histones, chromatin binding proteins, and RUNX1. The Y149F mutant CARM1 in particular, shows diminished dimerization and minimal methyltransferase activity. We confirmed RUNX1 as a key protein interaction with CARM1, by conducting a BioID screen, and showed that the non- phosphorylatable CARM1 mutant proteins did not bind RUNX1. Thus, in addition to promoting dimerization, the phosphorylation of CARM1 affects its localization, substrate binding, and methyltransferase activity.  
+
+We identified R223 and R319 of RUNX1 as residues asymmetrically dimethylated by CARM1, and showed that CARM1- Y149 and - Y334 phosphorylation enhanced the asymmetrical dimethylation of both R223- and R319- RUNX1. CARM1 can regulate hematopoietic cell differentiation through multiple mechanisms, including the generation of a repressor complex that contains asymmetrically dimethylated RUNX1- R223, and negatively regulates miR- 223 expression [3, 15, 64]. We now find that non- phosphorylatable CARM1 mutant cell lines show increased expression of miR- 223 and several other RUNX1- target genes (/D2 and
+
+<--- Page Split --->
+
+MIR144). Knock- in of a non- phosphorylatable CARM1 mutation downregulated the expression of BMI- 1, which is a regulator of self- renewal that plays a role in JAK2- V617F mutant hematopoietic stem cells [65], and in other cancer stem cell phenotypes. Given its substrate targets (e.g. BAF155 and RUNX1) and gene targets (e.g. BMI- 1 and ID2), CARM1 and phospho- CARM1 play a pivotal role in hematologic malignancies with the JAK2- V617F mutation, and in other settings as well.  
+
+The non- phosphorylatable CARM1 mutation (Y149F/Y334F) knock- in HEL cells show decreased cell growth with increased cell cycle arrest and apoptosis, likely due to the downregulation of gene expression associated with G2/M progression and anti- apoptosis, including BCL2 family members (BCL2 and BCL2A1) which have been implicated in controlling apoptosis in JAK2- V617F mutant myeloid malignancies [66- 68]. Similarly, the asymmetrical dimethylation of BAF155 by CARM1 has been shown to inhibit apoptosis of ovarian cancer cells through downregulation of pro- apoptotic gene expression (DAB2, DLC1, and NOXA) [69]. We could not detect a significant difference in the expression of these genes in WT- vs. Y149F/Y334F- CARM1 expressing HEL cells (data not shown), despite similar changes in the level of ADMA BAF155, confirming the cell- context- specific effects of arginine dimethylation.  
+
+The dependency of JAK2- V617F mutant AML cells on CARM1 is consistent with our previous studies showing that CARM1 is an essential gene for the growth of myeloid leukemia cells; further evidence was provided by an RNAi screen analysis conducted as part of the Dependency Map database (https://depmap.org/portal/) [70]. Non- phosphorylatable CARM1 mutant- expressing HEL cells showed significantly decreased cell growth, suggesting some dependency of HEL cells on CARM1 phosphorylation. This led us to evaluate the efficacy of inhibiting both JAK2 and CARM1 pathway, and we found that small- molecule inhibitors targeting CARM1 (EPZ025654) sensitized JAK2- mutant cells to JAK2 inhibitors.  
+
+In conclusion, CARM1 phosphorylation mediated by hyperactivated JAK2 regulates its methyltransferase activity and is required for maximal proliferation of myeloid neoplasms. Our results suggest a potential role of targeting both JAK2 and CARM1 in JAK2- V617F mutant myeloid malignancies.  
+
+## Material And Methods  
+
+Methodology is described in the Supplementary Methods file.  
+
+## Declarations  
+
+## Author contributions  
+
+H.I., S.M.G., and S.D.N. conceived the project. H.I. and S.D.N. designed the experiments and wrote the manuscript. H.I. conducted most of the experiments. A.K.M. and P.- J.H. assisted with sample preparation for mass spectrometry. P.- J.H. assisted with RUNX1 arginine methylation experiments. S.M.G. assisted with in vitro phosphorylation assay. A.K.M. and F.L. assisted with in vitro methylation assay. C.M., M.R., and J.S. contributed to the shRNA knockdown. R.G. assisted in immunoprecipitation and immunoblotting
+
+<--- Page Split --->
+
+assays to determine CARM1 phosphorylation in cells. J.S. assisted in ChiP- seq assays. D.B., C.M., and S.D. assisted immunoblotting to determine the subcellular fraction of CARM1. G.M.M. assisted with the library preparation for RNA- sequencing. C.C. and N.M. assisted with colony- forming assay. A.C.U. and S.S. performed the crystal structure analysis.  
+
+## Acknowledgements  
+
+We thank the members of the Nimer lab for their assistance and thoughtful input on the manuscript; especially Delphine Prou and Lauren Ashley Whitmore. We also thank the Oncogenomics Shared Resource at Sylvester Comprehensive Cancer Center for RNA- sequencing services and the Biostatistics and Bioinformatics Shared Resource for data analysis. This work was supported by funds from Sylvester Comprehensive Cancer Center, grant R01 CA251664- 01 and 1P30CA240139- 01 from the National Cancer Institute to S.D.N., 1F31CA254232- 01 from the National Cancer Institute to A.K.M., and the Translational Research Program Grant from the Leukemia and Lymphoma Society to S.D.N.  
+
+## References  
+
+1. Larsen SC, Sylvestersen KB, Mund A, Lyon D, Mullari M, Madsen VM, et al. Proteome-wide analysis of arginine monomethylation reveals widespread occurrence in human cells. Sci Signal. 2016;9(443):rs9.  
+2. Bedford MT, Clarke SG. Protein arginine methylation in mammals: who, what, and why. Mol Cell. 2009;33(1):1-13.  
+3. Greenblatt SM, Liu F, Nimer SD. Arginine methyltransferases in normal and malignant hematopoiesis. Exp Hematol. 2016;44(6):435-441.  
+4. Yang Y, Bedford M. Protein arginine methyltransferases and cancer. Nat Rev Cancer. 2013;13(1):37-50.  
+5. Gao WW, Xiao RQ, Peng BL, Xu HT, Shen HF, H MF, et al. Arginine methylation of HSP70 regulates retinoid acid-mediated RARβ2 gene activation. Proc Natl Acad Sci USA. 2015;112(26):E3327-E3336.  
+6. Hsu JM, Chen CT, Chou CK, Kuo HP, Li LY, Lin CY, et al. Crosstalk between Arg 1175 methylation and Tyr 1173 phosphorylation negatively modulates EGFR-mediated ERK activation. Nat Cell Biol. 2011;13(2):174-181.  
+7. Liao HW, Hsu JM, Xia W, Wang HL, Wang YN, Chang WC, et al. PRMT1-mediated methylation of the EGF receptor regulates signaling and cetuximab response. J Clin Invest. 2015;125(12):4529-4543.  
+8. Liu F, Ma, F, Wang Y, Hao L, Zeng H, Jia C, et al. PKM2 methylation by CARM1 activates aerobic glycolysis to promote tumorigenesis. Nat Cell Biol. 2017;19(11):1358-1370.  
+9. Poornima G, Shah S, Vignesh V, Parker R, Rajyaguru P. Arginine methylation promotes translation repression activity of elF4G-binding protein, Scd6. Nucleic Acids Re. 2016;44(19):9358-9368.
+
+<--- Page Split --->
+
+10. Wang YP, Zhou W, Wang J, Huang X, Zuo Y, Wang TS, et al. Arginine Methylation of MDH1 by CARM1 Inhibits Glutamine Metabolism and Suppresses Pancreatic Cancer. Mol Cell. 2016;64(4):673–687. 
+
+11. Zhao DY, Gish G, Braunschweig U, Li Y, Ni Z, Schmitges FW, et al. SMN and symmetric arginine dimethylation of RNA polymerase II C-terminal domain control termination. Nature. 2016;529(7584):48–53. 
+
+12. Chen D, Ma H, Hong H, Koh SS, Huang SM, Schurter BT, et al. Regulation of transcription by a protein methyltransferase. Science. 1999;284(5423):2174–2177. 
+
+13. Schurter BT, Koh SS, Chen D, Bunick GJ, Harp JM, Hanson BL, et al. Methylation of histone H3 by coactivator-associated arginine methyltransferase 1. Biochemistry. 2011;40(19):5747–5756. 
+
+14. Shishkova E, Zeng H, Liu F, Kwiecien NW, Herbert AS, Coon JJ, et al. Global mapping of CARM1 substrates defines enzyme specificity and substrate recognition. Nat Commun. 2017;8:15571. 
+
+15. Vu LP, Perna F, Wang L, Voza F, Figueroa ME, Tempst P, et al. PRMT4 blocks myeloid differentiation by assembling a methyl-RUNX1-dependent repressor complex. Cell Rep. 2013;5(6):1625–1638. 
+
+16. Ceschin DG, Walia M, Wenk SS, Duboé C, Gaudon C, Xiao Y, et al. Methylation specifies distinct estrogen-induced binding site repertoires of CBP to chromatin. Genes Dev. 2011;25(11):1132–1146. 
+
+17. Chevilalrd-Briet M, Trouche D, Vandel L. Control of CBP co-activating activity by arginine methylation. EMBO J. 2002;21(20):5457–5466. 
+
+18. Xu W, Chen H, Asahara H, Tini M, Emerson BM, Montminy M, et al. A transcriptional switch mediated by cofactor methylation. Science. 2001;294(5551):2507–2511. 
+
+19. Naeem H, Cheng D, Zhao Q, Underhill C, Tini M, Bedford MT, et al. The activity and stability of the transcriptional coactivator p/CIP/SRC-3 are regulated by CARM1-dependent methylation. Mol Cell Biol. 2007;27(1):120–134. 
+
+20. Feng Q, Yi Ping, Wong J, O'Malley BW. Signaling within a coactivator complex: methylation of SRC-3/AIB1 is a molecular switch for complex disassembly. Mol Cell Biol. 2006;26(21):7846–7857. 
+
+21. Lee J, Bedford MT. PABP1 identified as an arginine methyltransferase substrate using high-density protein arrays. EMBO Rep. 2002;3(3):268–273. 
+
+22. Cheng D, Côté J, Shaaban S, Bedford MT. The arginine methyltransferase CARM1 regulates the coupling of transcription and mRNA processing. Mol Cell. 2007;25(1):71–83. 
+
+23. Kowenz-Leutz Elisabeth, Pless Ole, Dittmar G, Knoblich M, Leutz A. Crosstalk between C/EBPbeta phosphorylation, arginine methylation, and SWI/SNF/Mediator implies an indexing transcription factor code. EMBO J. 2010;29(6):1105–1115. 
+
+24. Greenblatt SM, Man N, Hamard PJ, Asai T, Karl D, Martinez C, et al. CARM1 Is Essential for Myeloid Leukemogenesis but Dispensable for Normal Hematopoiesis. Cancer Cell. 2018;34(5):868. 
+
+25. Hong H, Kao C, Jeng MH, Eble JN, Koch MO, Gardner TA, et al. Aberrant expression of CARM1, a transcriptional coactivator of androgen receptor, in the development of prostate carcinoma and androgen-independent status. Cancer. 2004;101(1):83–89.
+
+<--- Page Split --->
+
+26. Majumder S, Liu Y, Ford OH, Mohler JL, Whang YE. Involvement of arginine methyltransferase CARM1 in androgen receptor function and prostate cancer cell viability. Prostate. 2006;66(12):1292–1301.  
+
+27. Al-Dhaheri M, Wu J, Skliris GP, Li J, Higashimoto K, Wang Y, et al. CARM1 is an important determinant of ERα-dependent breast cancer cell differentiation and proliferation in breast cancer cells. Cancer Res. 2011;71(6):2118–2128.  
+
+28. Lahusen T, Henke RT, Kagan BL, Wellstein A, Riegel AT. The role and regulation of the nuclear receptor co-activator AIB1 in breast cancer. Breast Cancer Rs Treat. 2009;116(2):225–237.  
+
+29. Frietze S, Lupien M, Silver PA, Brown M. CARM1 regulates estrogen-stimulated breast cancer growth through up-regulation of E2F1. Cancer Res. 2008;68(1):301–306.  
+
+30. Kim YR, Lee BK, Park RY, Nguyen NT, Bae JA, Kwon DD, et al. Differential CARM1 expression in prostate and colorectal cancers. BMC Cancer. 2010;10:197.  
+
+31. Ou CY, LaBonte MJ, Manegold PC, So AY, Ianculescu I, Gerke DS, et al. A coactivator role of CARM1 in the dysregulation of \(\beta\)-catenin activity in colorectal cancer cell growth and gene expression. Mol Cancer Res. 2011;9(5):660–670.  
+
+32. Osada S, Suzuki S, Yoshimi C, Matsumoto M, Shirai T, Takahashi S, et al. Elevated expression of coactivator-associated arginine methyltransferase 1 is associated with early hepatocarcinogenesis. Oncol Rep. 2013;30(4):1669–1674.  
+
+33. Zhong XY, Yuan XM, Xu YY, Yin M, Yan WW, Zou SW, et al. CARM1 Methylates GAPDH to Regulate Glucose Metabolism and Is Suppressed in Liver Cancer. Cell Rep. 2018;24(12):3207–3223.  
+
+34. Feng Q, He B, Jung SY, Song Y, Qin J, Tsai SY, et al. Biochemical control of CARM1 enzymatic activity by phosphorylation. J Biol Chem. 2009;284(52):36167–36174.  
+
+35. Carascossa S, Dudek P, Cenni B, Briand PA, Picard D. CARM1 mediates the ligand-independent and tamoxifen-resistant activation of the estrogen receptor alpha by cAMP. Genes Dev. 2010;24(7):708–719.  
+
+36. Higashimoto K, Kuhn P, Desai D, Cheng X, Xu W. Phosphorylation-mediated inactivation of coactivator-associated arginine methyltransferase 1. Proc Natl Acad Sci USA. 2007;104(3):12318–12323.  
+
+37. Kuhn P, Chumanov R, Wang Y, Ge Y, Burgess RR, Xu W. Auto­methylation of CARM1 allows coupling of transcription and mRNA splicing. Nucleic Acids Res. 2011;39(7):2717–2726.  
+
+38. Ward AC, Touw I, Yoshimura A. The Jak-Stat pathway in normal and perturbed hematopoiesis. Blood. 2000;95(1):19–29.  
+
+39. Khwaja A. The role of Janus kinases in haemopoiesis and haematological malignancy. Br J Haematol. 2006;134(4):366–384.  
+
+40. Silvennoinen O, Witthuhn BA, Quelle FW, Cleveland JL, Yi T, Ihle JN. Structure of the murine Jak2 protein-tyrosine kinase and its role in interleukin 3 signal transduction. Proc Natl Acad Sci USA. 1993;90(18):8429–8433.
+
+<--- Page Split --->
+
+41. Withthun BA, Quelle FW, Silvennoinen O, Yi T, Tang B, Miura O, et al. JAK2 associates with the erythropoietin receptor and is tyrosine phosphorylated and activated following stimulation with erythropoietin. Cell. 1993;74(2):227–236.  
+42. Parganas E, Wang D, Stravoopodis D, Topham DJ, Marine JC, Teglund S, et al. Jak2 is essential for signaling through a variety of cytokine receptors. Cell. 1998;93(3):385–395.  
+43. Rane SG, Reddy EP. JAKs, STATs and Src kinases in hematopoiesis. Oncogene, 2002;21(21):3334–3358.  
+44. Silvennoinen O, Hubbard SR. Molecular insights into regulation of JAK2 in myeloproliferative neoplasms. Blood. 2015;125(22):3388–3392.  
+45. Shan Y, Gnanasambandan K, Ungureanu D, Kim ET, Hammarén H, Yamashita K, et al. Molecular basis for pseudokinase-dependent autoinhibition of JAK2 tyrosine kinase. Nat Struct Mol Biol. 2014;21(7):579–584.  
+46. Bandaranayake RM, Ungureanu D, Shan Y, Shaw DE, Silvennoinen O, Hubbard SR. Crystal structures of the JAK2 pseudokinase domain and the pathogenic mutant V617F. Nat Struct Mol Biol. 2012;19(8):754–759.  
+47. Hammarén HM, Ungureanu D, Grisouard J, Skoda RC, Hubbard SR, Silvennoinen O. ATP binding to the pseudokinase domain of JAK2 is critical for pathogenic activation. Proc Natl Acad Sci USA. 2015;112(15):4642–4647.  
+48. James C, Ugo V, Couédic JL, Steak J, Delhommeay F, Lacout C, et al. A unique clonal JAK2 mutation leading to constitutive signalling causes polycythaemia vera. Nature. 2005;434(7037):1144–1148.  
+49. Levine RL, Wadleigh M, Cools J, Ebert BL, Wernig G, Huntly B, et al. Activating mutation in the tyrosine kinase JAK2 in polycythemia vera, essential thrombocythemia, and myeloid metaplasia with myelofibrosis. Cancer Cell. 2005;7(4):387–397.  
+50. Aynardi, Manur R, Hess PR, Chekol S, Morrissette JD, Babushol D, et al. JAK2 V617F-positive acute myeloid leukaemia (AML): a comparison between de novo AML and secondary AML transformed from an underlying myeloproliferative neoplasm. A study from the Bone Marrow Pathology Group. Br J Haematol. 2018;182(1):78–85.  
+51. Papaemmanuil E, Gerstung M, Malcovati L, Tauro S, Gundem G, Loo PV, et al. Clinical and biological implications of driver mutations in myelodysplastic syndromes. Blood. 2013;122(22):3616–3627.  
+52. Koppikar P, Bhagwat N, Kilpivaara O, Manshouri T, Adli M, Hricik T, et al. Heterodimeric JAK-STAT activation as a mechanism of persistence to JAK2 inhibitor therapy. Nature. 2012;489(7414):155–159.  
+53. Meyer S, Keller MD, Chiu S, Koppikar P, Guryanova OA, Rapaport F, et al. CHZ868, a Type II JAK2 Inhibitor, Reverses Type I JAK Inhibitor Persistence and Demonstrates Efficacy in Myeloproliferative Neoplasms. Cancer Cell. 2015;28(1):15–28.  
+54. Troffer-Charlier N, Cura V, Hassenboehler P, Moras D, Cavarelli J. Functional insights from structures of coactivator-associated arginine methyltransferase 1 domains. EMBO J. 2007;26(20):4391–4401.
+
+<--- Page Split --->
+
+55. Blank U, Karlsson S. The role of Smad signaling in hematopoiesis and translational hematology. Leukemia. 2011;25(9):1379-1388.  
+
+56. Walz C, Crowley B, Hudon HE, Gramlich JL, Neuberg DS, Podar K, et al. Activated Jak2 with the V617F point mutation promotes G1/S phase transition. J Biol Chem. 2006;281(26):18177-18183.  
+
+57. Score J, Curtis C, Waghorn K, Stalder M, Jotterand M, Grand FH, et al. Identification of a novel imatinib responsive KIF5B-PDGFRA fusion gene following screening for PDGFRA overexpression in patients with hypereosinophilia. Leukemia. 2006;20(5):827-832.  
+
+58. Parker JE, Mufti GJ, Rasool F, Mijovic A, Devereux S, Pagliuca A, et al. The role of apoptosis, proliferation, and the Bcl-2-related proteins in the myelodysplastic syndromes and acute myeloid leukemia secondary to MDS. Blood. 2000;96(12):3932-3938.  
+
+59. Jenal M, Batliner J, Reddy VA, Haferlach T, Tobler A, Fey MF, et al. The anti-apoptotic gene BCL2A1 is a novel transcriptional target of PU.1. Leukemia. 2010;24(5):1073-1076.  
+
+60. Luo X, Xu L, Wu X, Tan H, Liu L. Decreased SATB1 expression promotes AML cell proliferation through NF-κB activation. Cancer Cell Int. 2019;19:134.  
+
+61. Keita M, Bachvarova M, Morin C, Plante M, Gregoire J, Renaud MC, et al. The RUNX1 transcription factor is expressed in serous epithelial ovarian carcinoma and contributes to cell proliferation, migration and invasion. Cell Cycle. 2013;12(6):972-986.  
+
+62. Kohrs N, Kolodziej S, Kuvardina ON, Herglotz J, Yillah J, Herkt S, et al. MiR144/451 Expression Is Repressed by RUNX1 During Megakaryopoiesis and Disturbed by RUNX1/ETO. PLoS Genet. 2016;12(3):e1005946.  
+
+63. Vainchenker W, Constantinescu SN. JAK/STAT signaling in hematological malignancies. Oncogene. 2013;32(21):2601-2613.  
+
+64. Huber FM, Greenblatt SM, Davenport AM, Martinez C, Xu Y, Vu LP, et al. Histone-binding of DPF2 mediates its repressive role in myeloid differentiation. Proc Natl Acad Sci USA. 2017;114(23):6016-6021.  
+
+65. Shepherd MS, Li J, Wilson NK, Oedekoven CA, Li J, Belmonte M, et al. Single-cell approaches identify the molecular network driving malignant hematopoietic stem cell self-renewal. Blood. 2018;132(8):791-803.  
+
+66. Tognon R, Gasparotto EP, Neves RP, Nunes NS, Ferreira A, Palma PV, et al. Deregulation of apoptosis-related genes is associated with PRV1 overexpression and JAK2 V617F allele burden in Essential Thrombocythemia and Myelofibrosis. J Hematol Oncol. 2012;5:2.  
+
+67. Waibel M, Solomon VS, Knight DA, Ralli RA, Kim SK, Banks KM, et al. Combined targeting of JAK2 and Bcl-2/Bcl-xL to cure mutant JAK2-driven malignancies and overcome acquired resistance to JAK2 inhibitors. Cell Rep. 2013;5(4):1047-1059.  
+
+68. Meyer SC, Levine RL. Molecular pathways: molecular basis for sensitivity and resistance to JAK kinase inhibitors. Clin Cancer Res. 2014;20(8):2051-2059.  
+
+69. Karakashev S, Zhu H, Wu S, Yokoyama Y, Bitler B, Park PH, et al. CARM1-expressing ovarian cancer depends on the histone methyltransferase EZH2 activity. Nat Commun. 2018;9(1):631.
+
+<--- Page Split --->
+
+70. MaFarland J, Ho Z, Kugener G, Dempster J, Montgomery P, Bryan J, et al. Improved estimation of cancer dependencies from large-scale RNAi screens using model-based normalization and data integration. Nat Commun. 2018;9(1):4610. 
+
+## Figures 
+
+<center>Figure 1</center> 
+
+![](images/Figure_1.jpg)
+
+ 
+
+![](images/Figure_2.jpg)
+
+ 
+
+<center>Figure 1</center>
+
+<--- Page Split --->
+
+## JAK2 phosphorylates CARM1 in vitro  
+
+(A) JAK2 phosphorylates CARM1 in an in vitro kinase assay, in which active JAK2 kinase and recombinant CARM1 proteins were used; PAK1 was used as a positive control. The amounts of protein in the reaction are indicated. The phosphorylation of CARM1 was completely abolished by the JAK2 inhibitor (ruxolitinib, RUX) (lanes 7 and 8).  
+
+(B) Peptide fragments in the mass spectrometry analysis were generated from proteolytic cleavage of CARM1 following in vitro kinase assays in the presence of active JAK2 kinase. Tyrosine-149 (Y149) along with the series of y- and b-ions, including the phosphorylated residue, is shown as the phosphorylated peptide (EESSAVQpYF).  
+
+(C) Peptide fragments in the mass spectrometry analysis were generated from proteolytic cleavage of CARM1 following in vitro kinase assays in the absence of active JAK2 kinase. The peptide fragment around Y149 residue (EESSAVQYF) is shown without phosphorylation of tyrosine, indicating no gain in molecular weight of 80 Da (i.e. the weight of PO4).  
+
+(D) Peptide fragments in the mass spectrometry analysis were generated from proteolytic cleavage of CARM1 following in vitro kinase assays in the presence of active JAK2 kinase. Tyrosine-334 (Y334) along with the series of y- and b-ions, including the phosphorylated residue, is shown as the phosphorylated peptide (GAAVDEpYFR).  
+
+(E) Peptide fragments in the mass spectrometry analysis were generated from proteolytic cleavage of CARM1 following in vitro kinase assays in the absence of active JAK2 kinase. The peptide fragment around Y334 residue (GAAVDEYFR) is shown without phosphorylation of tyrosine, indicating no gain in molecular weight of 80 Da.  
+
+(F) The regions containing amino acid residues 149 and Y334 are located within the core catalytic domain (residue 140-480) of CARM1. Residues 28-140 in CARM1 are highly homologous to a family of Drosophila-Enabled/vasodilator-stimulated phosphoprotein homology 1 (EVH1) domains, which specifically bind to target proline-rich sequences with low affinity and high specificity.
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 2 </center>  
+
+## JAK2-V617F promotes the tyrosine phosphorylation of CARM1 in myeloid leukemia cells  
+
+(A) The expression of CARM1 protein and Y149/Y334 phosphorylated CARM1 protein was assessed in 14 myeloid leukemia cell lines and human CD34+ cord blood cells, by immunoblotting analysis. 
+(B) Phosphorylation of Y149 and Y334 of CARM1 in HEL cells is abolished following treatment with the JAK2 inhibitor (RUX at low concentration to avoid severe apoptosis; 500 nM), as is phosphorylation of
+
+<--- Page Split --->
+
+tyrosine- 694 in STAT5.  
+
+(C) Immunoprecipitation was performed using HEL, SET2, and K562 cells that express HA-tagged CARM1, with an anti-HA antibody. Immunoblotting with anti-HA and anti-JAK2 antibodies revealed the interaction between JAK2 and CARM1.  
+
+(D) HEL and UKE-1 cells harboring homozygous JAK2-V617F mutations had phosphorylated CARM1-Y149/Y334 and (auto) phosphorylated JAK2-Y1007/Y1008, while SET2 cells harboring heterozygous JAK2-V617F mutation did not.  
+
+(E) Proteins were immunoprecipitated from HEL cell extracts that express HA-tagged CARM1, using an anti-HA antibody; immunoblotting was then performed using an anti-HA antibody, anti-phosphorylated JAK2 rabbit antibody, or anti-JAK2 mouse antibody.  
+
+(F) Extracts from JAK1 or TYK2 knockout HEL cells were immunoblotted using phosphospecific anti-CARM1, JAK2, and STAT5 antibodies.
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 3 </center>  
+
+Biochemical regulation of CARM1 enzymatic activity by tyrosine phosphorylation.  
+
+(A) Based on the crystal structures of CARM1, CARM1-Y334 phosphorylation (middle) increased the interaction of the region containing Y334 itself with substrate compared with non-phosphorylated Y334
+
+<--- Page Split --->
+
+(left). CARM1- Y149 phosphorylation (right) also increased the binding of the region containing Y149 itself with substrate.  
+
+(B) CARM1- Y334 and -Y149 phosphorylation impairs the loss of binding between CARM1 methionine-259 (Met259) and substrate. The decreased Met259 binding increases the interaction of glutamic acid-257 (Glu257) and glutamic acid-266 (Glu266) with substrates (middle) in the presence of Y334 phosphorylation, compared to non-phosphorylated Y334 (left). Furthermore, the loss of methionine-259 binding increases the interaction of glutamine-158 (Gln158) and aspartic acid-161 (Asn161) with substrates in the presence of Y149 phosphorylation (right).  
+
+(C) The mutant CARM1 Y149F, Y334F, and Y149F/Y334F proteins show reduced methyltransferase activity for histone H3.1, compared to wild-type (WT) CARM1, in in vitro methylation assay. CARM1 and MYC protein lanes are shown to demonstrate equal loading (Top). Autoradiograph of the methylated \(^3\mathrm{H}\) -histone H3.1 (Middle). Coomassie staining shows histone H3.1 used in the assay (Bottom). Western blotting shows the relative amount of CARM1 and MYC.  
+
+(D) Phosphorylated CARM1 shows increased methyltransferase activity for histone H3.1 in in vitro methylation assay (Top). Autoradiograph of the methylated \(^3\mathrm{H}\) -histone H3.1 (Middle). Coomassie staining shows histone H3.1 used in the assay (Bottom). Western blotting shows the relative amount of CARM1 and phosphorylated CARM1.  
+
+(E) CARM1-Y149 and -Y334 localize at dimerization arm and helix \(\alpha X,\) respectively. These residues lie close to the dimerization interface in the modeled CARM1 structure.  
+
+(F) Co-immunoprecipitation of HA- and MYC-tagged CARM1 from 293T cell extracts transiently transfected with plasmid expressing HA-tagged WT and MYC-tagged WT or mutant CARM1. HA-tagged WT CARM1 was immunoprecipitated from cell extracts with anti-HA antibodies, and then the coimmunoprecipitated MYC-tagged CARM1 was probed with anti-MYC antibodies. The levels of MYC-tagged CARM1 Y149F and Y149F/Y334F from the HA immunoprecipitates were lower than those of MYC-tagged CARM1 WT.  
+
+(G) Subcellular fractionations of HEL cells and UKE-1 cells were immunoblotted using anti-total CARM1, CARM1-pY334, and -pY149 antibodies; cytoplasmic extraction, CYE; nuclear soluble extraction, NSE; and chromatin-bound extraction, CBE. The left lane represents the expression levels of the indicated proteins of whole-cell lysates (WCE). The bar graph on the right represents the ratio of cytoplasmic, nuclear, or chromatin-binding CARM1-pY334 and -pY149 to total cytoplasmic, nuclear, or chromatin-binding CARM1, respectively (bands inside the boxes). Data represent the mean ± SD. n=3, **p<0.01, *** p<0.001; unpaired two-tailed Student's t-test.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Figure 4 </center>  
+
+## Identification of RUNX1 as CARM1-interacting proteins by Proximity BiolD proteomics  
+
+(A) Schematic diagram demonstrating BiolD (proximity-dependent biotin identification) approach for the identification of CARM1-interacting proteins.
+
+<--- Page Split --->
+
+(B) Metascape enrichment network visualization showing the intra-cluster and inter-cluster similarities of enrichment terms, up to nine terms per clusters, in HEL cells. Terms are defined according to GO/KEGG terms, canonical pathways, and hallmark gene sets.  
+
+The connecting pairs of nodes are created with Kappa score \(>0.3\) . Terms containing more genes tend to have a more significant \(P\) - value; the darker color of the node indicates the more statistically significant \(P\) - value.  
+
+(C) Scatter plot comparing mean-fold change for CARM1-BirA\* fusion vs. BirA\* alone with abundance in published negative control AP-MS datasets (%CRAPome). Green dots represent proteins (i) with a cutoff frequency of \(\geq 80\%\) CRAPome and the average spectral count fold change \(\geq 1.2\) or (ii) with a cutoff frequency of \(< 80\%\) CRAPome but the average spectral count fold change \(\geq 3.0\). Known substrates of CARM1 are indicated as red, and E2F-targets, histone binding proteins, and MYC-targets are shown in yellow, blue, and violet, respectively. See also supplemental data 1 (HEL cells) and 2 (K562 cells).  
+
+(D) Proteins were immunoprecipitated from HEL cell extracts that express MYC-tagged CARM1 (WT and non-phosphorylatable mutants), using an anti-MYC antibody; immunoblotting was then performed using an anti-MYC antibody and anti-RUNX1 mouse antibody.  
+
+(E) Doxycycline-inducible short hairpin RNAs (shRNAs) directed against CARM1 decreased CARM1 protein levels and the ADMA levels of RUNX1-R223 and -R319 as well as well-established targets, such as PABP1-R455/R460 and BAF155-R1064.  
+
+(F) Clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR-associated protein-9 (Cas9)-mediated non-phosphorylatable CARM1 mutants decreased the ADMA levels of RUNX1-R223 and -R319 as well as PABP1 and BAF155.  
+
+(G) Expression of total and asymmetry dimethylated RUNX1, PABP1, and BAF155 were assessed in HEL cells treated with RUX 250 nM or DMSO control for 5 days. Fresh media with RUX or DMSO was added on day 0, 2, and 4. Quantification of the ADMA levels of RUNX1, BAF155, and PABP1 at 5 days after RUX treatment are shown in the right panels. Data represent the mean ± SD. \(n = 3\) , \(*p< 0.05\) , \(**p< 0.01\) ; one-way ANOVA.  
+
+(H) The levels of ADMA RUNX1, BAF155, and PABP1 were measured in HEL cells treated with RUX (or DMSO) after EPZ025654 treatment for 5 days followed by a wash-out phase lasting up to 3 days (labeled as day 8).
+
+<--- Page Split --->
+![](images/Figure_unknown_0.jpg)
+
+<center>Figure 5 </center>  
+
+## Functional analysis of non-phosphorylatable mutant CARM1  
+
+(A) Cell proliferation assays of non-phosphorylatable CARM1 mutant knock-in HEL cells, where cell numbers were measured using cell-counting apparatus. \(\mathrm{n} = 3\) , \(^{**}p< 0.01\) .
+
+<--- Page Split --->
+
+(B) The flow cytometry analysis of BrdU-stained HEL cells expressing non-phosphorylatable CARM1 mutants. Mean fractions ± s.d. in sub G1, G0/G1, S, and G2/M populations. n=3, \*p<0.05, \*\*p<0.01.  
+
+(C) Heatmap shows the differentially expressed coding genes at 2-fold cut-off, representing replicates of HEL cells expressing CARM1 WT or two independent cells expressing CAMR1-Y149F/Y334F double mutation (Y149F/Y334F-1 and Y149F/Y334F-2).  
+
+(D) Gene ontology analysis of significant downregulated genes in HEL cells expressing CARM1-Y149F/Y334F compared to CARM1-WT.  
+
+(E) Heatmaps of FDR (q < 0.25) values from GSEA of hallmark gene set collections.  
+
+(F) Representative GSEA plot depicting the downregulation of G2/M checkpoint and apoptosis/antiapoptosis pathways.  
+
+(G) Volcano plot representing gene expression changes triggered by CARM1-Y149F/Y334F mutation knock-in in HEL cells. Genes associated with apoptosis/anti-apoptosis, G2/M checkpoints, stemness in hematopoietic stem cells, and RUNX1-target are shown in red, yellow, bale, and violet, respectively. The red dots indicate upregulated genes in HEL cells expressing CARM1-Y149F/Y334F, whereas the blue dots indicated downregulated genes.  
+
+(H) Representative GSEA plot depicting the downregulation of "hematopoietic stem cell up" signature.  
+
+(I) qRT-PCR analysis showing BMI-1 and CD34 in HEL cells expressing CARM1 WT, Y149F, Y334F, and Y149F/Y334F mutation. Mean and SD are expressed as a percentage of HPRT-1 expression.  
+
+(J) qRT-PCR analysis showing ID2 and MIR144 in HEL cells expressing CARM1 WT, Y149F, Y334F, and Y149F/Y334F mutation. Mean and SD are expressed as a percentage of HPRT-1 expression.  
+
+(K) Heat map of total R319-RUNX1 or asymmetrically dimethylated R319-RUNX1 binding tag intensity by ChIP-seq analysis for HEL cells expressing CARM1 WT, Y149F, Y334F, or Y149F/Y334F mutant proteins.  
+
+(L) ChIP-seq analyses were performed to assess total RUNX1 and asymmetrically dimethylated R319-RUNX1 chromatin binding. Target occupancies at the ID2 gene are shown in IGV genome browser tracks.
+
+<--- Page Split --->
+![](images/Figure_unknown_1.jpg)
+
+<center>B </center>  
+
+![](images/Figure_6.jpg)
+
+<center>C </center>  
+
+![](images/Figure_7.jpg)
+
+<center>Figure 6 </center>  
+
+## Inhibition of CARM1 targets cells harboring phosphorylated CARM1 mediated by JAK2-V617F mutant  
+
+(A) Western blot assessment of phosphorylation in JAK2, STAT5, ERK, and AKT, and asymmetric demethylated arginine in RUNX1, BAF155, and PABP1 in HEL and UKE-1 cells treated with 5 days with increasing concentrations of EPZ025654 (μM).
+
+<--- Page Split --->
+
+(B) Excess over Bliss plots (Bliss method) showing synergistic effects between EPZ025654 and RUX were visualized in the calculated 2D synergy maps. Red and green areas represent synergistic (synergy score \(> + 10\) ), addictive (synergy score \(0 + 10\) ), and antagonistic effect ( \(< - 10\) ), respectively. In 2D synergy maps, white rectangles show the maximum synergy area in each cell.  
+
+(C) The colony formation of HEL, UKE-1, and SET2 cells treated with DMSO (control), RUX, EPZ025654, or a combination of RUX and EPZ025654. The concentration of RUX was applied based on the IC50 values for each cell line. Representative pictures of colonies on semi-solid methylcellulose media are shown on the upper panels. Quantification of the number of colonies at 14 days after plating are shown in the lower panels. Data represent the mean ± SD. \(n = 4\) , \(p< 0.05\) , \(**p< 0.01\) , \(***p< 0.001\) ; one-way ANOVA.  
+
+![PLACEHOLDER_28_0]
+
+<center>Figure 7 </center>  
+
+## A schematic model showing JAK2-CARM1 axis  
+
+JAK2- V617F mutant kinase, when activated by JAK2, JAK1, or TYK2, strongly phosphorylates CARM1- Y149 and - Y334, increasing its methyltransferase activity and the asymmetrical dimethylation of its substrates, including histone 3 and RUNX1. CARM1 phosphorylation promotes cell- cycle progression and inhibits apoptosis, and regulates the genes associated with stemness (BMI- 1).  
+
+## Supplementary Files
+
+<--- Page Split --->
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- NatCommunSupplementalData.xlsx- NatCommunSupplementalFigXTextXTab.docx
+
+<--- Page Split --->

preprint/preprint__7ea08d3a6b99d16e7f9a32724c54deb41c2325e4e80d5feab1f63b193d958db8/preprint__7ea08d3a6b99d16e7f9a32724c54deb41c2325e4e80d5feab1f63b193d958db8_det.mmd

@@ -0,0 +1,684 @@
+<|ref|>title<|/ref|><|det|>[[44, 107, 905, 177]]<|/det|>
+# Tyrosine phosphorylation of CARM1 promotes its enzymatic activity and alters its target specificity  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 228, 238]]<|/det|>
+Hidehiro Itonaga  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 244, 217, 283]]<|/det|>
+Adnan Mookhtiar  Harvard University  
+
+<|ref|>text<|/ref|><|det|>[[44, 290, 228, 330]]<|/det|>
+Sarah Greenblatt  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 336, 399, 376]]<|/det|>
+Fan Liu  https://orcid.org/0000- 0003- 4142- 5139  
+
+<|ref|>text<|/ref|><|det|>[[44, 382, 228, 422]]<|/det|>
+Concepcion Martinez  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 428, 228, 468]]<|/det|>
+Renata Grozovsky  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 474, 228, 514]]<|/det|>
+Daniel Bilbao  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 520, 228, 560]]<|/det|>
+Masai Rains  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 566, 250, 606]]<|/det|>
+Pierre-Jacques Hamard  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 613, 120, 630]]<|/det|>
+Jun Sun  
+
+<|ref|>text<|/ref|><|det|>[[44, 635, 228, 675]]<|/det|>
+Afoma Umeano  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 681, 456, 722]]<|/det|>
+Stephanie Duffort  Miller School of Medicine, University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 728, 585, 769]]<|/det|>
+Chuan Chen  University of Miami  https://orcid.org/0000- 0001- 5701- 0857  
+
+<|ref|>text<|/ref|><|det|>[[44, 775, 228, 815]]<|/det|>
+Na Man  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 821, 228, 860]]<|/det|>
+Gloria Mas  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 866, 585, 907]]<|/det|>
+Stephan Schurer  University of Miami  https://orcid.org/0000- 0001- 7180- 0978
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[44, 42, 280, 88]]<|/det|>
+Stephen Nimer snimer@med.miami.edu  
+
+<|ref|>text<|/ref|><|det|>[[52, 115, 584, 135]]<|/det|>
+University of Miami https://orcid.org/0000- 0003- 2439- 7586  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 176, 102, 194]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 214, 800, 234]]<|/det|>
+Keywords: JAK2- V617F, CARM1, protein arginine methyltransferases, chromatin, RUNX1  
+
+<|ref|>text<|/ref|><|det|>[[44, 252, 296, 271]]<|/det|>
+Posted Date: July 12th, 2022  
+
+<|ref|>text<|/ref|><|det|>[[44, 290, 475, 309]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 1807575/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 327, 911, 370]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 388, 531, 408]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 444, 914, 487]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on April 22nd, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 47689- 4.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 159, 68]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[41, 82, 958, 331]]<|/det|>
+Coactivator- associated arginine methyltransferase 1 (CARM1) is overexpressed in cancer, and it has emerged as an important target in acute myeloid leukemia and other hematologic malignancies. Janus kinase 2 (JAK2), that is activated by mutation in a variety of myeloid malignancies, can dictate chromatin structure via multiple effects. Here, we find that the hyperactivated JAK2- V617F mutant kinase phosphorylates CARM1, increasing its methyltransferase activity and altering its target specificity. Phospho- CARM1 binds and methylates the RUNX1 transcription factor, and the asymmetric dimethylation of R223 and R319 in RUNX1 is lost in engineered to express only non- phosphorylatable CARM1 mutant proteins in JAK2- V617F+ cell lines. The decreased CARM1 activity found in these cell lines impairs cycle progression and induces apoptosis. We have established a link between activated JAK2 and CARM1 activity, and demonstrate that dual targeting of JAK2 and CARM1 is more effective than monotherapy in phospho- CARM1+ cell lines.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 355, 207, 380]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[41, 394, 953, 597]]<|/det|>
+Protein arginine methylation is an essential protein post- translational modification, with \(\sim 7\%\) of arginine residues being methylated, which is comparable to the \(9\%\) of serine residues that are phosphorylated and the \(7\%\) of lysine residues that are ubiquitinated [1]. Protein arginine methyltransferases (PRMTs) catalyze monomethylation, asymmetric dimethylation, or symmetric dimethylation reactions on arginine residues [2], and are classified as class I (asymmetric dimethyl arginine; ADMA), class II (symmetric dimethyl arginine; SDMA) and class III methyltransferases (monomethyl arginine; MMA) [2- 4]. PRMTs are ubiquitously expressed and they regulate multiple cellular processes, including transcription, RNA splicing, DNA replication, DNA repair, protein translation, and cellular metabolism, thereby affecting cell growth, proliferation and differentiation [5- 11].  
+
+<|ref|>text<|/ref|><|det|>[[41, 613, 955, 794]]<|/det|>
+Coactivator- associated arginine methyltransferase 1 (CARM1), also known as PRMT4, was originally identified as a coactivator for steroid hormone receptors [12]. CARM1 is a type I protein arginine methyltransferase (PRMT), that catalyzes the asymmetric dimethylation of arginine residues in histones, such as H3R17 and H3R26, which are thought to promote transcription [13, 14]. Furthermore, CARM1 also catalyzes the ADMA of non- histone substrates, including transcription factors like RUNX1 [15], histone acetyltransferases (e.g. p300 and CBP) [16- 18], the steroid receptor co- activator AIB1 [19, 20], RNA binding proteins (e.g. PABP1) [21], RNA splicing factors (e.g. SAP49, SmB, and U1C) [22], and components of the SWI/SNF chromatin remodeling complex (e.g. BAF155) [16, 23].  
+
+<|ref|>text<|/ref|><|det|>[[41, 810, 944, 945]]<|/det|>
+We previously demonstrated that CARM1 blocks the myeloid differentiation of normal hematopoietic stem/progenitor cells (HSPCs) by promoting the assembly of a repressive RUNX1 complex [15], and that Carm1 knockout in adult mouse HSPCs prevents the development of acute myeloid leukemia (AML), driven by either the AML1- ETO or MLL- AF9 oncogenes, but only modestly decreases long- term hematopoietic stem cell (HSC) numbers [24]. Carm1 knockout also abrogated the maintenance of AML, suggesting that CARM1 inhibition could have therapeutic efficacy in AML [24]. CARM1 is overexpressed
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 951, 88]]<|/det|>
+in a variety of cancers, including breast, lung, colorectal, liver, and prostate cancer [25–33] and because it drives key oncogenic processes, it could be a therapeutic target in these diseases as well.  
+
+<|ref|>text<|/ref|><|det|>[[40, 105, 947, 445]]<|/det|>
+There have been several reports of post- translational modifications (PTMs) of CARM1, including its serine phosphorylation and arginine auto- methylation [34–37]. Yet, little is known about how or whether post- translational modifications of CARM1 affect its role in promoting cancer development. We have been examining how oncogenic signaling pathways affect the activity of epigenetic modifiers, including tyrosine kinases such as Janus- activated kinases (JAK) family (JAK1, JAK2, JAK3, and TYK2) which are activated by cytokine receptors and other cell surface receptors [38–42]. Activated JAK2 signals through downstream effectors including the signal transducers and activators of transcription (STAT) transcription factors (TFs), and the Ras- mitogen- activated protein kinase (MAPK) and phosphoinositide 3- kinase (PI3K) pathways, which regulate hematopoietic cell differentiation, cell proliferation, and apoptosis [39, 43, 44]. A single somatic mutation, V617F (exon 14) in the JH2 domain of JAK2, which disrupts the JH1- JH2 autoinhibitory interaction, leading to JAK2 hyperactivation [45–47], is a common hallmark of the BCR/ABL1- negative myeloproliferative neoplasms. It is found in > 95% of polycythemia vera patients, with homozygous mutations found in patients with longstanding disease. It is also found in \(\sim 30–50\%\) of patients with essential thrombocythemia and primary myelofibrosis, and in \(\sim 1–4\%\) of adult AML and myelodysplastic syndrome patients [48–51].  
+
+<|ref|>text<|/ref|><|det|>[[41, 460, 949, 664]]<|/det|>
+We now report that when it is phosphorylated and activated, the JAK2- V617F kinase phosphorylates CARM1 on tyrosine- 149 and – 334 (which are located within the CARM1 catalytic domain), promoting its methyltransferase activity and its nuclear and chromatin localization. Based on multiple in vitro and in vivo studies, we find that CARM1 phosphorylation alters its substrate specificity and its effects on CARM1 target gene selection. We also see important biological differences in the ability of phosphorylatable vs. non- phosphorylatable CARM1 to support leukemia cell proliferation and demonstrate the therapeutic relevance of targeting both JAK2- V617F and CARM1. These results provide novel insights into the regulation of chromatin structure by tyrosine kinases and into the pathogenesis of myeloid neoplasms.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 688, 144, 712]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 727, 600, 748]]<|/det|>
+## Tyrosine residues in CARM1 are phosphorylated by JAK2 in vitro  
+
+<|ref|>text<|/ref|><|det|>[[41, 765, 953, 946]]<|/det|>
+To determine whether CARM1 is a direct substrate of JAK2, and identify potential CARM1 tyrosine phosphorylation sites, we performed cell- free in vitro kinase assays using GST- tagged CARM1 protein as the substrate for a recombinant active form of JAK2 kinase, using recombinant PAK1 (a known JAK2 substrate) as a positive control (Fig. 1A). The JAK2- dependent tyrosine phosphorylation of GST- tagged CARM1 was readily identified (lanes 4 and 5), as incorporation of a JAK2 inhibitor (ruxolitinib, RUX) in the kinase assay completely abrogated CARM1 phosphorylation (lane 7). We next identified tyrosine- 149 (Y149) and tyrosine- 334 (Y334) as the sites of JAK2 phosphorylation by subjecting the in vitro phosphorylated, recombinant GST- tagged CARM1 protein to mass spectrometry analysis (Fig. 1B- E and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 953, 88]]<|/det|>
+S1A). These tyrosine residues are located within the core catalytic domain of CARM1 (Fig. 1F), and highly conserved in CARM1, from African clawed frog to human (Figure S1B and S1C).  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 105, 662, 126]]<|/det|>
+## CARM1 phosphorylation is mediated by JAK2 in myeloid leukemia cells  
+
+<|ref|>text<|/ref|><|det|>[[41, 144, 960, 301]]<|/det|>
+We generated two phospho- tyrosine specific rabbit polyclonal antibodies, against either Y149 phosphorylated or Y334 phosphorylated CARM1 (Figure S2A); these antibodies recognize in vitro phosphorylated recombinant GST- tagged CARM1 protein, but not the unphosphorylated protein, in a dose- dependent manner (Figure S2B and S2C). Furthermore, these antibodies recognize MYC- tagged wild- type (WT) CARM1, but not the non- phosphorylatable MYC- tagged CARM1 mutant proteins (that contain Y149F, Y334F, or Y149F/Y334F amino acid substitutions), based on the in vitro assays (with or without active JAK2 kinase) using the immunoprecipitants from K562 cell lysates (Figure S2D).  
+
+<|ref|>text<|/ref|><|det|>[[41, 317, 955, 548]]<|/det|>
+We compared the level of CARM1 protein expression and CARM1- Y149 and - Y334 phosphorylation in 14 myeloid leukemia cell lines, using normal human \(\mathrm{CD34^{+}}\) cord blood (CB) cells as control. As previously described [15, 24], we found that CARM1 expression levels are higher in nearly all of these myeloid leukemia cell lines, than in the \(\mathrm{CD34^{+}}\) CB cells (Figure 2A). HEL cells, that express JAK2- V617F, showed the highest level of phosphorylated CARM1- Y149 and - Y334 (lane 2), while SET2 cells, which also express JAK2- V617F, showed abundant CARM1 protein but a lower relative amount of phosphorylated CARM1 protein (lane 3) than the HEL cells. K562 cells (lane 8) had abundant CARM1, but less phosphorylated CARM1. Treating HEL cells with the JAK2 inhibitor (RUX, 500 nM) decreased the phosphorylation of CARM1- Y149 and - Y334, and the phosphorylation of STAT5, confirming that JAK2 is the relevant kinase phosphorylating CARM1 (Figure 2B and S2E).  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 564, 498, 584]]<|/det|>
+## Activated JAK2 binds to and phosphorylates CARM1  
+
+<|ref|>text<|/ref|><|det|>[[41, 601, 955, 919]]<|/det|>
+To determine whether JAK2 protein interacts with CARM1 in vivo, we transduced HEL, SET2, and K562 cells with an HA- tagged CARM1 construct. HA- tagged CARM1 was then immunoprecipitated, using an anti- HA antibody, followed by immunoblotting with an anti- JAK2 antibody. We were able to detect the direct binding of JAK2 to CARM1 in HEL cells and to a lesser extent in SET2 cells, but weakly in K562 cells (Figure 2C). To determine what accounts for these differences, we focused on the phosphorylation status of JAK2 in these cells, as HEL cells have a bi- allelic JAK2- V617F mutation, while SET2 cells have a mono- allelic JAK2- V617F mutation (Figure S3A). Biallelic mutations lead to a higher level of JAK2 autophosphorylation, leading us to hypothesize that the bi- allelic JAK2- V617F mutation in HEL cells may trigger the higher level of CARM1 phosphorylation. First, we confirmed that HEL cells have the highest level of JAK2 phosphorylation among the various myeloid leukemia cell lines tested, including SET2 cells (Figure S3B). UKE- 1 cells, which carry a bi- allelic JAK2- V617F mutation, also showed a high level of JAK2 auto- phosphorylation, and of CARM1- Y149/Y334 phosphorylation (Figure 2D). We then found that none of the other JAK family members (JAK1, JAK3 or TYK2) bound to CARM1, even when CARM1 was overexpressed in HEL cells (data not shown).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[40, 44, 950, 338]]<|/det|>
+We next examined whether JAK2 auto- phosphorylation affected the binding of JAK2 to CARM1 or its kinase activity over CARM1, using an anti- HA antibody for the immunoprecipitation and an antibody against Y1007/Y1008 phosphorylated JAK2 for immunoblotting. Phosphorylated JAK2 bound CARM1 in HEL cells that overexpress HA- tagged CARM1 (Figure 2E). Given studies showing that JAK1 and TYK2 activity can transphosphorylate JAK2 [52], we generated JAK1 and TYK2 knockout (KO) HEL cells using the clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR- associated protein- 9 (Cas9) nuclease system. KO of either JAK1 or TYK2 decreased the phosphorylation of JAK2 (by \(18 - 28\%\) ) and CARM1 (by \(38 - 55\%\) ) (Figure 2F), but neither KO reduced the binding of JAK2 to CARM1 (Figure S4A). To confirm that the binding of JAK2 to CARM1 is JAK2- phosphorylation independent, we used a type II JAK2 inhibitor, CHZ868 [53]. CHZ868 treatment abrogated JAK1- mediated JAK2- Y1007/Y1008 phosphorylation (Figure S4B), but it did not reduce the binding of JAK2 to CARM1. Taken together, these results indicate that JAK2 activation, through transphosphorylation by JAK1 and TYK2, enhances JAK2 tyrosine kinase activity against CARM1, without affecting the binding of JAK2 to CARM1.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 354, 701, 375]]<|/det|>
+## Tyrosine phosphorylation of CARM1 increases its methyltransferase activity  
+
+<|ref|>text<|/ref|><|det|>[[40, 391, 953, 760]]<|/det|>
+To better understand how phosphorylation of CARM1 affects its substrate binding and methyltransferase activity, we analyzed published crystal structures of CARM1 and PABP1. CARM1- Y149 and - Y334 phosphorylation are predicted to increase the binding of CARM1 to its unmethylated substrates (Figure 3A and 3B). Y149 and Y334 phosphorylation could also affect the methyltransferase activity of CARM1, so we purified MYC- tagged CARM1 protein from 293T cells transfected with WT or mutant CARM1 (using the non- phosphorylatable Y149F, Y334F, and Y149F/Y334F CARM1 proteins) and examined the ability of these CARM1 proteins to methylate histone H3.1 arginine- 17 (R17) in vitro and in vivo. Upon incubation of purified CARM1 with S- [methyl- \(^{14}\mathrm{C}]\) - adenosyl- methionine and recombinant histone H3.1 in an in vitro methylation assay, while WT- CARM1 had significant methyltransferase activity against H3.1, none of the non- phosphorylatable CARM1 proteins had methyltransferase activity against histone H3.1 (Figure 3C). Next, we examined whether incubating CARM1 with active JAK2 kinase affected its ability to methylate histone H3.1 in vitro. We confirmed the phosphorylation of CARM1, and then incubated histone H3.1 with S- [methyl- \(^{14}\mathrm{C}]\) - adenosyl- methionine and phosphorylated or non- phosphorylated CARM1. The addition of JAK2 kinase increased CARM1 methyltransferase activity on histone H3.1 (Figure 3D, lane 9 vs. lane 7). The mass spectrometry analysis further confirmed the increased methylation level of histone H3R17 by phosphorylated CARM1, compared to non- phosphorylated CARM1 (Figure S5).  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 775, 587, 796]]<|/det|>
+## The Y149F mutation in CARM1 impairs it's in vivo dimerization  
+
+<|ref|>text<|/ref|><|det|>[[42, 813, 953, 949]]<|/det|>
+CARM1 dimer formation involves interactions between the so- called dimerization arm (residues 300- 338) and helices \(\alpha \mathrm{X}\) , \(\alpha \mathrm{Y}\) , \(\alpha \mathrm{Z}\) , \(\alpha \mathrm{A}\) , and \(\alpha \mathrm{B}\) (residues 144- 232) [54]. Given that the sites of CARM1 phosphorylation (Y149 and Y334) are located within regions involved in CARM1 dimerization (Figure 3E), we demonstrated that the JAK2 inhibitor (RUX, \(10 \mu \mathrm{M}\) ) decreased the dimerization of CARM1 in 293T cells co- transfected with HA- tagged WT- CARM1 and MYC- tagged CARM1 constructs (Figure S6A). To investigate whether each tyrosine phosphorylation site in CARM1 affects dimerization, we co- transfected
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 45, 944, 180]]<|/det|>
+HA- tagged WT- CARM1 and MYC- tagged CARM1 (WT and Y- to- F mutated) constructs into 293T cells, immunoprecipitated with anti- HA antibodies, and probed with anti- MYC antibodies. MYC- tagged CARM1- WT and - Y334F mutant CARM1 were readily detectable in the HA immunoprecipitates (Figure 3F, lane 7 and 8). However, the Y149F mutant and the Y149F/Y334F double mutant CARM1 proteins reduced the binding ability to HA- tagged WT, compared to MYC- tagged WT CARM1 (Figure 3F, lane 9 and 10 vs. lane 7). Thus, it appears that phosphorylation of Y149 in CARM1 can promotes its dimerization.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 196, 617, 217]]<|/det|>
+## CARM1 tyrosine phosphorylation promotes its nuclear localization  
+
+<|ref|>text<|/ref|><|det|>[[42, 234, 953, 390]]<|/det|>
+We investigated whether phosphorylation affects the subcellular localization of CARM1 using an anti- CARM1 antibody. CARM1 localizes mainly in the cytoplasm of HEL, SET2, and K562 cells, with some CARM1 found in the nuclear soluble fraction (Figure S6B). Then, using CARM1 phospho- specific antibodies, and HEL and UKE- 1 cells, we found a significantly higher proportion of Y149 and Y334 phosphorylated CARM1 in the nucleus, and in the chromatin fraction of both HEL cells and UKE- 1 cells (Figure 3G). These results indicated that the tyrosine phosphorylation of CARM1 contributes to its nuclear and chromatin localization.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 408, 528, 428]]<|/det|>
+## Identification of CARM1-interacting proteins using BiolD  
+
+<|ref|>text<|/ref|><|det|>[[41, 444, 940, 853]]<|/det|>
+To capture the proteins which are strongly, weakly, or transiently associated with phosphorylated CARM1, we engineered HEL cells (with abundant phosphorylated CARM1) and K562 cells (with primarily non- phosphorylated CARM1) to express the proximity- dependent biotin identification (BiolD) system (Figure 4A and S7A). Using shotgun mass spectrometry, we identified 128 and 60 proteins that significantly interact with CARM1 in HEL and K562 cells, respectively (supplemental data and Figure S7B). CARM1- associated proteins clustered into groups of proteins identified as being involved in purine nucleotide metabolism, E2F targets, MYC targets, histone and chromatin binding proteins, and ribonucleoprotein complex biogenesis, based on enriched pathway analysis of Metascape (metascape.org) for both cell lines (Figure 4B and S7C). We identified many previously known CARM1- associated proteins or substrates, including NUDT4, ADAR, RUNX1, SON, and CARM1 in HEL cells [14, 15] (Figure 4C), and confirmed the high intensity of RUNX1 fragments interacting with CARM1- BirA\* fusion in HEL cells but not in K562 cells, using target mass spectrometry (Figure S7D), which suggests that CARM1 phosphorylation enhances the binding of CARM1 to RUNX1. To test this hypothesis, we generated HEL cells overexpressing MYC- tagged CARM1 (WT, Y334F, Y149F, and Y149F/Y334F mutants), and immunoprecipitated the WT or non- phosphorylatable mutant proteins, using an antibody against the MYC- tag. WT- CARM1, but none of the non- phosphorylatable CARM1 mutants were able to pull down RUNX1 (Figure 4D), confirming the role of CARM1 phosphorylation in regulating this interaction with RUNX1.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 870, 688, 890]]<|/det|>
+## RUNX1 R223 and R319 are strongly methylated by phosphorylated CARM1  
+
+<|ref|>text<|/ref|><|det|>[[42, 908, 936, 951]]<|/det|>
+Having identified arginine- 223 (R223) in RUNX1 as a site of CARM1 asymmetric dimethylation [15], we examined whether the tyrosine phosphorylation of CARM1 affects its ability to methylate RUNX1. Using
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 45, 940, 224]]<|/det|>
+mass spectrometry and an in vitro methylation assay, we confirmed R223 as a CARM1 target site and also identified arginine- 319 (R319) in RUNX1 as another potential CARM1 methylation site (Figure S8A). We next generated an asymmetric dimethylation specific anti- RUNX1- R319 antibody, and used this antibody and a previously published asymmetric dimethylation specific anti- RUNX1- R223 antibody [15] (Figure S8B) to show that RUNX1- R223 and - R319 are indeed methylated by CARM1 in vivo. Both residues, R223 which is located C- terminal to the RUNX1b DNA binding domain and R319 which is located within the activation domain of RUNX1b, are evolutionally conserved among mammals (Figure S8C- E).  
+
+<|ref|>text<|/ref|><|det|>[[41, 241, 951, 353]]<|/det|>
+To prove that CARM1 is the relevant methyltransferase, we created doxycycline- inducible CARM1 knockdown (KD) HEL cells, using three different small hairpin RNAs (shRNAs) that significantly decrease CARM1 RNA and protein expression (Figure S9). KD of CARM1 by all three shRNAs significantly decreased the level of RUNX1- R223me2a and RUNX1- R319me2a; they also decreased the ADMA levels of BAF155 and PABP1 (Figure 4E).  
+
+<|ref|>text<|/ref|><|det|>[[41, 369, 950, 573]]<|/det|>
+To define the role of CARM1 phosphorylation on the asymmetric dimethylation of RUNX1 (and its other substrates), we generated isogenic HEL cells lines carrying homozygous CARM1 non- phosphorylatable mutation (Y149F or Y334F single mutations, or Y149F/Y334F double mutation), using the CRISPR/Cas9 nuclease system (Figure S10). We then examined the HEL cells harboring non- phosphorylatable CARM1 mutations (Y149F or Y334F single mutations, or the Y149F/Y334F double mutation), and found that mutation of either site abrogates the asymmetric dimethylation of RUNX1- R223 and RUNX1- R319 (Figure 4F). HEL cells containing either of the non- phosphorylatable single mutations also showed decreased levels of asymmetrically dimethylated BAF155 and PABP1, while HEL cells with the CARM1 double mutation showed much lower levels of ADMA BAF155 and PABP1 than those with single mutations.  
+
+<|ref|>text<|/ref|><|det|>[[42, 589, 944, 678]]<|/det|>
+To confirm that JAK2 dependent tyrosine phosphorylation of CARM1 is responsible for the increased ADMA of RUNX1 and other CARM1 substrates in HEL cells, we treated HEL cells with RUX (Figure 4G). RUNX1- R223 and - R319 dimethylation was reduced after 5 days of RUX exposure. A modest decrease in BAF155 dimethylation and a greater decrease in PABP1 dimethylation were also seen.  
+
+<|ref|>text<|/ref|><|det|>[[41, 695, 953, 922]]<|/det|>
+To better understand the kinetics of methylation and re- methylation of CARM1 substrates, we treated HEL cells with a potent and selective CARM1 inhibitor (EPZ025654) for 5 days, and monitored cell growth and substrate methylation over the subsequent five days. While BAF155 and PABP1 methylation was restored after a 1- day EPZ025654- free period, RUNX1 re- methylation was first observed three days after EPZ025654 removal (Figure S11). We next assessed the re- methylation levels of BAF155, PABP1, and RUNX1 in HEL cells treated with RUX (or DMSO) after the removal of EPZ025654. RUX significantly impaired the re- methylation of RUNX1, but not that of BAF155 or PABP1 (Figure 4H). These results indicate the different requirements and time course of the effect of JAK2 on the ADMA of various CARM1 substrates; RUNX1 dimethylation is most sensitive to the JAK2 kinase inhibitor, while BAF155 dimethylation is only sensitive to the CARM1 inhibitor.  
+
+<|ref|>sub_title<|/ref|><|det|>[[42, 940, 490, 959]]<|/det|>
+## Biological consequence of CARM1 phosphorylation
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 44, 951, 270]]<|/det|>
+We used the non- phosphorylatable CARM1 mutant knock- in HEL cells to examine the biological effects of CARM- Y149 and - Y334 phosphorylation on cell behavior. While the single mutant cells grew normally, the double CARM1 mutant (Y149F/Y334F)- expressing HEL cells showed reduced proliferation (Figure 5A). Cell cycle analysis revealed that all three cell lines harboring homozygous single or double mutant CARM1 had a decreased S- phase fraction and an increased G2/M fraction (consistent with significant G2/M arrest) (Figure 5B). We observed an increase in apoptosis (indicated by an increased sub- G1 fraction and annexin V- positive cells) particularly in cells harboring the CARM1- Y149F/Y334F double mutation (Figure 5B and S12). These results demonstrate that phosphorylation of Y149 and Y334 in CARM1 regulates the proliferation of HEL cells, affecting both cell cycle and apoptosis to varying degrees.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 286, 630, 308]]<|/det|>
+## Distinct transcriptional profiles regulated by phosphorylated CARM1  
+
+<|ref|>text<|/ref|><|det|>[[41, 323, 951, 757]]<|/det|>
+To understand the molecular consequences of CARM1 phosphorylation, we examined the gene expression profiles of the non- phosphorylatable CARM1 mutation knock- in cells, by RNA sequencing. We identified 1,505 differentially expressed genes ( \(\geq 1.5\) - fold change and adjusted p- value \(< 0.05\) ) between CARM1 WT and - Y149F/Y334F double mutations (Figure 5C), while the single CARM1- Y149F or Y334F knock- in cells showed fewer differentially expressed genes, when compared to CARM1 WT cells (Figure S13A and S13B). Gene ontology (GO) analysis identified that genes involved in G2/M cell cycle progression and apoptosis were negatively enriched in Y149F/Y334F double mutation knock- in HEL cells (Figure 5D). We also used gene set enrichment analysis (GSEA) to identify pathways differentially regulated by the lack of CARM1 phosphorylation and found that gene sets associated with G2/M cell cycle progression and anti- apoptosis were significantly downregulated in Y149F/Y334F mutation knock- in HEL cells (Figure 5E). Heatmaps of FDR (q < 0.25) values in three non- phosphorylatable CARM1 mutation knock- in cell lines are shown in Figure 5F, which show that G2/M checkpoints- associated gene sets were downregulated in Y149F/Y334F double and Y149 single mutations, while anti- apoptosis- associated gene sets were downregulated in Y149F/Y334F double mutation cells. Among the genes associated with G2/M cell cycle progression and anti- apoptosis that were differentially expressed (adjusted p- value \(< 0.05\) and FC>1.5) and relevant in myeloid malignancies [55- 60], we found SMAD3, CCND2, KIF5B, BCL2, BCL2A1, and SATB1 (Figure 5G). We confirmed the downregulation of both G2/M checkpoint (SMAD3, CCND2, and KIF5B) and anti- apoptosis (BCL2, BCL2A1, and SATB1) genes in the double CARM1 mutation knock- in HEL cells, using qRT- PCR (Figure S13C and S13D).  
+
+<|ref|>text<|/ref|><|det|>[[42, 771, 953, 952]]<|/det|>
+Consistent with our previous reports that CARM1 promotes a DPF2- containing repressor complex that repress miR- 223 expression [3, 15], we found increased expression of MIR223 in CARM1- Y149F cells and in the CARM1- Y149F/Y334F double mutation cells (Figure S13E). As MIR223 promotes myeloid differentiation, the decrease of its expression led us to evaluate the expression of HSPC stemness- related genes by RNA sequencing. Gene sets associated with HSPC stemness were decreased in CARM1- Y149F/Y334F mutation knock- in HEL cells (normalized enrichment score of - 1.36 with FDR q- value of 0.216) (Figure 5H), and two HSPC- associated genes, CD34 and BMI- 1, were downregulated in Y149F single and Y149F/Y334F double mutations knock- in, based on qRT- PCR (Figure 5I). RUNX (RUNX1, 2, and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 944, 180]]<|/det|>
+3)- target gene sets were not significantly altered in CARM1- Y149F/Y334F mutation knock- in HEL cells (FDR q- value of 0.717) (Figure S13F); however, three RUNX- target genes (ID2, MIR144, and RNF144A) were identified as a subset of core- enrichment genes with \(\geq 1.5\) - fold change and adjusted p- value \(< 0.05\) . Given that RUNX1 regulates the transcription of ID2 and MIR144 [61, 62], we independently evaluated their gene expression by qRT- PCR, and confirmed that the Y149F/Y334F double mutation knock- in induced the upregulation of ID2 and MIR144 mRNA (Figure 5J).  
+
+<|ref|>text<|/ref|><|det|>[[41, 196, 951, 444]]<|/det|>
+To investigate the global localization of RUNX1 on chromatin, we performed ChIP- seq analyses using antibodies against asymmetrically dimethylated R319- RUNX1 and total RUNX1. As we expected, knock- in of CARM1- Y149F or - Y334F single, or - Y149F/Y334F double mutation decreased overall signals of dimethylated RUNX1- R319 and, to a lesser degree, those of total RUNX1 (Figure 5K). We found that dimethylated R319- RUNX1 shared occupancy for ID2, MIR144, and MIR223 with less effect on total RUNX1 in HEL cells expressing CARM1- WT (Figure 5L and S14). In addition, CARM1 non- phosphorylatable mutation decreased the signals of dimethylated RUNX1- R319 within 5 kb of the transcription start sites for ID2, MIR144, and MIR223 with less effect on total R319- RUNX1. These results suggested that phosphorylated CARM1 regulates the chromatin binding of asymmetrically dimethylated R319- RUNX1 and to a lesser extent unmethylated RUNX1, leading to changes in RUNX1 target- gene expression.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 461, 323, 481]]<|/det|>
+## Targeting the JAK2-CARM1 axis  
+
+<|ref|>text<|/ref|><|det|>[[42, 499, 912, 565]]<|/det|>
+We have previously shown that CARM1 KD or inhibition reduced the cell proliferation of AML (Figure S15A), inducing G0/G1 cell cycle arrest (Figure S15B) and differentiation [24], and to a lesser degree apoptosis (Figure S15C).  
+
+<|ref|>text<|/ref|><|det|>[[40, 581, 950, 900]]<|/det|>
+We assessed the efficacy of EPZ025654 on a variety of cell lines and found a dose- dependent reduction in the proliferation of UKE- 1 cells, but not HEL cells. The half maximal inhibitory concentration (IC50) values of EPZ025654 were 23.7 μM, 1.8 μM, and 186.5 nM in HEL, UKE- 1, and SET2 cells, respectively (Figure S16A), while the IC50 values for RUX were 492 nM, 269 nM, and 56 nM in HEL, UKE- 1, and SET2 cells, respectively (Figure S16B). We assessed the level of asymmetric dimethyl RUNX1, BAF155, and PABP1 in three cell lines and found that EPZ025654 significantly reduced the levels of ADMA RUNX1 (at R223 and R319) (Figure S16C); it also reduced ADMA BAF155 and PABP1 levels, in a time- dependent manner, without affecting the phosphorylation of JAK2, CARM1, STAT5, ERK, or AKT (Figure 6A), which suggests that CARM1 inhibitors inhibit cell growth via distinct signaling pathways. To determine if the combination of EPZ025654 and RUX could be synergistic on certain cell lines, we examined cell proliferation after 6 days of EPZ025654 treatment and 2 days of RUX treatment. We observed a significant synergistic inhibition effect on HEL and UKE- 1 cells (based on a positive Bliss score), but only an addictive effect on SET2 cells (Figure 6B and S16D), which suggests that a synergistic effect of JAK2 inhibition, combined with EPZ025654, may occur primarily in cells that contain phosphorylated CARM1.  
+
+<|ref|>text<|/ref|><|det|>[[42, 916, 950, 959]]<|/det|>
+We checked efficacy of single vs. combination therapy on the colony- forming capacity of these cell lines, and found a significant reduction in the colony- forming potential of HEL and UKE- 1 cells using the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[43, 45, 919, 111]]<|/det|>
+combination therapy for 14 days, compared with either drug alone, which was not seen in SET2 cells (Figure 6C). This further suggests that JAK2 inhibitors have the potential ability to sensitize cells with phosphorylated CARM1 to CARM1 inhibition.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 134, 190, 160]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[42, 174, 955, 308]]<|/det|>
+Having identified the phosphorylation of Y149 and Y334 in CARM1 as novel PTMs mediated by JAK2- V617F mutants, we show that these PTMs increase the enzymatic activity and alter the cellular localization and target specificity of CARM1. CARM1 phosphorylation enhances its ability to block differentiation, and regulate apoptosis and cell cycling by controlling G2/M checkpoints (Fig. 7). Our work highlights the importance of these regulatory effects on the phenotypes driven by CARM1 in hematologic cells that express the JAK2- V617F oncogene.  
+
+<|ref|>text<|/ref|><|det|>[[41, 325, 950, 553]]<|/det|>
+We also demonstrated auto- phosphorylation and cross- phosphorylation of Y1007/Y1008 in the JAK2 activation loop by JAK2, JAK1, or TYK2 which more strongly promotes CARM1 tyrosine phosphorylation, especially in bi- allelic JAK2- V617F mutant cells. Thus, KD of either JAK1 or TYK2 decreased the phosphorylation levels of CARM1 by altering the ability of JAK2 to phosphorylate CARM1. However, phosphorylation of JAK2- V617F on Y1007/Y1108 did not affect the binding of JAK2 to CARM1. CARM1- Y149 and - Y334 phosphorylation is promoted by the active conformation of the mutant JAK2 protein, which is stimulated in the case of JAK2- V617F by the expression of type I cytokine receptors (e.g. EpoR, MPL, or G- CSFR), and inhibited by prolonged exposure to type I JAK2- inhibitors [54, 55]. Our data demonstrate that increased CARM1 tyrosine phosphorylation is a biological marker of cells with hyperactivated JAK2 (e.g. the JAK2- V617F mutant protein) [63].  
+
+<|ref|>text<|/ref|><|det|>[[41, 568, 958, 794]]<|/det|>
+We observed increased methyltransferase activity of phosphorylated CARM1 on histone 3.1, with reduced methyltransferase activity for the Y149F and Y334F mutations, which render CARM1 non- phosphorylatable. Our crystal structural model analysis shows that Y149 and Y334 phosphorylation does increase CARM1 binding to its substrates. The phosphorylation of Y149 and Y334 in CARM1 promotes its nuclear localization, allowing enhanced binding to nuclear substrates, including histones, chromatin binding proteins, and RUNX1. The Y149F mutant CARM1 in particular, shows diminished dimerization and minimal methyltransferase activity. We confirmed RUNX1 as a key protein interaction with CARM1, by conducting a BioID screen, and showed that the non- phosphorylatable CARM1 mutant proteins did not bind RUNX1. Thus, in addition to promoting dimerization, the phosphorylation of CARM1 affects its localization, substrate binding, and methyltransferase activity.  
+
+<|ref|>text<|/ref|><|det|>[[41, 810, 955, 945]]<|/det|>
+We identified R223 and R319 of RUNX1 as residues asymmetrically dimethylated by CARM1, and showed that CARM1- Y149 and - Y334 phosphorylation enhanced the asymmetrical dimethylation of both R223- and R319- RUNX1. CARM1 can regulate hematopoietic cell differentiation through multiple mechanisms, including the generation of a repressor complex that contains asymmetrically dimethylated RUNX1- R223, and negatively regulates miR- 223 expression [3, 15, 64]. We now find that non- phosphorylatable CARM1 mutant cell lines show increased expression of miR- 223 and several other RUNX1- target genes (/D2 and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 955, 157]]<|/det|>
+MIR144). Knock- in of a non- phosphorylatable CARM1 mutation downregulated the expression of BMI- 1, which is a regulator of self- renewal that plays a role in JAK2- V617F mutant hematopoietic stem cells [65], and in other cancer stem cell phenotypes. Given its substrate targets (e.g. BAF155 and RUNX1) and gene targets (e.g. BMI- 1 and ID2), CARM1 and phospho- CARM1 play a pivotal role in hematologic malignancies with the JAK2- V617F mutation, and in other settings as well.  
+
+<|ref|>text<|/ref|><|det|>[[41, 173, 953, 375]]<|/det|>
+The non- phosphorylatable CARM1 mutation (Y149F/Y334F) knock- in HEL cells show decreased cell growth with increased cell cycle arrest and apoptosis, likely due to the downregulation of gene expression associated with G2/M progression and anti- apoptosis, including BCL2 family members (BCL2 and BCL2A1) which have been implicated in controlling apoptosis in JAK2- V617F mutant myeloid malignancies [66- 68]. Similarly, the asymmetrical dimethylation of BAF155 by CARM1 has been shown to inhibit apoptosis of ovarian cancer cells through downregulation of pro- apoptotic gene expression (DAB2, DLC1, and NOXA) [69]. We could not detect a significant difference in the expression of these genes in WT- vs. Y149F/Y334F- CARM1 expressing HEL cells (data not shown), despite similar changes in the level of ADMA BAF155, confirming the cell- context- specific effects of arginine dimethylation.  
+
+<|ref|>text<|/ref|><|det|>[[41, 392, 944, 572]]<|/det|>
+The dependency of JAK2- V617F mutant AML cells on CARM1 is consistent with our previous studies showing that CARM1 is an essential gene for the growth of myeloid leukemia cells; further evidence was provided by an RNAi screen analysis conducted as part of the Dependency Map database (https://depmap.org/portal/) [70]. Non- phosphorylatable CARM1 mutant- expressing HEL cells showed significantly decreased cell growth, suggesting some dependency of HEL cells on CARM1 phosphorylation. This led us to evaluate the efficacy of inhibiting both JAK2 and CARM1 pathway, and we found that small- molecule inhibitors targeting CARM1 (EPZ025654) sensitized JAK2- mutant cells to JAK2 inhibitors.  
+
+<|ref|>text<|/ref|><|det|>[[42, 589, 950, 655]]<|/det|>
+In conclusion, CARM1 phosphorylation mediated by hyperactivated JAK2 regulates its methyltransferase activity and is required for maximal proliferation of myeloid neoplasms. Our results suggest a potential role of targeting both JAK2 and CARM1 in JAK2- V617F mutant myeloid malignancies.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 678, 339, 705]]<|/det|>
+## Material And Methods  
+
+<|ref|>text<|/ref|><|det|>[[44, 720, 572, 740]]<|/det|>
+Methodology is described in the Supplementary Methods file.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 763, 212, 789]]<|/det|>
+## Declarations  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 803, 356, 832]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[42, 848, 953, 958]]<|/det|>
+H.I., S.M.G., and S.D.N. conceived the project. H.I. and S.D.N. designed the experiments and wrote the manuscript. H.I. conducted most of the experiments. A.K.M. and P.- J.H. assisted with sample preparation for mass spectrometry. P.- J.H. assisted with RUNX1 arginine methylation experiments. S.M.G. assisted with in vitro phosphorylation assay. A.K.M. and F.L. assisted with in vitro methylation assay. C.M., M.R., and J.S. contributed to the shRNA knockdown. R.G. assisted in immunoprecipitation and immunoblotting
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 936, 134]]<|/det|>
+assays to determine CARM1 phosphorylation in cells. J.S. assisted in ChiP- seq assays. D.B., C.M., and S.D. assisted immunoblotting to determine the subcellular fraction of CARM1. G.M.M. assisted with the library preparation for RNA- sequencing. C.C. and N.M. assisted with colony- forming assay. A.C.U. and S.S. performed the crystal structure analysis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 164, 350, 196]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[42, 210, 951, 368]]<|/det|>
+We thank the members of the Nimer lab for their assistance and thoughtful input on the manuscript; especially Delphine Prou and Lauren Ashley Whitmore. We also thank the Oncogenomics Shared Resource at Sylvester Comprehensive Cancer Center for RNA- sequencing services and the Biostatistics and Bioinformatics Shared Resource for data analysis. This work was supported by funds from Sylvester Comprehensive Cancer Center, grant R01 CA251664- 01 and 1P30CA240139- 01 from the National Cancer Institute to S.D.N., 1F31CA254232- 01 from the National Cancer Institute to A.K.M., and the Translational Research Program Grant from the Leukemia and Lymphoma Society to S.D.N.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 389, 196, 415]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[55, 430, 955, 919]]<|/det|>
+1. Larsen SC, Sylvestersen KB, Mund A, Lyon D, Mullari M, Madsen VM, et al. Proteome-wide analysis of arginine monomethylation reveals widespread occurrence in human cells. Sci Signal. 2016;9(443):rs9.  
+2. Bedford MT, Clarke SG. Protein arginine methylation in mammals: who, what, and why. Mol Cell. 2009;33(1):1-13.  
+3. Greenblatt SM, Liu F, Nimer SD. Arginine methyltransferases in normal and malignant hematopoiesis. Exp Hematol. 2016;44(6):435-441.  
+4. Yang Y, Bedford M. Protein arginine methyltransferases and cancer. Nat Rev Cancer. 2013;13(1):37-50.  
+5. Gao WW, Xiao RQ, Peng BL, Xu HT, Shen HF, H MF, et al. Arginine methylation of HSP70 regulates retinoid acid-mediated RARβ2 gene activation. Proc Natl Acad Sci USA. 2015;112(26):E3327-E3336.  
+6. Hsu JM, Chen CT, Chou CK, Kuo HP, Li LY, Lin CY, et al. Crosstalk between Arg 1175 methylation and Tyr 1173 phosphorylation negatively modulates EGFR-mediated ERK activation. Nat Cell Biol. 2011;13(2):174-181.  
+7. Liao HW, Hsu JM, Xia W, Wang HL, Wang YN, Chang WC, et al. PRMT1-mediated methylation of the EGF receptor regulates signaling and cetuximab response. J Clin Invest. 2015;125(12):4529-4543.  
+8. Liu F, Ma, F, Wang Y, Hao L, Zeng H, Jia C, et al. PKM2 methylation by CARM1 activates aerobic glycolysis to promote tumorigenesis. Nat Cell Biol. 2017;19(11):1358-1370.  
+9. Poornima G, Shah S, Vignesh V, Parker R, Rajyaguru P. Arginine methylation promotes translation repression activity of elF4G-binding protein, Scd6. Nucleic Acids Re. 2016;44(19):9358-9368.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[46, 42, 952, 92]]<|/det|>
+10. Wang YP, Zhou W, Wang J, Huang X, Zuo Y, Wang TS, et al. Arginine Methylation of MDH1 by CARM1 Inhibits Glutamine Metabolism and Suppresses Pancreatic Cancer. Mol Cell. 2016;64(4):673–687. 
+
+<|ref|>text<|/ref|><|det|>[[50, 96, 901, 160]]<|/det|>
+11. Zhao DY, Gish G, Braunschweig U, Li Y, Ni Z, Schmitges FW, et al. SMN and symmetric arginine dimethylation of RNA polymerase II C-terminal domain control termination. Nature. 2016;529(7584):48–53. 
+
+<|ref|>text<|/ref|><|det|>[[50, 167, 950, 209]]<|/det|>
+12. Chen D, Ma H, Hong H, Koh SS, Huang SM, Schurter BT, et al. Regulation of transcription by a protein methyltransferase. Science. 1999;284(5423):2174–2177. 
+
+<|ref|>text<|/ref|><|det|>[[50, 216, 920, 259]]<|/det|>
+13. Schurter BT, Koh SS, Chen D, Bunick GJ, Harp JM, Hanson BL, et al. Methylation of histone H3 by coactivator-associated arginine methyltransferase 1. Biochemistry. 2011;40(19):5747–5756. 
+
+<|ref|>text<|/ref|><|det|>[[50, 265, 900, 308]]<|/det|>
+14. Shishkova E, Zeng H, Liu F, Kwiecien NW, Herbert AS, Coon JJ, et al. Global mapping of CARM1 substrates defines enzyme specificity and substrate recognition. Nat Commun. 2017;8:15571. 
+
+<|ref|>text<|/ref|><|det|>[[50, 314, 944, 357]]<|/det|>
+15. Vu LP, Perna F, Wang L, Voza F, Figueroa ME, Tempst P, et al. PRMT4 blocks myeloid differentiation by assembling a methyl-RUNX1-dependent repressor complex. Cell Rep. 2013;5(6):1625–1638. 
+
+<|ref|>text<|/ref|><|det|>[[50, 363, 945, 406]]<|/det|>
+16. Ceschin DG, Walia M, Wenk SS, Duboé C, Gaudon C, Xiao Y, et al. Methylation specifies distinct estrogen-induced binding site repertoires of CBP to chromatin. Genes Dev. 2011;25(11):1132–1146. 
+
+<|ref|>text<|/ref|><|det|>[[50, 412, 950, 455]]<|/det|>
+17. Chevilalrd-Briet M, Trouche D, Vandel L. Control of CBP co-activating activity by arginine methylation. EMBO J. 2002;21(20):5457–5466. 
+
+<|ref|>text<|/ref|><|det|>[[50, 461, 947, 505]]<|/det|>
+18. Xu W, Chen H, Asahara H, Tini M, Emerson BM, Montminy M, et al. A transcriptional switch mediated by cofactor methylation. Science. 2001;294(5551):2507–2511. 
+
+<|ref|>text<|/ref|><|det|>[[50, 510, 926, 576]]<|/det|>
+19. Naeem H, Cheng D, Zhao Q, Underhill C, Tini M, Bedford MT, et al. The activity and stability of the transcriptional coactivator p/CIP/SRC-3 are regulated by CARM1-dependent methylation. Mol Cell Biol. 2007;27(1):120–134. 
+
+<|ref|>text<|/ref|><|det|>[[50, 582, 937, 626]]<|/det|>
+20. Feng Q, Yi Ping, Wong J, O'Malley BW. Signaling within a coactivator complex: methylation of SRC-3/AIB1 is a molecular switch for complex disassembly. Mol Cell Biol. 2006;26(21):7846–7857. 
+
+<|ref|>text<|/ref|><|det|>[[50, 632, 933, 675]]<|/det|>
+21. Lee J, Bedford MT. PABP1 identified as an arginine methyltransferase substrate using high-density protein arrays. EMBO Rep. 2002;3(3):268–273. 
+
+<|ref|>text<|/ref|><|det|>[[50, 681, 905, 725]]<|/det|>
+22. Cheng D, Côté J, Shaaban S, Bedford MT. The arginine methyltransferase CARM1 regulates the coupling of transcription and mRNA processing. Mol Cell. 2007;25(1):71–83. 
+
+<|ref|>text<|/ref|><|det|>[[50, 731, 920, 797]]<|/det|>
+23. Kowenz-Leutz Elisabeth, Pless Ole, Dittmar G, Knoblich M, Leutz A. Crosstalk between C/EBPbeta phosphorylation, arginine methylation, and SWI/SNF/Mediator implies an indexing transcription factor code. EMBO J. 2010;29(6):1105–1115. 
+
+<|ref|>text<|/ref|><|det|>[[50, 804, 930, 847]]<|/det|>
+24. Greenblatt SM, Man N, Hamard PJ, Asai T, Karl D, Martinez C, et al. CARM1 Is Essential for Myeloid Leukemogenesis but Dispensable for Normal Hematopoiesis. Cancer Cell. 2018;34(5):868. 
+
+<|ref|>text<|/ref|><|det|>[[50, 853, 907, 919]]<|/det|>
+25. Hong H, Kao C, Jeng MH, Eble JN, Koch MO, Gardner TA, et al. Aberrant expression of CARM1, a transcriptional coactivator of androgen receptor, in the development of prostate carcinoma and androgen-independent status. Cancer. 2004;101(1):83–89.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[46, 44, 955, 94]]<|/det|>
+26. Majumder S, Liu Y, Ford OH, Mohler JL, Whang YE. Involvement of arginine methyltransferase CARM1 in androgen receptor function and prostate cancer cell viability. Prostate. 2006;66(12):1292–1301.  
+
+<|ref|>text<|/ref|><|det|>[[46, 118, 956, 183]]<|/det|>
+27. Al-Dhaheri M, Wu J, Skliris GP, Li J, Higashimoto K, Wang Y, et al. CARM1 is an important determinant of ERα-dependent breast cancer cell differentiation and proliferation in breast cancer cells. Cancer Res. 2011;71(6):2118–2128.  
+
+<|ref|>text<|/ref|><|det|>[[46, 189, 956, 232]]<|/det|>
+28. Lahusen T, Henke RT, Kagan BL, Wellstein A, Riegel AT. The role and regulation of the nuclear receptor co-activator AIB1 in breast cancer. Breast Cancer Rs Treat. 2009;116(2):225–237.  
+
+<|ref|>text<|/ref|><|det|>[[46, 238, 945, 281]]<|/det|>
+29. Frietze S, Lupien M, Silver PA, Brown M. CARM1 regulates estrogen-stimulated breast cancer growth through up-regulation of E2F1. Cancer Res. 2008;68(1):301–306.  
+
+<|ref|>text<|/ref|><|det|>[[46, 287, 900, 330]]<|/det|>
+30. Kim YR, Lee BK, Park RY, Nguyen NT, Bae JA, Kwon DD, et al. Differential CARM1 expression in prostate and colorectal cancers. BMC Cancer. 2010;10:197.  
+
+<|ref|>text<|/ref|><|det|>[[46, 336, 953, 402]]<|/det|>
+31. Ou CY, LaBonte MJ, Manegold PC, So AY, Ianculescu I, Gerke DS, et al. A coactivator role of CARM1 in the dysregulation of \(\beta\)-catenin activity in colorectal cancer cell growth and gene expression. Mol Cancer Res. 2011;9(5):660–670.  
+
+<|ref|>text<|/ref|><|det|>[[46, 408, 944, 474]]<|/det|>
+32. Osada S, Suzuki S, Yoshimi C, Matsumoto M, Shirai T, Takahashi S, et al. Elevated expression of coactivator-associated arginine methyltransferase 1 is associated with early hepatocarcinogenesis. Oncol Rep. 2013;30(4):1669–1674.  
+
+<|ref|>text<|/ref|><|det|>[[46, 480, 930, 523]]<|/det|>
+33. Zhong XY, Yuan XM, Xu YY, Yin M, Yan WW, Zou SW, et al. CARM1 Methylates GAPDH to Regulate Glucose Metabolism and Is Suppressed in Liver Cancer. Cell Rep. 2018;24(12):3207–3223.  
+
+<|ref|>text<|/ref|><|det|>[[46, 529, 951, 573]]<|/det|>
+34. Feng Q, He B, Jung SY, Song Y, Qin J, Tsai SY, et al. Biochemical control of CARM1 enzymatic activity by phosphorylation. J Biol Chem. 2009;284(52):36167–36174.  
+
+<|ref|>text<|/ref|><|det|>[[46, 579, 944, 644]]<|/det|>
+35. Carascossa S, Dudek P, Cenni B, Briand PA, Picard D. CARM1 mediates the ligand-independent and tamoxifen-resistant activation of the estrogen receptor alpha by cAMP. Genes Dev. 2010;24(7):708–719.  
+
+<|ref|>text<|/ref|><|det|>[[46, 650, 936, 715]]<|/det|>
+36. Higashimoto K, Kuhn P, Desai D, Cheng X, Xu W. Phosphorylation-mediated inactivation of coactivator-associated arginine methyltransferase 1. Proc Natl Acad Sci USA. 2007;104(3):12318–12323.  
+
+<|ref|>text<|/ref|><|det|>[[46, 721, 955, 765]]<|/det|>
+37. Kuhn P, Chumanov R, Wang Y, Ge Y, Burgess RR, Xu W. Auto­methylation of CARM1 allows coupling of transcription and mRNA splicing. Nucleic Acids Res. 2011;39(7):2717–2726.  
+
+<|ref|>text<|/ref|><|det|>[[46, 771, 954, 815]]<|/det|>
+38. Ward AC, Touw I, Yoshimura A. The Jak-Stat pathway in normal and perturbed hematopoiesis. Blood. 2000;95(1):19–29.  
+
+<|ref|>text<|/ref|><|det|>[[46, 821, 877, 864]]<|/det|>
+39. Khwaja A. The role of Janus kinases in haemopoiesis and haematological malignancy. Br J Haematol. 2006;134(4):366–384.  
+
+<|ref|>text<|/ref|><|det|>[[46, 870, 912, 938]]<|/det|>
+40. Silvennoinen O, Witthuhn BA, Quelle FW, Cleveland JL, Yi T, Ihle JN. Structure of the murine Jak2 protein-tyrosine kinase and its role in interleukin 3 signal transduction. Proc Natl Acad Sci USA. 1993;90(18):8429–8433.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[47, 44, 940, 935]]<|/det|>
+41. Withthun BA, Quelle FW, Silvennoinen O, Yi T, Tang B, Miura O, et al. JAK2 associates with the erythropoietin receptor and is tyrosine phosphorylated and activated following stimulation with erythropoietin. Cell. 1993;74(2):227–236.  
+42. Parganas E, Wang D, Stravoopodis D, Topham DJ, Marine JC, Teglund S, et al. Jak2 is essential for signaling through a variety of cytokine receptors. Cell. 1998;93(3):385–395.  
+43. Rane SG, Reddy EP. JAKs, STATs and Src kinases in hematopoiesis. Oncogene, 2002;21(21):3334–3358.  
+44. Silvennoinen O, Hubbard SR. Molecular insights into regulation of JAK2 in myeloproliferative neoplasms. Blood. 2015;125(22):3388–3392.  
+45. Shan Y, Gnanasambandan K, Ungureanu D, Kim ET, Hammarén H, Yamashita K, et al. Molecular basis for pseudokinase-dependent autoinhibition of JAK2 tyrosine kinase. Nat Struct Mol Biol. 2014;21(7):579–584.  
+46. Bandaranayake RM, Ungureanu D, Shan Y, Shaw DE, Silvennoinen O, Hubbard SR. Crystal structures of the JAK2 pseudokinase domain and the pathogenic mutant V617F. Nat Struct Mol Biol. 2012;19(8):754–759.  
+47. Hammarén HM, Ungureanu D, Grisouard J, Skoda RC, Hubbard SR, Silvennoinen O. ATP binding to the pseudokinase domain of JAK2 is critical for pathogenic activation. Proc Natl Acad Sci USA. 2015;112(15):4642–4647.  
+48. James C, Ugo V, Couédic JL, Steak J, Delhommeay F, Lacout C, et al. A unique clonal JAK2 mutation leading to constitutive signalling causes polycythaemia vera. Nature. 2005;434(7037):1144–1148.  
+49. Levine RL, Wadleigh M, Cools J, Ebert BL, Wernig G, Huntly B, et al. Activating mutation in the tyrosine kinase JAK2 in polycythemia vera, essential thrombocythemia, and myeloid metaplasia with myelofibrosis. Cancer Cell. 2005;7(4):387–397.  
+50. Aynardi, Manur R, Hess PR, Chekol S, Morrissette JD, Babushol D, et al. JAK2 V617F-positive acute myeloid leukaemia (AML): a comparison between de novo AML and secondary AML transformed from an underlying myeloproliferative neoplasm. A study from the Bone Marrow Pathology Group. Br J Haematol. 2018;182(1):78–85.  
+51. Papaemmanuil E, Gerstung M, Malcovati L, Tauro S, Gundem G, Loo PV, et al. Clinical and biological implications of driver mutations in myelodysplastic syndromes. Blood. 2013;122(22):3616–3627.  
+52. Koppikar P, Bhagwat N, Kilpivaara O, Manshouri T, Adli M, Hricik T, et al. Heterodimeric JAK-STAT activation as a mechanism of persistence to JAK2 inhibitor therapy. Nature. 2012;489(7414):155–159.  
+53. Meyer S, Keller MD, Chiu S, Koppikar P, Guryanova OA, Rapaport F, et al. CHZ868, a Type II JAK2 Inhibitor, Reverses Type I JAK Inhibitor Persistence and Demonstrates Efficacy in Myeloproliferative Neoplasms. Cancer Cell. 2015;28(1):15–28.  
+54. Troffer-Charlier N, Cura V, Hassenboehler P, Moras D, Cavarelli J. Functional insights from structures of coactivator-associated arginine methyltransferase 1 domains. EMBO J. 2007;26(20):4391–4401.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[47, 42, 920, 87]]<|/det|>
+55. Blank U, Karlsson S. The role of Smad signaling in hematopoiesis and translational hematology. Leukemia. 2011;25(9):1379-1388.  
+
+<|ref|>text<|/ref|><|det|>[[47, 92, 936, 137]]<|/det|>
+56. Walz C, Crowley B, Hudon HE, Gramlich JL, Neuberg DS, Podar K, et al. Activated Jak2 with the V617F point mutation promotes G1/S phase transition. J Biol Chem. 2006;281(26):18177-18183.  
+
+<|ref|>text<|/ref|><|det|>[[47, 142, 940, 210]]<|/det|>
+57. Score J, Curtis C, Waghorn K, Stalder M, Jotterand M, Grand FH, et al. Identification of a novel imatinib responsive KIF5B-PDGFRA fusion gene following screening for PDGFRA overexpression in patients with hypereosinophilia. Leukemia. 2006;20(5):827-832.  
+
+<|ref|>text<|/ref|><|det|>[[48, 214, 930, 282]]<|/det|>
+58. Parker JE, Mufti GJ, Rasool F, Mijovic A, Devereux S, Pagliuca A, et al. The role of apoptosis, proliferation, and the Bcl-2-related proteins in the myelodysplastic syndromes and acute myeloid leukemia secondary to MDS. Blood. 2000;96(12):3932-3938.  
+
+<|ref|>text<|/ref|><|det|>[[48, 286, 949, 332]]<|/det|>
+59. Jenal M, Batliner J, Reddy VA, Haferlach T, Tobler A, Fey MF, et al. The anti-apoptotic gene BCL2A1 is a novel transcriptional target of PU.1. Leukemia. 2010;24(5):1073-1076.  
+
+<|ref|>text<|/ref|><|det|>[[48, 336, 894, 382]]<|/det|>
+60. Luo X, Xu L, Wu X, Tan H, Liu L. Decreased SATB1 expression promotes AML cell proliferation through NF-κB activation. Cancer Cell Int. 2019;19:134.  
+
+<|ref|>text<|/ref|><|det|>[[48, 386, 924, 454]]<|/det|>
+61. Keita M, Bachvarova M, Morin C, Plante M, Gregoire J, Renaud MC, et al. The RUNX1 transcription factor is expressed in serous epithelial ovarian carcinoma and contributes to cell proliferation, migration and invasion. Cell Cycle. 2013;12(6):972-986.  
+
+<|ref|>text<|/ref|><|det|>[[48, 457, 919, 525]]<|/det|>
+62. Kohrs N, Kolodziej S, Kuvardina ON, Herglotz J, Yillah J, Herkt S, et al. MiR144/451 Expression Is Repressed by RUNX1 During Megakaryopoiesis and Disturbed by RUNX1/ETO. PLoS Genet. 2016;12(3):e1005946.  
+
+<|ref|>text<|/ref|><|det|>[[48, 528, 945, 575]]<|/det|>
+63. Vainchenker W, Constantinescu SN. JAK/STAT signaling in hematological malignancies. Oncogene. 2013;32(21):2601-2613.  
+
+<|ref|>text<|/ref|><|det|>[[48, 579, 943, 647]]<|/det|>
+64. Huber FM, Greenblatt SM, Davenport AM, Martinez C, Xu Y, Vu LP, et al. Histone-binding of DPF2 mediates its repressive role in myeloid differentiation. Proc Natl Acad Sci USA. 2017;114(23):6016-6021.  
+
+<|ref|>text<|/ref|><|det|>[[48, 650, 947, 719]]<|/det|>
+65. Shepherd MS, Li J, Wilson NK, Oedekoven CA, Li J, Belmonte M, et al. Single-cell approaches identify the molecular network driving malignant hematopoietic stem cell self-renewal. Blood. 2018;132(8):791-803.  
+
+<|ref|>text<|/ref|><|det|>[[48, 722, 947, 793]]<|/det|>
+66. Tognon R, Gasparotto EP, Neves RP, Nunes NS, Ferreira A, Palma PV, et al. Deregulation of apoptosis-related genes is associated with PRV1 overexpression and JAK2 V617F allele burden in Essential Thrombocythemia and Myelofibrosis. J Hematol Oncol. 2012;5:2.  
+
+<|ref|>text<|/ref|><|det|>[[48, 796, 927, 865]]<|/det|>
+67. Waibel M, Solomon VS, Knight DA, Ralli RA, Kim SK, Banks KM, et al. Combined targeting of JAK2 and Bcl-2/Bcl-xL to cure mutant JAK2-driven malignancies and overcome acquired resistance to JAK2 inhibitors. Cell Rep. 2013;5(4):1047-1059.  
+
+<|ref|>text<|/ref|><|det|>[[48, 868, 912, 914]]<|/det|>
+68. Meyer SC, Levine RL. Molecular pathways: molecular basis for sensitivity and resistance to JAK kinase inhibitors. Clin Cancer Res. 2014;20(8):2051-2059.  
+
+<|ref|>text<|/ref|><|det|>[[48, 917, 936, 963]]<|/det|>
+69. Karakashev S, Zhu H, Wu S, Yokoyama Y, Bitler B, Park PH, et al. CARM1-expressing ovarian cancer depends on the histone methyltransferase EZH2 activity. Nat Commun. 2018;9(1):631.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[50, 45, 916, 111]]<|/det|>
+70. MaFarland J, Ho Z, Kugener G, Dempster J, Montgomery P, Bryan J, et al. Improved estimation of cancer dependencies from large-scale RNAi screens using model-based normalization and data integration. Nat Commun. 2018;9(1):4610. 
+
+<|ref|>sub_title<|/ref|><|det|>[[42, 136, 144, 161]]<|/det|>
+## Figures 
+
+<|ref|>image_caption<|/ref|><|det|>[[52, 188, 118, 203]]<|/det|>
+<center>Figure 1</center> 
+
+<|ref|>image<|/ref|><|det|>[[52, 214, 600, 415]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[52, 435, 950, 900]]<|/det|>
+ 
+
+<|ref|>image_caption<|/ref|><|det|>[[42, 932, 118, 950]]<|/det|>
+<center>Figure 1</center>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 44, 367, 65]]<|/det|>
+## JAK2 phosphorylates CARM1 in vitro  
+
+<|ref|>text<|/ref|><|det|>[[42, 82, 950, 172]]<|/det|>
+(A) JAK2 phosphorylates CARM1 in an in vitro kinase assay, in which active JAK2 kinase and recombinant CARM1 proteins were used; PAK1 was used as a positive control. The amounts of protein in the reaction are indicated. The phosphorylation of CARM1 was completely abolished by the JAK2 inhibitor (ruxolitinib, RUX) (lanes 7 and 8).  
+
+<|ref|>text<|/ref|><|det|>[[42, 189, 955, 279]]<|/det|>
+(B) Peptide fragments in the mass spectrometry analysis were generated from proteolytic cleavage of CARM1 following in vitro kinase assays in the presence of active JAK2 kinase. Tyrosine-149 (Y149) along with the series of y- and b-ions, including the phosphorylated residue, is shown as the phosphorylated peptide (EESSAVQpYF).  
+
+<|ref|>text<|/ref|><|det|>[[42, 294, 930, 385]]<|/det|>
+(C) Peptide fragments in the mass spectrometry analysis were generated from proteolytic cleavage of CARM1 following in vitro kinase assays in the absence of active JAK2 kinase. The peptide fragment around Y149 residue (EESSAVQYF) is shown without phosphorylation of tyrosine, indicating no gain in molecular weight of 80 Da (i.e. the weight of PO4).  
+
+<|ref|>text<|/ref|><|det|>[[42, 402, 954, 492]]<|/det|>
+(D) Peptide fragments in the mass spectrometry analysis were generated from proteolytic cleavage of CARM1 following in vitro kinase assays in the presence of active JAK2 kinase. Tyrosine-334 (Y334) along with the series of y- and b-ions, including the phosphorylated residue, is shown as the phosphorylated peptide (GAAVDEpYFR).  
+
+<|ref|>text<|/ref|><|det|>[[42, 508, 933, 598]]<|/det|>
+(E) Peptide fragments in the mass spectrometry analysis were generated from proteolytic cleavage of CARM1 following in vitro kinase assays in the absence of active JAK2 kinase. The peptide fragment around Y334 residue (GAAVDEYFR) is shown without phosphorylation of tyrosine, indicating no gain in molecular weight of 80 Da.  
+
+<|ref|>text<|/ref|><|det|>[[42, 614, 933, 705]]<|/det|>
+(F) The regions containing amino acid residues 149 and Y334 are located within the core catalytic domain (residue 140-480) of CARM1. Residues 28-140 in CARM1 are highly homologous to a family of Drosophila-Enabled/vasodilator-stimulated phosphoprotein homology 1 (EVH1) domains, which specifically bind to target proline-rich sequences with low affinity and high specificity.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[50, 60, 808, 750]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[48, 46, 94, 58]]<|/det|>
+<center>Figure 2 </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 803, 808, 825]]<|/det|>
+## JAK2-V617F promotes the tyrosine phosphorylation of CARM1 in myeloid leukemia cells  
+
+<|ref|>text<|/ref|><|det|>[[42, 842, 933, 944]]<|/det|>
+(A) The expression of CARM1 protein and Y149/Y334 phosphorylated CARM1 protein was assessed in 14 myeloid leukemia cell lines and human CD34+ cord blood cells, by immunoblotting analysis. 
+(B) Phosphorylation of Y149 and Y334 of CARM1 in HEL cells is abolished following treatment with the JAK2 inhibitor (RUX at low concentration to avoid severe apoptosis; 500 nM), as is phosphorylation of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[44, 46, 240, 64]]<|/det|>
+tyrosine- 694 in STAT5.  
+
+<|ref|>text<|/ref|><|det|>[[42, 83, 955, 149]]<|/det|>
+(C) Immunoprecipitation was performed using HEL, SET2, and K562 cells that express HA-tagged CARM1, with an anti-HA antibody. Immunoblotting with anti-HA and anti-JAK2 antibodies revealed the interaction between JAK2 and CARM1.  
+
+<|ref|>text<|/ref|><|det|>[[42, 166, 923, 233]]<|/det|>
+(D) HEL and UKE-1 cells harboring homozygous JAK2-V617F mutations had phosphorylated CARM1-Y149/Y334 and (auto) phosphorylated JAK2-Y1007/Y1008, while SET2 cells harboring heterozygous JAK2-V617F mutation did not.  
+
+<|ref|>text<|/ref|><|det|>[[42, 250, 926, 315]]<|/det|>
+(E) Proteins were immunoprecipitated from HEL cell extracts that express HA-tagged CARM1, using an anti-HA antibody; immunoblotting was then performed using an anti-HA antibody, anti-phosphorylated JAK2 rabbit antibody, or anti-JAK2 mouse antibody.  
+
+<|ref|>text<|/ref|><|det|>[[42, 332, 916, 375]]<|/det|>
+(F) Extracts from JAK1 or TYK2 knockout HEL cells were immunoblotted using phosphospecific anti-CARM1, JAK2, and STAT5 antibodies.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[45, 53, 800, 789]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 802, 117, 820]]<|/det|>
+<center>Figure 3 </center>  
+
+<|ref|>text<|/ref|><|det|>[[44, 843, 750, 864]]<|/det|>
+Biochemical regulation of CARM1 enzymatic activity by tyrosine phosphorylation.  
+
+<|ref|>text<|/ref|><|det|>[[42, 881, 936, 924]]<|/det|>
+(A) Based on the crystal structures of CARM1, CARM1-Y334 phosphorylation (middle) increased the interaction of the region containing Y334 itself with substrate compared with non-phosphorylated Y334
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 911, 88]]<|/det|>
+(left). CARM1- Y149 phosphorylation (right) also increased the binding of the region containing Y149 itself with substrate.  
+
+<|ref|>text<|/ref|><|det|>[[41, 105, 951, 240]]<|/det|>
+(B) CARM1- Y334 and -Y149 phosphorylation impairs the loss of binding between CARM1 methionine-259 (Met259) and substrate. The decreased Met259 binding increases the interaction of glutamic acid-257 (Glu257) and glutamic acid-266 (Glu266) with substrates (middle) in the presence of Y334 phosphorylation, compared to non-phosphorylated Y334 (left). Furthermore, the loss of methionine-259 binding increases the interaction of glutamine-158 (Gln158) and aspartic acid-161 (Asn161) with substrates in the presence of Y149 phosphorylation (right).  
+
+<|ref|>text<|/ref|><|det|>[[41, 256, 944, 370]]<|/det|>
+(C) The mutant CARM1 Y149F, Y334F, and Y149F/Y334F proteins show reduced methyltransferase activity for histone H3.1, compared to wild-type (WT) CARM1, in in vitro methylation assay. CARM1 and MYC protein lanes are shown to demonstrate equal loading (Top). Autoradiograph of the methylated \(^3\mathrm{H}\) -histone H3.1 (Middle). Coomassie staining shows histone H3.1 used in the assay (Bottom). Western blotting shows the relative amount of CARM1 and MYC.  
+
+<|ref|>text<|/ref|><|det|>[[41, 387, 936, 478]]<|/det|>
+(D) Phosphorylated CARM1 shows increased methyltransferase activity for histone H3.1 in in vitro methylation assay (Top). Autoradiograph of the methylated \(^3\mathrm{H}\) -histone H3.1 (Middle). Coomassie staining shows histone H3.1 used in the assay (Bottom). Western blotting shows the relative amount of CARM1 and phosphorylated CARM1.  
+
+<|ref|>text<|/ref|><|det|>[[42, 495, 920, 538]]<|/det|>
+(E) CARM1-Y149 and -Y334 localize at dimerization arm and helix \(\alpha X,\) respectively. These residues lie close to the dimerization interface in the modeled CARM1 structure.  
+
+<|ref|>text<|/ref|><|det|>[[41, 555, 933, 688]]<|/det|>
+(F) Co-immunoprecipitation of HA- and MYC-tagged CARM1 from 293T cell extracts transiently transfected with plasmid expressing HA-tagged WT and MYC-tagged WT or mutant CARM1. HA-tagged WT CARM1 was immunoprecipitated from cell extracts with anti-HA antibodies, and then the coimmunoprecipitated MYC-tagged CARM1 was probed with anti-MYC antibodies. The levels of MYC-tagged CARM1 Y149F and Y149F/Y334F from the HA immunoprecipitates were lower than those of MYC-tagged CARM1 WT.  
+
+<|ref|>text<|/ref|><|det|>[[40, 706, 954, 864]]<|/det|>
+(G) Subcellular fractionations of HEL cells and UKE-1 cells were immunoblotted using anti-total CARM1, CARM1-pY334, and -pY149 antibodies; cytoplasmic extraction, CYE; nuclear soluble extraction, NSE; and chromatin-bound extraction, CBE. The left lane represents the expression levels of the indicated proteins of whole-cell lysates (WCE). The bar graph on the right represents the ratio of cytoplasmic, nuclear, or chromatin-binding CARM1-pY334 and -pY149 to total cytoplasmic, nuclear, or chromatin-binding CARM1, respectively (bands inside the boxes). Data represent the mean ± SD. n=3, **p<0.01, *** p<0.001; unpaired two-tailed Student's t-test.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[50, 55, 680, 789]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 46, 80, 55]]<|/det|>
+<center>Figure 4 </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 838, 783, 860]]<|/det|>
+## Identification of RUNX1 as CARM1-interacting proteins by Proximity BiolD proteomics  
+
+<|ref|>text<|/ref|><|det|>[[42, 878, 944, 921]]<|/det|>
+(A) Schematic diagram demonstrating BiolD (proximity-dependent biotin identification) approach for the identification of CARM1-interacting proteins.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 946, 111]]<|/det|>
+(B) Metascape enrichment network visualization showing the intra-cluster and inter-cluster similarities of enrichment terms, up to nine terms per clusters, in HEL cells. Terms are defined according to GO/KEGG terms, canonical pathways, and hallmark gene sets.  
+
+<|ref|>text<|/ref|><|det|>[[42, 128, 949, 194]]<|/det|>
+The connecting pairs of nodes are created with Kappa score \(>0.3\) . Terms containing more genes tend to have a more significant \(P\) - value; the darker color of the node indicates the more statistically significant \(P\) - value.  
+
+<|ref|>text<|/ref|><|det|>[[42, 211, 944, 348]]<|/det|>
+(C) Scatter plot comparing mean-fold change for CARM1-BirA\* fusion vs. BirA\* alone with abundance in published negative control AP-MS datasets (%CRAPome). Green dots represent proteins (i) with a cutoff frequency of \(\geq 80\%\) CRAPome and the average spectral count fold change \(\geq 1.2\) or (ii) with a cutoff frequency of \(< 80\%\) CRAPome but the average spectral count fold change \(\geq 3.0\). Known substrates of CARM1 are indicated as red, and E2F-targets, histone binding proteins, and MYC-targets are shown in yellow, blue, and violet, respectively. See also supplemental data 1 (HEL cells) and 2 (K562 cells).  
+
+<|ref|>text<|/ref|><|det|>[[42, 363, 949, 430]]<|/det|>
+(D) Proteins were immunoprecipitated from HEL cell extracts that express MYC-tagged CARM1 (WT and non-phosphorylatable mutants), using an anti-MYC antibody; immunoblotting was then performed using an anti-MYC antibody and anti-RUNX1 mouse antibody.  
+
+<|ref|>text<|/ref|><|det|>[[42, 446, 951, 512]]<|/det|>
+(E) Doxycycline-inducible short hairpin RNAs (shRNAs) directed against CARM1 decreased CARM1 protein levels and the ADMA levels of RUNX1-R223 and -R319 as well as well-established targets, such as PABP1-R455/R460 and BAF155-R1064.  
+
+<|ref|>text<|/ref|><|det|>[[42, 529, 950, 594]]<|/det|>
+(F) Clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR-associated protein-9 (Cas9)-mediated non-phosphorylatable CARM1 mutants decreased the ADMA levels of RUNX1-R223 and -R319 as well as PABP1 and BAF155.  
+
+<|ref|>text<|/ref|><|det|>[[42, 611, 953, 722]]<|/det|>
+(G) Expression of total and asymmetry dimethylated RUNX1, PABP1, and BAF155 were assessed in HEL cells treated with RUX 250 nM or DMSO control for 5 days. Fresh media with RUX or DMSO was added on day 0, 2, and 4. Quantification of the ADMA levels of RUNX1, BAF155, and PABP1 at 5 days after RUX treatment are shown in the right panels. Data represent the mean ± SD. \(n = 3\) , \(*p< 0.05\) , \(**p< 0.01\) ; one-way ANOVA.  
+
+<|ref|>text<|/ref|><|det|>[[42, 740, 955, 808]]<|/det|>
+(H) The levels of ADMA RUNX1, BAF155, and PABP1 were measured in HEL cells treated with RUX (or DMSO) after EPZ025654 treatment for 5 days followed by a wash-out phase lasting up to 3 days (labeled as day 8).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[45, 50, 617, 790]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 800, 118, 818]]<|/det|>
+<center>Figure 5 </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 841, 568, 861]]<|/det|>
+## Functional analysis of non-phosphorylatable mutant CARM1  
+
+<|ref|>text<|/ref|><|det|>[[44, 880, 896, 923]]<|/det|>
+(A) Cell proliferation assays of non-phosphorylatable CARM1 mutant knock-in HEL cells, where cell numbers were measured using cell-counting apparatus. \(\mathrm{n} = 3\) , \(^{**}p< 0.01\) .
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 44, 904, 88]]<|/det|>
+(B) The flow cytometry analysis of BrdU-stained HEL cells expressing non-phosphorylatable CARM1 mutants. Mean fractions ± s.d. in sub G1, G0/G1, S, and G2/M populations. n=3, \*p<0.05, \*\*p<0.01.  
+
+<|ref|>text<|/ref|><|det|>[[42, 105, 943, 172]]<|/det|>
+(C) Heatmap shows the differentially expressed coding genes at 2-fold cut-off, representing replicates of HEL cells expressing CARM1 WT or two independent cells expressing CAMR1-Y149F/Y334F double mutation (Y149F/Y334F-1 and Y149F/Y334F-2).  
+
+<|ref|>text<|/ref|><|det|>[[42, 188, 872, 231]]<|/det|>
+(D) Gene ontology analysis of significant downregulated genes in HEL cells expressing CARM1-Y149F/Y334F compared to CARM1-WT.  
+
+<|ref|>text<|/ref|><|det|>[[42, 248, 763, 270]]<|/det|>
+(E) Heatmaps of FDR (q < 0.25) values from GSEA of hallmark gene set collections.  
+
+<|ref|>text<|/ref|><|det|>[[42, 286, 909, 331]]<|/det|>
+(F) Representative GSEA plot depicting the downregulation of G2/M checkpoint and apoptosis/antiapoptosis pathways.  
+
+<|ref|>text<|/ref|><|det|>[[42, 347, 949, 459]]<|/det|>
+(G) Volcano plot representing gene expression changes triggered by CARM1-Y149F/Y334F mutation knock-in in HEL cells. Genes associated with apoptosis/anti-apoptosis, G2/M checkpoints, stemness in hematopoietic stem cells, and RUNX1-target are shown in red, yellow, bale, and violet, respectively. The red dots indicate upregulated genes in HEL cells expressing CARM1-Y149F/Y334F, whereas the blue dots indicated downregulated genes.  
+
+<|ref|>text<|/ref|><|det|>[[42, 476, 920, 498]]<|/det|>
+(H) Representative GSEA plot depicting the downregulation of "hematopoietic stem cell up" signature.  
+
+<|ref|>text<|/ref|><|det|>[[42, 514, 920, 558]]<|/det|>
+(I) qRT-PCR analysis showing BMI-1 and CD34 in HEL cells expressing CARM1 WT, Y149F, Y334F, and Y149F/Y334F mutation. Mean and SD are expressed as a percentage of HPRT-1 expression.  
+
+<|ref|>text<|/ref|><|det|>[[42, 574, 923, 618]]<|/det|>
+(J) qRT-PCR analysis showing ID2 and MIR144 in HEL cells expressing CARM1 WT, Y149F, Y334F, and Y149F/Y334F mutation. Mean and SD are expressed as a percentage of HPRT-1 expression.  
+
+<|ref|>text<|/ref|><|det|>[[42, 634, 945, 679]]<|/det|>
+(K) Heat map of total R319-RUNX1 or asymmetrically dimethylated R319-RUNX1 binding tag intensity by ChIP-seq analysis for HEL cells expressing CARM1 WT, Y149F, Y334F, or Y149F/Y334F mutant proteins.  
+
+<|ref|>text<|/ref|><|det|>[[42, 695, 945, 740]]<|/det|>
+(L) ChIP-seq analyses were performed to assess total RUNX1 and asymmetrically dimethylated R319-RUNX1 chromatin binding. Target occupancies at the ID2 gene are shown in IGV genome browser tracks.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[56, 45, 641, 310]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[46, 320, 62, 330]]<|/det|>
+<center>B </center>  
+
+<|ref|>image<|/ref|><|det|>[[55, 335, 641, 508]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[46, 520, 62, 530]]<|/det|>
+<center>C </center>  
+
+<|ref|>image<|/ref|><|det|>[[55, 520, 641, 784]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 803, 116, 821]]<|/det|>
+<center>Figure 6 </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 843, 920, 863]]<|/det|>
+## Inhibition of CARM1 targets cells harboring phosphorylated CARM1 mediated by JAK2-V617F mutant  
+
+<|ref|>text<|/ref|><|det|>[[43, 881, 920, 946]]<|/det|>
+(A) Western blot assessment of phosphorylation in JAK2, STAT5, ERK, and AKT, and asymmetric demethylated arginine in RUNX1, BAF155, and PABP1 in HEL and UKE-1 cells treated with 5 days with increasing concentrations of EPZ025654 (μM).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 955, 134]]<|/det|>
+(B) Excess over Bliss plots (Bliss method) showing synergistic effects between EPZ025654 and RUX were visualized in the calculated 2D synergy maps. Red and green areas represent synergistic (synergy score \(> + 10\) ), addictive (synergy score \(0 + 10\) ), and antagonistic effect ( \(< - 10\) ), respectively. In 2D synergy maps, white rectangles show the maximum synergy area in each cell.  
+
+<|ref|>text<|/ref|><|det|>[[42, 151, 955, 263]]<|/det|>
+(C) The colony formation of HEL, UKE-1, and SET2 cells treated with DMSO (control), RUX, EPZ025654, or a combination of RUX and EPZ025654. The concentration of RUX was applied based on the IC50 values for each cell line. Representative pictures of colonies on semi-solid methylcellulose media are shown on the upper panels. Quantification of the number of colonies at 14 days after plating are shown in the lower panels. Data represent the mean ± SD. \(n = 4\) , \(p< 0.05\) , \(**p< 0.01\) , \(***p< 0.001\) ; one-way ANOVA.  
+
+<|ref|>image<|/ref|><|det|>[[45, 280, 930, 688]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 714, 117, 733]]<|/det|>
+<center>Figure 7 </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 756, 454, 776]]<|/det|>
+## A schematic model showing JAK2-CARM1 axis  
+
+<|ref|>text<|/ref|><|det|>[[42, 794, 950, 882]]<|/det|>
+JAK2- V617F mutant kinase, when activated by JAK2, JAK1, or TYK2, strongly phosphorylates CARM1- Y149 and - Y334, increasing its methyltransferase activity and the asymmetrical dimethylation of its substrates, including histone 3 and RUNX1. CARM1 phosphorylation promotes cell- cycle progression and inhibits apoptosis, and regulates the genes associated with stemness (BMI- 1).  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 905, 311, 931]]<|/det|>
+## Supplementary Files
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[44, 45, 765, 64]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 82, 485, 128]]<|/det|>
+- NatCommunSupplementalData.xlsx- NatCommunSupplementalFigXTextXTab.docx
+
+<--- Page Split --->

preprint/preprint__7ea7ef9b459a9c05ffe0bbce2c11c10d905560d4f8d0e1620fe90f75c4f3407c/images_list.json

@@ -0,0 +1,70 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1: Mouse EndoU is expressed during thymocyte development and has a calcium-activated RNase activity. (A) EndoU expression levels in developing thymocyte populations. The color indicates fractional expression of the mRNA across 211 measured cell types. Data from the Immunological Genome Project [63]. (B) RNase activity in WT and EndoU KO cell lysates. Cytoplasmic lysates from the indicated WT, KO or rescue cell lines were incubated +/- 5 mM \\(\\mathrm{Ca^{2 + }}\\) for 15 minutes at \\(37^{\\circ}\\mathrm{C}\\) . RNA was extracted, 5 μg were run on an 8 % urea-PAGE gel, and visualized by SYBR Green II. WT-HA denotes a WT EndoU rescue construct with a C-terminal HA tag. (C) Immunoprecipitated EndoU cleavage activity on a defined RNA substrate. (D) Mutation of presumed catalytic site residues abolishes EndoU cleavage activity. (E) EndoU RNase activity is specifically activated by \\(\\mathrm{Ca^{2 + }}\\) ions. (F) Domain structure of EndoU and homologs.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 4
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2: Crystal structure of calcium-activated EndoU. (A) Overview of calcium-activated EndoU structure. Calcium binding site (1) to (4) as refered in the text are indicated. (B) Close-up view of EndoU calcium binding sites. Residues coordinating calcium directly (pink) or indirectly through water-mediated contacts (cyan) are highlighted. (C) Intramolecular signaling between calcium binding site (1) and remote catalytic residues. Catalytic residue (yellow) are highlighted, along with residues coordinating calcium directly (pink) or indirectly (cyan). (D) Structural change upon EndoU allosteric activation by calcium.",
+    "footnote": [],
+    "bbox": [
+      [
+        118,
+        87,
+        884,
+        830
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3: Enzymatic activity of EndoU and its variants. (A) RNA degradation assays. Comparison of mutants for calcium-binding sites (magenta), the bridging residue E290 (cyan), and catalytic residues (yellow) with wild-type EndoU over a 2 hrs degradation assay. (B) Enzymatic progress curve. Example of fit for a first-order reaction model \\(A \\times \\exp (-k \\times t)\\) with wild-type EndoU. (C) Relative reaction rates of EndoU variants compared to wild-type. (D) Calcium binding to EndoU monitored through Molecular Dynamics. Each plot displays the distance between the E284 side-chain carboxylate and a calcium ion throughout the simulation. Five calcium ions were introduced in the simulation box.",
+    "footnote": [],
+    "bbox": [
+      [
+        113,
+        213,
+        883,
+        626
+      ]
+    ],
+    "page_idx": 37
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4: NMR mapping of the RNA binding interface on calcium-activated EndoU. Overlay of (A) \\(^{15}\\mathrm{N}\\) or (B) \\(^{13}\\mathrm{C}\\) SOFAST-HMQC spectra from isolated \\(^{13}\\mathrm{C}^{15}\\mathrm{N}\\) -EndoU (350 \\(\\mu \\mathrm{M}\\) , green) and upon successive additions of 2' fluorinated RNA. (C) Combined \\(^{1}\\mathrm{H}\\) - \\(^{15}\\mathrm{N}\\) chemical shift perturbations between calcium-activated \\(^{13}\\mathrm{C}^{15}\\mathrm{N}\\) -EndoU and in complex with 2'F RNA. (D) RNA binding interface mapping on calcium-activated EndoU. (E) Electrostatic surface potential of calcium-activated EndoU.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        88,
+        875,
+        850
+      ]
+    ],
+    "page_idx": 38
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5: Schematic representation of eukaryotic EndoU activation upon calcium and substrate binding.",
+    "footnote": [],
+    "bbox": [
+      [
+        115,
+        265,
+        888,
+        675
+      ]
+    ],
+    "page_idx": 39
+  }
+]

preprint/preprint__7ea7ef9b459a9c05ffe0bbce2c11c10d905560d4f8d0e1620fe90f75c4f3407c/preprint__7ea7ef9b459a9c05ffe0bbce2c11c10d905560d4f8d0e1620fe90f75c4f3407c.mmd

@@ -0,0 +1,491 @@
+
+# Molecular Basis for the Calcium-Dependent Activation of the Ribonuclease EndoU  
+
+Florian Malard  INSERM U1212  
+
+Kristen Dias  University of California Riverside  
+
+Margaux Baudy  INSERM U1212, ARNA unit  
+
+Stéphane Thore  INSERM U1212, ARNA unit  
+
+Brune Vialet  University of Bordeaux  
+
+Philippe Barthélémy  https://orcid.org/0000- 0003- 3917- 0579  
+
+Sébastien Fribourg  Univ. de Bordeaux, Institut Européen de Chimie et Biologie  
+
+Fedor Karginov  University of California Riverside  
+
+Sebastien Campagne  
+
+sebastien.campagne@inserm.fr  
+
+INSERM U1212, ARNA unit  https://orcid.org/0000- 0002- 0094- 4760  
+
+## Article  
+
+Keywords: EndoU, calcium, RNA, allostery, ribonuclease.  
+
+Posted Date: July 15th, 2024  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 4654759/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.
+
+<--- Page Split --->
+
+Version of Record: A version of this preprint was published at Nature Communications on April 1st, 2025. See the published version at https://doi.org/10.1038/s41467-025-58462-6.
+
+<--- Page Split --->
+
+# Molecular Basis for the Calcium-Dependent Activation  
+
+# of the Ribonuclease EndoU  
+
+Florian Malard \(^{1,2}\) , Kristen Dias \(^{3}\) , Margaux Baudy \(^{1,2}\) , Stéphane Thore \(^{1}\) , Brune Vialet \(^{1}\) , Philippe Barthélémy \(^{1}\) , Sébastien Fribourg \(^{1,*}\) , Fedor V Karginov \(^{3,*}\) , and Sébastien Campagne \(^{1,2,*}\)  
+
+\(^{1}\) Univ. Bordeaux, CNRS, INSERM, ARNA, UMR 5320, U1212, F- 33000  
+
+Bordeaux, France  
+
+\(^{2}\) Univ. Bordeaux, CNRS, INSERM, IECB, US1, UAR 3033, F- 33600 Pessac,  
+
+France  
+
+\(^{3}\) Department of Molecular, Cell and Systems Biology, Institute for Integrative Genome Biology, University of California at Riverside, Riverside, CA, 92521,  
+
+USA  
+
+\* Correspondence should be addressed to sebastien.fribourg@inserm.fr, fedor.karginov@ucr.edu and sebastien.campagne@inserm.fr
+
+<--- Page Split --->
+
+## Abstract  
+
+Ribonucleases (RNases) are ubiquitous enzymes that process or degrade RNA, essential for cellular functions and immune responses. The EndoU- like superfamily includes endoribonucleases conserved across bacteria, eukaryotes, and certain viruses, with an ancient evolutionary link to the ribonuclease A- like superfamily. Both bacterial EndoU and animal RNase A share a similar fold and function independently of cofactors. In contrast, the eukaryotic EndoU catalytic domain requires divalent metal ions for catalysis, possibly due to an N- terminal extension near the catalytic core. In this study, we used biophysical and computational techniques along with in vitro assays to investigate the calcium- dependent activation of human EndoU. We determined the crystal structure of EndoU bound to calcium and found that calcium binding remote from the catalytic triad triggers water- mediated intramolecular signaling and structural changes, activating the enzyme through allostery. Calcium- binding involves residues from both the catalytic core and the N- terminal extension, indicating that the N- terminal extension interacts with the catalytic core to modulate activity in response to calcium. Our findings suggest that similar mechanisms may be present across all eukaryotic EndoUs, highlighting a unique evolutionary adaptation that connects endoribonuclease activity to cellular signaling in eukaryotes.  
+
+Keywords: EndoU, calcium, RNA, allostery, ribonuclease.
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+  
+
+Graphical abstract: Calcium binds at the interface between the catalytic core and N- terminal extension in eukaryotic EndoU catalytic domains, activating the catalytic site at a distance via allostery.
+
+<--- Page Split --->
+
+## Introduction  
+
+Ribonucleases (RNases) are nucleases that catalyze the processing or degradation of RNA. Found in all organisms, RNases play vital roles in various cellular processes, including maturing both coding and non- coding RNAs, combating RNA viruses, and contributing to sophisticated immune strategies like RNA interference [1, 2, 3]. For example, RNases catalyze mRNA decay in general pathways (XRN1, exosome/DIS3L) or as part of apoptotic cascades (RNase L, DIS3L2), carry out unconventional splicing or tRNA cleavage during stress (IRE1, angiogenin), or catabolize extracellular RNAs (RNase A). Among RNases, the cellular roles of those that cleave endonucleolytically have been increasingly recognized [4]. RNases can be constitutively active (RNase A, angiogenin), or stimulated by ligand binding (RNase L) or cellular signaling events, such as phosphorylation (IRE1). The ribonuclease A- like domain superfamily (IPR036816 [5]) is the most well- known RNase domain, with many pioneering studies in the \(20^{\mathrm{th}}\) century [6, 7]: it was the first directly sequenced enzyme [8], the first enzyme for which a catalytic mechanism was proposed based on experimental data [9], and one of the first solved three- dimensional structures [10]. Despite its significant impact in enzyme research, it is important to note that the RNase- A- like domain is only found in vertebrates, raising questions about its deeper evolutionary ancestors or relatives [6].  
+
+The endoribonuclease EndoU- like (Endoribonuclease specific for Uridylate) superfamily (IPR037227 [5]) is a poorly understood group of RNases found in bacteria, eukaryotes and viruses. Notably, a structural similarity between a bacterial EndoU- like toxin and vertebrate RNase A was identified [11]. Furthermore, recent studies uncovered an ancient evolutionary link between the Ribonuclease A and EndoU families, suggesting that the animal RNase A gene could have evolved either through significant alteration of an EndoU gene, or by horizontal acquisition of a prokaryotic ribonuclease [6]. XendoU, the founding member of the EndoU- like superfamily (IPR037227 [5]), was initially identified in Xenopus laevis oocyte extracts as an enzyme that releases small nucleolar RNAs from introns [12, 13, 14]. In vitro studies demonstrated that XendoU is an endonuclease that cleaves single- stranded RNA preferentially at 5' of uridylates [15]. In eukaryotes, XendoU defines
+
+<--- Page Split --->
+
+a distinct EndoU family (IPR018998 [5], PF09412 [16]) that lacks sequence homology with other known RNases, and is broadly conserved across \*Arabidopsis thaliana\*, \*Drosophila melanogaster\*, \*Mus musculus\*, \*Homo sapiens\*, and other species [14, 15]. Human EndoU (hEndoU) was first identified as human placental protein 11 (PP11) due to its prevalence in the placenta [17, 18]. It is also now recognized as a biomarker in various cancers, including squamous cell carcinomas, ovarian adenocarcinoma, non- trophoblastic tumors and breast cancers [19, 20, 21, 22, 23, 24]. In human cells, EndoU has been proposed to be involved in RNA cleavage, ribonucleoprotein particle removal, and endoplasmic reticulum network organization [25, 26]. Across other eukaryotes, EndoU has been implicated in pro- apoptotic processes in mouse B cells, neuron survival in fruit flies, and synaptic remodeling in nematodes [27, 28, 29]. The more distant bacterial EndoU- like ribonucleases are common in microbial warfare as toxins [30].  
+
+Members of the EndoU- like superfamily (IPR037227 [5]) exhibit notable differences in their activation requirements. For instance, it is well characterized that EndoU- like bacterial toxins and arterial Nsp11 do not need any cofactors for activation, analogous to vertebrate RNase A [11, 31]. In contrast, studies have shown that purified forms of XendoU and coronaviral Nsp15 require millimolar concentrations of \(\mathrm{Ca}^{2 + }\) or \(\mathrm{Mn}^{2 + }\) [15, 25, 32, 33]. The crystal structure of the endoribonuclease XendoU in the absence of divalent metals has been solved [34], suggesting a catalytic site arrangement similar to that of vertebrate RNase A, specifically featuring a catalytic His- His- Lys triad [34]. However, the structural basis for the metal- dependent activation of eukaryotic EndoUs could not be explained by the crystal structure of XendoU, which represents the inactive state of the endonuclease in the absence of a cofactor [34]. Bacterial and metal- independent viral EndoUs share a smaller, C- terminal catalytic domain compared to eukaryotic EndoUs. Because eukaryotic EndoUs contain an N- terminal extension within this catalytic domain that correlates with \(\mathrm{Ca}^{2 + }\) dependence, we hypothesized that it may bind calcium and control the activity of the catalytic core through allostery.  
+
+In this study, we elucidated the molecular mechanism of EndoU activation by calcium. First, we es
+
+<--- Page Split --->
+
+tablished a thymocyte cell line model to confirm the dependence of EndoU for calcium in both cell extract and recombinant forms. Next, we used biophysical methods to detect an allosteric change upon activation by calcium and to solve the structure of active EndoU. Our structural analysis revealed a calcium- stabilized interaction network involving residues from both the eukaryotic- specific N- terminal extension and the catalytic core of EndoU, ultimately leading to the activation of the catalytic triad. Our findings provide unprecedented atomic- level insights into a metal ion- activated member of the EndoU- like superfamily (IPR037227 [5]), addressing a longstanding question in the study of eukaryotic EndoUs, which are of significant interest due to their switchable endonuclease activity.
+
+<--- Page Split --->
+
+## Materials and Methods  
+
+## Cell culture  
+
+VL3- 3M2 mouse thymic lymphoma cells [35] were cultured in RPMI 1640 (Corning) supplemented with \(10\mathrm{mM}\) HEPES, \(50\mu \mathrm{L}\beta\) - mercaptoethanol, \(1\mathrm{x}\) penicillin/streptomycin, and \(10\%\) fetal bovine serum (FBS). The Platinum- E (Plat- E) retroviral packaging cell line was cultured in DMEM (Corning) supplemented with \(10\%\) FBS (Corning) and \(10\mathrm{units.ml^{- 1}}\) of penicillin/streptomycin (Gibco). All cells were grown at \(37^{\circ}\mathrm{C}\) in an atmosphere containing \(5\%\) \(\mathrm{CO_2}\)  
+
+## VL3-3M2 TCR activation  
+
+Cell culture 6- well plates were pre- incubated overnight at \(37^{\circ}\mathrm{C}\) with \(1\mathrm{mL}\) of PBS, either with or without \(5\mu \mathrm{g.mL^{- 1}}\) of anti- CD3e/CD28 or anti- CD3/CD4 antibodies. The PBS was then aspirated, and \(5*10^{5}\) cells in \(2\mathrm{mL}\) of media were added. For PMA/ionomycin stimulation, concentrations of \(20\mathrm{ng.mL^{- 1}}\) and \(500\mathrm{ng.mL^{- 1}}\) were used, respectively. Total RNA was extracted using Trizol 24 hours later, and RT- qPCR measurements were conducted for EndoU, Rag1, and CD5, normalized against a \(\beta\) - actin control. Fold changes were calculated relative to an unstimulated control.  
+
+## EndoU knockout cell generation  
+
+EndoU KO VL3- 3M2 cells were generated as described [36]. sgRNAs designed to target intron 1 and exon 11 of the EndoU locus (Table S1) were cloned into the pSpCas9(BB)/pX330 Cas9- sgRNA expression plasmid (Addgene #42230). A neomycin resistance cassette flanked by two 900 bp homology regions to intron 1 and exon 11 were assembled into the pUC- 19 vector as previously described [36]. The Cas9- sgRNA expression plasmids and the homology arm vector were electroporated into \(10^{7}\) VL3- 3M2 cells at 340 V for 47 ms in Opti- MEM (Gibco). Neomycin selection was applied after two days. Clonal cells were subsequently generated and screened via PCR using genomic DNA as the template. This involved primers (Table S1) to detect genomic DNA (positive
+
+<--- Page Split --->
+
+control for WT and KO, gDNA F/R, 800 bp amplicon), primers to verify the presence of the WT allele (EndoU validation F/R, 1079 bp amplicon), and primers to identify the KO allele (EndoU validation F/Resistance R, 999 bp amplicon).  
+
+## Mouse EndoU tagged and mutant constructs  
+
+The mouse EndoU cDNA (NM_001168693) was PCR amplified from VL3- 3M2 cDNA with primers containing XhoI (forward) and BglII (reverse) restriction sites and ligated into the pMSCV- PIG or pRL- TK vectors. The Q5 site- directed mutagenesis kit (NEB Cat. E0554S) was used to add a C- terminal FLAG- HA tag, or to create the E285A;H286A catalytically dead mutant version, in pMSCV- PIG.  
+
+## Viral production and stable integration of EndoU rescue constructs  
+
+VL3- 3M2 clonal EndoU knockout cells were rescued through viral integration of the above EndoU constructs. Plat- E cells were calcium- phosphate transfected with \(10 \mu \mathrm{g}\) of pMSCV- PIG and \(2.5 \mu \mathrm{g}\) VSVG to produce amphitropic VSVG- pseudotyped retrovirus.  
+
+## RT-qPCR  
+
+RNA was extracted from whole cells using ribozol followed by two phenol chloroform (pH 5.2) extractions. Superscript II reverse transcriptase was used for cDNA synthesis with \(1 \mu \mathrm{g}\) of total RNA as template. TaqMan probes against EndoU (Cat. 4351372) were used in the RT- qPCR.  
+
+## Cell lysis  
+
+Cell lysis was carried out by first washing the cells once with PBS buffer and then resuspending them in hypotonic lysis buffer (10 mM Tris- HCl pH 7.5, 10 mM KCl, 5 mM DTT, protease inhibitor). The cells were subsequently incubated on ice for 20 minutes. Isotonicity was restored by adjusting the KCl concentration to \(100 \mathrm{mM}\) using a \(5 \mathrm{X}\) supplemental buffer (450 mM KCl,
+
+<--- Page Split --->
+
+0.08 U. \(\mu \mathrm{l}^{- 1}\) RNaseIN). In certain experiments, lysates were centrifuged at \(17 000 \mathrm{g}\) for 20 minutes to separate the cytoplasmic fraction and collect the supernatant.  
+
+## Immunoprecipitations  
+
+Immunoprecipitations were carried out using protein A Dynabeads. Beads were prepared by incubation with \(16.7 \mu \mathrm{g}.\mathrm{ml}^{- 1}\) anti- mouse \(\mathrm{Fc}\gamma\) bridging antibody and \(16.7 \mu \mathrm{g}.\mathrm{ml}^{- 1}\) mouse anti- HA.11 antibody, sequentially. Cell lysates were incubated with prepared beads for 1 hour at room temperature. To equalize the amount of EndoU across reactions, an excess of cell lysate over bead capacity was used, and saturation of EndoU binding was verified by western blot.  
+
+## On-bead mouse EndoU RNase assays  
+
+In a total volume of \(10 \mu \mathrm{L}\) , reactions consisted of (unless used as a variable) \(2 \mathrm{mM}\) calcium, \(100 \mathrm{mM}\) Tris- HCl (pH 7.5), \(10 \mathrm{mM}\) NaCl, \(5 \mu \mathrm{g}\) of total cytoplasmic RNA, or \(1 \mu \mathrm{M}\) of specific RNA oligo (Table S2, typically 50 mer 1), and the immunoprecipitated EndoU. Reactions were incubated at \(37^{\circ} \mathrm{C}\) , RNA was extracted and run on an \(8 \%\) urea- PAGE gel, and visualized by SYBR Green II. Densitometry was used to quantify substrate degradation using Quantity One (BioRad). Experiments were done in triplicate from distinct samples, with central tendencies expressed as means and variations as standard deviation.  
+
+## Production of human EndoU  
+
+The open reading frame (ORF) encoding the catalytic domain of human EndoU (135- 410) was sub- cloned into the pET24b(+) plasmid \((\mathrm{Kan}^{\mathrm{R}})\) downstream of the GB1 protein ORF followed by a hexahistidine tag and a TEV protease cleavage site. Expression of EndoU was achieved in Escherichia coli BL21 Rosetta (DE3) pLysS. The bacteria were grown in rich LB medium or in M9 minimal medium supplemented with \(^{15}\mathrm{N}\) - labeled \(\mathrm{NH_4Cl}\) (1 g. \(\mathrm{L}^{- 1}\) ) and \(^{13}\mathrm{C}\) - labeled glucose (2 g. \(\mathrm{L}^{- 1}\) ) to achieve uniform isotope labeling. The cultures were grown at \(37^{\circ} \mathrm{C}\) until reaching an \(\mathrm{OD}_{600}\) of
+
+<--- Page Split --->
+
+approximately 0.6. Subsequently, protein expression was induced using \(0.25 \mathrm{mM}\) IPTG at \(15^{\circ} \mathrm{C}\) over 16 hours. The bacteria were harvested by centrifugation (5000 g, 10 min, \(4^{\circ} \mathrm{C}\) ), and the resulting pellets were resuspended in ice- cold lysis buffer (20 mM Tris pH 8, 500 mM NaCl, 250 \(\mu \mathrm{L} \cdot \mathrm{L}^{- 1} \beta\) - mercaptoethanol). This buffer was further supplemented with \(1 \mathrm{mg} \cdot \mathrm{mL}^{- 1}\) lysozyme and \(10 \mu \mathrm{L} \cdot \mathrm{L}^{- 1} \mathrm{DNase}\) (NEB). Cell lysis was achieved by sonication, running three cycles of 5 minutes each at \(20 \%\) amplitude, with 20- second on/off intervals. The lysate was clarified by centrifugation (20000 g, 30 min, \(4^{\circ} \mathrm{C}\) ) and the supernatant was loaded onto a gravity- flow histidine affinity chromatography column equilibrated with loading buffer (20 mM Tris pH 8, 500 mM NaCl, 250 \(\mu \mathrm{L} \cdot \mathrm{L}^{- 1} \beta\) - mercaptoethanol). The column was washed with \(15 \mathrm{mM}\) imidazole (10 CV), and the protein was eluted with \(300 \mathrm{mM}\) imidazole (5 CV). The eluted protein was then dialyzed against TEV digestion buffer (10 mM Tris pH 8, 250 mM NaCl, \(125 \mu \mathrm{L} \cdot \mathrm{L}^{- 1} \beta\) - mercaptoethanol) over 16 hours, in the presence of His \(_{6}\) - TEV protease (1:10 w/w ratio) to digest the GB1- His \(_{6}\) tag. Post- digestion, EndoU was isolated from the flow- through fraction following its loading onto a gravity- flow histidine affinity chromatography column, and washing with the loading buffer (5 CV). The resulting protein was concentrated, and a large excess of EDTA (250 mM) was added to chelate potential divalent cations. Further purification was achieved using a Superdex 75 column pre- equilibrated with storage buffer (10 mM Tris pH 7, 50 mM NaCl, \(1 \mathrm{mM}\) TCEP). Finally, EndoU was concentrated to a concentration of \(500 \mu \mathrm{M}\) . It was used immediately for enzymatic assays, while it was stored at \(- 80^{\circ} \mathrm{C}\) for other experiments. Point mutants were generated using the QuickChange protocol [37] and EndoU variants were purified using the same protocol as the wild type protein. The sequences of the oligonucleotides are given (Table S3).  
+
+## Oligonucleotides synthesis  
+
+Oligonucleotides were synthesized using the \(\beta\) - phosphoramidite method with an H8 automated synthesizer (K&A Labs, Germany) on a micromolar scale. For the synthesis of \(2^{\circ} \mathrm{F}\) RNA analogs, sequences started with a Unylinker solid support (Glen Research), and nucleotides were added sequentially using \(2^{\circ} \mathrm{F}\) phosphoramidites. For the synthesis of \(3^{\circ}\) labeled Cyanine 5 RNA, the dye was
+
+<--- Page Split --->
+
+directly attached to the support, and RNA monomers were used. All phosphoramidites and the Cyanine 5 solid support were purchased from LINK (Scotland). Deprotection of the oligonucleotides was performed according to the suppliers protocols. The concentrated crude oligonucleotides were then resuspended in water. The sample concentration was determined from the absorbance at 260 nm and the molar extinction coefficient of the oligonucleotide. This value was calculated using the Integrated DNA Technology online oligo analyzer tool, which uses the standard nearest neighbor method.  
+
+## Nuclear Magnetic Resonance  
+
+Nuclear Magnetic Resonance (NMR) spectroscopy was used to analyze protein structure and dynamics. Experiments were performed using either a Bruker AVIll NMR spectrometer at 700 MHz with a room- temperature probe, or a Bruker Avance NEO spectrometer at 800 MHz with a cryogenic \(5\mathrm{mm}\) TCI \(^{1}\mathrm{H - }^{13}\mathrm{C / }^{15}\mathrm{N / }^{2}\mathrm{H}\) Z- gradient probe. These experiments were carried out at \(35^{\circ}\mathrm{C}\) in a minimal buffer composed of \(10\mathrm{mM}\) Tris (pH 7), \(50\mathrm{mM}\) NaCl, \(1\mathrm{mM}\) TCEP, and \(10\%\) \(\mathrm{D}_2\mathrm{O}\) for field frequency lock. We acquired 2D \(^{1}\mathrm{H - }^{15}\mathrm{N}\) and \(^{1}\mathrm{H - }^{13}\mathrm{C}\) correlation spectra using the SOFAST- HMQC experiment scheme [38]. Sequence- specific backbone assignments of \(^{15}\mathrm{N}^{13}\mathrm{C}\) - labeled calcium- activated EndoU were achieved via classical 3D triple resonance experiments based on the BEST- TROSY principle [39, 40]. The same approach was applied to EndoU bound to RNA targets. Spectra processing was conducted with Topspin 4 (Bruker) and analyzed using CARA [41] and CCPNMR software 2.4 [42]. Combined \(^{1}\mathrm{H - }^{15}\mathrm{N}\) chemical shift perturbations \((\Delta \delta_{\mathrm{comb}})\) were calculated as \(\Delta \delta_{comb} = \sqrt{\Delta\delta^{1}H + 0.14} * \Delta \delta^{15}N\) , where \(\Delta \delta^{1}H\) and \(\Delta \delta^{15}N\) are the chemical shift perturbations (in ppm) for \(^{1}\mathrm{H}\) and \(^{15}\mathrm{N}\) resonances, respectively [43]. NMR titrations to map the RNA binding surface on calcium- bound EndoU were performed using a non- cleavable, \(2^{\prime}\) - fluorinated RNA obtained in house via solid- phase synthesis with the following sequence: \(5^{\prime}\) - AAGUCC- 3'.
+
+<--- Page Split --->
+
+## Structure Determination  
+
+Structure DeterminationA sample of the catalytic domain of human EndoU, spanning residues 135 to 410, was prepared at a concentration of \(12 \mathrm{mg}.\mathrm{mL}^{- 1}\) in a buffer containing \(10 \mathrm{mM}\) Tris pH 7, \(50 \mathrm{mM}\) NaCl, \(1 \mathrm{mM}\) TCEP, and \(20 \mathrm{mM}\) \(\mathrm{CaCl}_2\) . The crystallization of EndoU was carried out at \(20^{\circ}\mathrm{C}\) using the MCSG4 matrix screen, specifically condition F6, which comprises \(0.1 \mathrm{M}\) sodium acetate, \(0.1 \mathrm{M}\) HEPES pH 7.5, and \(22 \%\) PEG 4k. The resulting crystals were flash- frozen in liquid nitrogen using a cryoprotectant solution identical to the crystallization condition but supplemented with \(20 \%\) ethylene glycol. Diffraction data were collected at the SOLEIL synchrotron on the PX1 beamline and processed using XDS [44]. Molecular replacement was conducted with Phaser [45] from the Phenix suite [46], using the AlphaFold 2 [47] predicted structure of the human EndoU protein as the model. This process identified two molecules per asymmetric unit, which were subsequently refined using Phenix and BUSTER [48]. Detailed crystallographic data and refinement statistics are presented in the supplementary materials (Table S4).  
+
+## Enzymatic RNA Degradation Assays  
+
+Enzymatic RNA Degradation AssaysRNA degradation assays on the human catalytic domain were carried out to assess the relative activity of EndoU and its variants. An RNA sequence, 5'- CAGGUUUCCCCAACGAAAAAAAAAA- 3', was obtained in- house via solid- phase synthesis. The RNA was labeled at the 3' end with a Cyanine- 5 (Cy5) fluorescent probe for detection purposes. In each assay, EndoU or one of its variants was prepared at a final concentration of \(1 \mathrm{nM}\) in presence of \(1 \mu \mathrm{M}\) of the RNA. The enzymatic reaction was initiated by introducing \(2 \mathrm{mM}\) \(\mathrm{CaCl}_2\) into the mixture. Samples were collected at 15 time points: 0, 1, 3, 5, 10, 15, 20, 25, 30, 40, 50, 60, 80, 100, and 120 minutes. The reaction was terminated at each time point with an excess of EDTA to chelate calcium ions in order to prevent EndoU activation and further RNA degradation. RNA degradation was monitored by resolving the samples on a polyacrylamide gel containing \(6 \mathrm{M}\) urea, followed by electrophoresis at \(250 \mathrm{V}\) for 50 minutes. The gel was then scanned with a fluorescence scanner. We used the GelAnalyzer
+
+<--- Page Split --->
+
+software [49] to integrate band intensities, which were normalized relative to the zero time point. Each assay was conducted in triplicate to ensure the reproducibility of the results. Measurements were taken from distinct samples for each replicate; central tendencies are expressed as means, and variations as standard deviations. Data were processed and analyzed using custom Python scripts. A first- order reaction model, \(A * exp(- k * t)\) , was used to fit the enzymatic progress curve, using the curve_fit function from the scipy.optimize module for regression [50]. The fitted reaction rate \(k\) characterizes the activity of each EndoU variant. To enable comparison across different variants, this reaction rate was subsequently expressed in relative terms with respect to the wild- type EndoU, yielding a dimensionless parameter.  
+
+## Molecular Dynamics  
+
+Molecular dynamics simulations were performed using the GROMACS software package (version 2022.1) [51]. System preparation was achieved through CHARMM- GUI and the Input Generator module [52, 53], which was also used to apply single amino- acid substitutions for EndoU variants. To monitor the stability of the calcium binding sites, we used the crystal structure of EndoU bound to calcium ions as input. To monitor the binding of calcium to apo- EndoU, we removed calcium ions from the crystal structure and used the resulting structure as input. Calcium ions were then reintroduced into the system as salt ions. To propose an ensemble of models of calcium- activated EndoU in complex with RNA, we used the AlphaFold (AF) 3 [54] webserver with the human EndoU sequence (135- 410), a \((\mathrm{U})_6\) RNA, and four calcium ions as inputs. The top- ranked model accurately reproduced each of the calcium binding sites, with an RMSD of 0.313 Å between the crystal structure of calcium- activated EndoU and the corresponding part of the AF3 model. Therefore, we created a hybrid model comprising the experimental structure of calcium- activated EndoU in complex with the AF3- modeled bound RNA. This resulting model was used as input for MD simulations. For all simulations, we used the CHARMM36m force field [55] and the TIP3P water model. Each system was solvated in a cubic box with a 1.0 nm buffer zone between the protein and the box edge, and 50 mM NaCl was added with adjustments to neutralize the system. After down
+
+<--- Page Split --->
+
+loading the generated inputs, energy minimization was executed in GROMACS using the steepest descent algorithm until the maximum force was below \(1000 \mathrm{kJ} \cdot \mathrm{mol}^{- 1} \cdot \mathrm{nm}^{- 1}\) , and then equilibration was done under NVT conditions for \(125 \mathrm{ps}\) . Particle Mesh Ewald was used for long- range electrostatics, with a cutoff of \(1.0 \mathrm{nm}\) for van der Waals interactions, and a time step of \(2 \mathrm{fs}\) was applied. The production phase of the simulations was carried out for \(1 \mu \mathrm{s}\) under NPT conditions at a temperature of \(35^{\circ} \mathrm{C}\) and a pressure of \(1 \mathrm{bar}\) . The output was then analyzed for various parameters using the built- in tools of GROMACS.  
+
+## SEC-SAXS experiments  
+
+SEC- SAXS experimentsSEC- SAXS experiments were conducted on the SWING beamline at the SOLEIL synchrotron (Saint- Aubin, France). All procedures were carried out at a temperature of \(35^{\circ} \mathrm{C}\) using a buffer composed of \(10 \mathrm{mM}\) Tris pH 7, \(50 \mathrm{mM}\) NaCl, and \(1 \mathrm{mM}\) TCEP. EndoU was prepared to a concentration of \(500 \mu \mathrm{M}\) . A volume of \(75 \mu \mathrm{L}\) was injected onto a size exclusion column (Bio- SEC 3 Agilent \(100 \AA\) ), and was then eluted directly into the SAXS flow- through capillary cell at a flow rate of \(0.3 \mathrm{mL} \cdot \mathrm{min}^{- 1}\) . SAXS data were collected using an EigerX 4M detector situated \(2 \mathrm{m}\) away, using the definition of the momentum transfer \(q: q = 4 \pi \sin (\theta) / \lambda\) , where \(2 \theta\) represents the scattering angle and \(\lambda\) the X- ray wavelength (1.033 \(\AA\) for these experiments). The overall SEC- SAXS setup has been described in earlier publications [56, 57, 58]. A total of 900 SAXS frames were continuously recorded during elution, each with a duration of \(1.99 \mathrm{s}\) and a \(0.01 \mathrm{s}\) dead time between frames. 180 frames were collected before the dead volume to account for buffer scattering. Data reduction to absolute units, buffer subtraction, and averaging of identical frames corresponding to the elution peak were performed using the in- house SWING software FOXTROT [57] and BioXTAS [59]. BioXTAS was also employed to compute the gyration ratio and to estimate the molecular weight based on the volume of correlation [60]. The fitting of the EndoU homology model to the experimental SAXS data were accomplished through the Crysol software, part of the ATSAS Suite [61, 62].
+
+<--- Page Split --->
+
+## Intrinsic Fluorescence  
+
+To assess the impact of calcium binding on the tertiary structure of EndoU, we measured the intrinsic fluorescence of the protein with a temperature- controlled spectrofluorometer (FS5, Edinburgh Instruments). Protein samples were prepared at \(10~\mu \mathrm{M}\) in a buffer of \(10~\mathrm{mM}\) Tris pH 7, \(50~\mathrm{mM}\) NaCl, and \(1\mathrm{mM}\) TCEP, and their fluorescence emission spectra were recorded at \(35^{\circ}\mathrm{C}\) . Emissions from \(300\) to \(525~\mathrm{nm}\) were recorded to detect fluorescence from tryptophan, tyrosine, and phenylalanine. Slit widths for excitation and emission were set at \(5\mathrm{nm}\) , and each spectrum was an average of three scans, corrected for buffer baseline.
+
+<--- Page Split --->
+
+## Results  
+
+## EndoU Expression and RNase Activity in a Thymocyte Model  
+
+In mammals, EndoU expression is limited to specific cell types. Analysis of Immunological Genome Project data [63] on mRNA from 211 mouse hematopoietic cell types revealed strong EndoU expression in developing thymocytes, starting at the double negative (DN) 2- 3 transition and progressing through the double positive (DP) stages (Fig. 1 A). EndoU expression is absent in the later stages: single positive thymocytes that survive selection and circulating T cells. Outside the hematopoietic system, EndoU protein staining in human samples [64] showed cytoplasmic expression in stratified squamous epithelia (e.g., skin, esophagus, cervix) and the trophoblast layer in the placenta.  
+
+For molecular and biochemical analysis of EndoU, we used the mouse thymic lymphoma cell line VL3- 3M2 [35], which resembles double positive thymocytes with high EndoU expression (Fig. S1 A). Upon stimulation with PMA/ionomycin, anti- CD3/CD28, or anti- CD3/CD4, the cell line shows further maturation, including downregulation of Rag1 and EndoU and upregulation of CD5 (Fig. S1 B). EndoU was knocked out in VL3- 3M2 cells using CRISPR/Cas9. We confirmed the deletion of the genomic region (Fig. S1 C) and the loss of EndoU mRNA (Fig. S1 D). We assayed endogenous ribonuclease activity in WT and EndoU KO VL3- 3M2 extracts, based on experiments in Xenopus laevis egg extracts [25]. Incubation of cytoplasmic extracts at \(37^{\circ}\mathrm{C}\) for 15 minutes without divalent metals caused no RNA degradation (Fig. 1 B). However, WT extracts with \(5\mathrm{mM}\) \(\mathrm{Ca}^{2 + }\) showed robust RNA cleavage, which was absent in EndoU KO extracts and rescued by expressing WT or HA- tagged EndoU in KO cells (Fig. 1 B). Thus, VL3- 3M2 extracts have strong EndoU- and \(\mathrm{Ca}^{2 + }\) - dependent ribonuclease activity.  
+
+To confirm the role and enzymatic properties of EndoU, we used HA- tagged EndoU rescue cells for on- bead in vitro cleavage assays with immunoprecipitated EndoU. Since mouse EndoU cleaved various RNA sequences (Fig. S2 A), we used an arbitrary substrate (50 mer 1) for subsequent ex
+
+<--- Page Split --->
+
+periments (Table S2). Time course measurements (Fig. S1 B) were used to calculate initial reaction rates (Fig. S2 C). Mutation of two critical residues [65] abolished activity (Fig. 1 D), confirming the role of EndoU in RNA cleavage. We showed that only \(\mathrm{Ca^{2 + }}\) stimulated cleavage, unlike \(\mathrm{Mn^{2 + }}\) or other divalent metals (Fig. 1 E, S2 B), with optimal activity at \(1 - 2 \mathrm{mM} \mathrm{Ca^{2 + }}\) . EndoU showed little dependence on \(\mathrm{Na^{+}}\) or \(\mathrm{K^{+}}\) and sustained activity across pH 4- 8 (Fig. S2 D, E, F). These results indicate EndoU is a calcium- activated ribonuclease targeting a large repertoire of RNAs.  
+
+## Structural Basis for EndoU Activation by Calcium Ions  
+
+To examine structural changes in EndoU upon \(\mathrm{Ca^{2 + }}\) activation, we first computed a homology model of human EndoU using XendoU crystal structure [34] as template via SWISS- MODEL [66]. The apo- EndoU structure is globular, with a predominantly \(\beta\) - sheet catalytic core and an \(\alpha\) - helical bundle N- terminal extension (Fig. S3 A). We expressed the XendoU catalytic domain (Fig. 1 D) of human EndoU with \(^{15}\mathrm{N}\) labeling for NMR spectroscopy. The \(^{15}\mathrm{N}\) SOFAST- HMQC spectrum of apo- EndoU showed well- dispersed signals but fewer than expected, suggesting conformational exchange in the \(\mu \mathrm{s}\) - ms range (Fig. S3 B). Size Exclusion Chromatography with Small- Angle X- ray Scattering (SEC- SAXS) validated the correct folding of recombinant EndoU, matching theoretical predictions (Fig. S3 C). Furthermore, the crystal program [67] showed strong correlation between experimental and theoretical SAXS data, validating the structural model (Fig. S3 D).  
+
+We then studied the effect of divalent metal ions on EndoU structure and dynamics by comparing \(^{15}\mathrm{N}\) SOFAST- HMQC spectra of apo- EndoU to metal- bound states. Saturating concentrations of magnesium, nickel, or strontium caused signal loss in the NMR spectra, suggesting either protein aggregation or increased conformational exchange (Fig. S4 A, B, C). In contrast, saturating calcium restored a set of well- dispersed peaks in the NMR spectrum (Fig. S4 D). At sub- saturating calcium levels, we observed chemical shift perturbations and a shift in the fluorescence spectrum of the protein (Fig. S5 A, B). Intriguingly, the addition of a \(2'\) - fluorinated nonhydrolyzable RNA in the presence of sub- saturating calcium produced effects similar to those observed with saturating calcium, including the restoration of a set of well- dispersed peaks (Fig. S5 C). This was not
+
+<--- Page Split --->
+
+observed in the absence of calcium, where the substrate analog did not significantly alter the NMR spectrum (Fig. S5 D). These findings suggests a two- step activation process involving local structural changes at lower calcium concentrations and the abrogation of conformational exchange in the \(\mu \mathrm{s}\) - ms range at higher calcium concentrations or upon binding of a substrate analog.  
+
+To elucidate the effect of calcium, we crystallized EndoU with an excess of calcium and solved its structure at 1.7 Å resolution. The structure revealed five calcium ions, with one aiding crystal packing and four potentially activating the protein (Fig. 2 A, S6). Each calcium ion is coordinated by seven oxygen atoms from acidic side- chains, backbone carbonyl groups, or protein- stabilized water molecules (Fig. 2 B). Sites (1) and (3) include residues from both the catalytic core and the eukaryotic- specific N- terminal extension (Fig. S7), with site (1) located 12.8 Å away from the catalytic triad (H285, H300, K343). Comparing apo- XendoU, apo- EndoU, and calcium- activated EndoU structures revealed conformational changes upon calcium binding that mainly cluster nearby protein loops (Fig. 2 C, D). The side- chain of E290, located midway between site (1) and the catalytic triad, flips to engage with a water molecule in the calcium coordination network. This correlates with the side- chain rotation of catalytic H285, then locked by an electrostatic interaction with E290 (Fig. 2 C). The bonding correlates with a disorder- to- order transition in the loop carrying E290, forming a \(\beta\) - hairpin and stabilizing the catalytic site (Fig. 2 D). Our data show a model for the calcium- dependent regulation of EndoU through allostery, with site (1) and residue E290 as key mediators in the intramolecular signaling leading to EndoU activation.  
+
+## Experimental Validation of EndoU Activation Model  
+
+To validate our structure- based model for calcium- mediated EndoU activation, we first designed variants with altered calcium- binding sites. RNA degradation assays were conducted for each variant (Fig. 3 A). Without calcium or wild- type EndoU, no RNA cleavage was detected, whereas their presence led to almost complete RNA degradation over time. The degradation data fitted a first- order reaction model, providing a kinetic parameter describing the reaction (Fig. 3 B). Disrupting calcium binding sites (2) or (4) with mutations E226A or D330A led to RNA degradation rates
+
+<--- Page Split --->
+
+similar to the wild- type (Fig. 3 A, C). In contrast, disruption of calcium binding sites (1) or (3) with mutations E284A or D179A abolished enzymatic activity. Interestingly, calcium binding sites (1) and (3) are defined by residues from both the catalytic core and the eukaryote- specific N- terminal extension, while this is not the case for sites (2) and (4) (Fig. S7).  
+
+Our results clearly indicate that the eukaryote- specific N- terminal extension of EndoU contributes to calcium sensing, thereby enabling allosteric regulation. Even though the crystal structure of calcium- activated EndoU could explain the role of calcium binding site (1) in this process, it was not the case for site (3). We hypothesized that calcium binding to site (1) could be promoted by a prior binding event at site (3) and relied on Molecular Dynamics (MD) experiments to test this hypothesis. With the wild- type EndoU, we observed calcium binding to site (1) within less than 100 ns simulation time (Fig. 3 D). Disrupting site (3) with mutation D179A resulted in no stable binding at site (1) for any of the five calcium ions added in the \(1 \mu \mathrm{s}\) simulation. This suggests that that cooperative binding of calcium at sites (1) and (3) leads to EndoU activation.  
+
+We further proposed that calcium sensing information at site (1) was communicated to the remote catalytic site through water- mediated intramolecular signaling events enabled by key residue E290 (Fig. 2 C). Disruption of the intramolecular signaling cascade with mutation E290A completely abrogated activity, while charge- conservative mutation E290D resulted in a 4- fold increase in the enzymatic reaction rate (Fig. 3A, C). Consistent with our structural model, a negatively charged side- chain at position 290 is required for the allosteric activation of EndoU by calcium. In this model, E290 locks the catalytic H285 side- chain in an active conformation. Accordingly, substitution of catalytic H285 with alanine completely abrogated enzymatic activity, as observed for catalytic mutant H300A, underscoring the importance of the histidine pair in catalysis. Substitution of catalytic residue K343 by alanine resulted in nearly half reduction of enzymatic activity, consistent with the role of K343 as a stabilizer of reaction intermediates. Overall, mutagenesis experiments corroborate the residue assignments proposed in our structure- based model for calcium- mediated activation of EndoU.
+
+<--- Page Split --->
+
+## Calcium-Activated EndoU in Complex with an RNA Analog  
+
+To experimentally determine the RNA- binding surface of calcium- activated EndoU, we first performed 3D triple resonance NMR experiments for backbone chemical shift assignment, successfully assigning \(89.3\%\) (251 out of 281 residues) of the backbone resonances (Fig. S8). Excluding the N- terminal GGSEFA sequence and nine proline residues, the assignment coverage increased to \(94.3\%\) . Next, we recorded \(^{15}\mathrm{N}\) and \(^{13}\mathrm{C}\) SOFAST- HMQC spectra of calcium- activated \(^{13}\mathrm{C}^{15}\mathrm{N}\) - labeled EndoU with a \(2^{\prime}\) - fluorinated RNA substrate analog (Fig. 4 A, B). Severe line broadening in a subset of crosspeaks and additional spectral changes in the rapid exchange regime were observed (Fig. 4 A). We calculated \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) chemical shift perturbations between RNA- bound and unbound EndoU, noting particularly the residues that disappeared in the bound state (Fig. 4 C). All disappearing residues are located in the C- terminal catalytic core, covering the \(\beta\) - sheet surface composed of two independent \(\beta\) - sheets and a short \(\alpha\) - helix (Fig. 4 D). The strongest chemical shift perturbations also correspond to residues in this area. Electrostatic analysis indicated that the \(\beta\) - sheet is highly basic, suitable for RNA binding (Fig. 4 E). In contrast, the N- terminal \(\alpha\) - helical region showed minimal changes upon RNA binding, with no disappearing resonances, and displayed a neutral or acidic surface charge. These data support that RNA binds to the conserved catalytic core of the protein, involving an extended, basic \(\beta\) - sheet- rich groove for RNA binding.  
+
+To model the complex between calcium- activated EndoU and RNA, we used AlphaFold (AF) 3 [54] with the primary sequence of EndoU, a \((\mathrm{U})_6\) RNA, and four calcium ions as inputs. We controlled that a \(2^{\prime}\mathrm{F}(\mathrm{U})_6\) RNA analog interacts with calcium- activated EndoU (Fig. S9). Using the top- ranked AF model as input, we then conducted a \(1 \mu \mathrm{s}\) Molecular Dynamics (MD) simulation using the top- ranked AF model as input. Over the trajectory, we clustered structures based on RNA conformations to obtain a final ensemble of RNA- bound calcium- activated EndoU models (Fig. S10 A). All models consistently reproduced the crystal structure of calcium- activated EndoU, with a mean RMSD of \(1.18 \pm 0.29 \mathrm{\AA}\) . The intermolecular interface with the \((\mathrm{U})_6\) RNA is defined by the cleft between the two front \(\beta\) - sheets, where nucleotide \(\mathrm{U}_2\) anchors in close proximity to the catalytic
+
+<--- Page Split --->
+
+triad, exposing its sugar 2'OH for nucleophilic attack by residue H300 (Fig. S10 B, S11). Conformational heterogeneity is observed at the interface with EndoU across the models for the rest of the RNA sequence, consistent with the experimental NMR data showing line broadening for residues in the RNA- binding region due to conformational sampling in the \(\mu \mathrm{s}\) - ms range. Overall, the ensemble of models aligns well with our experimental data, providing a robust structural hypothesis for calcium- activated, RNA- bound EndoU.
+
+<--- Page Split --->
+
+## Discussion  
+
+In this work, we elucidated the molecular basis for the \(\mathrm{Ca^{2 + }}\) - dependent activation of human EndoU, with implications for the entire eukaryotic EndoU family due to the conserved sequence and structure of its catalytic domain across eukarya. Our findings, in conjunction with existing data, suggest an allosteric rather than catalytic requirement for a divalent metal in EndoU cleavage (Fig. 5).  
+
+Indeed, a common evolutionary origin with the RNase A family has been proposed based on structural and distant sequence similarities [6]. RNase A enzymes use a catalytic triad of two histidines and a lysine [68], and the Xenopus EndoU structure suggested a mechanistic similarity, where conserved His and Lys residues mark the proposed catalytic site [34]. Mutation of the corresponding residues in human EndoU supported these assignments (Fig. 3 C). Then, both EndoU and RNase A leaves a \(5^{\prime}\) - OH product, which is characteristic of metal- independent endonuclease catalytic catalysis [69]. Also, bacterial and some viral EndoU homologs do not require divalent metal ions [31, 11]. In the context of eukaryotic EndoU, our experimental data identified that calcium binding was necessary for cleavage activity in both mouse and human EndoU (Fig. 2, 3 C), with coordinating residues from both the conserved catalytic core and the eukaryote- specific N- terminal extension. We explained how the binding of calcium at the interface between the two regions could trigger water- mediated intramolecular signaling, ultimately leading to local structural changes that result in the positioning of catalytic residues in an active conformation. Therefore, our work supports a model where calcium is not directly involved in catalysis but rather activates the catalytic His- His- Lys triad through allostery. At the same time, we assign a role to the eukaryotic- specific N- terminal extension in sensing calcium at relevant sites, enabling allosteric regulation by calcium ions.  
+
+Eukaryotic EndoU proteins primarily depend on calcium for activation, as our data for mouse and human EndoU indicate exclusive activation by \(\mathrm{Ca^{2 + }}\) , consistent with Xenopus EndoU findings [25]. Although early Xenopus studies also reported \(\mathrm{Mn^{2 + }}\) requirements, \(\mathrm{Ca^{2 + }}\) - dependent cleavage was observed [13], and the C. elegans endu- 2 homolog is activated by both \(\mathrm{Ca^{2 + }}\) and \(\mathrm{Mn^{2 + }}\) [70]. The
+
+<--- Page Split --->
+
+physiological relevance of \(\mathrm{Mn}^{2 + }\) activation is unclear due to its low concentration in mammalian tissues [71, 72, 73], whereas \(\mathrm{Ca}^{2 + }\) concentrations vary significantly, making EndoU localization crucial for understanding \(\mathrm{Ca}^{2 + }\) - mediated activation. Animal EndoUs contain an N- terminal signal peptide, and mammalian EndoUs have two somatomedin B (SmB) domains rich in disulfide bonds, suggesting secretion or ER association [74, 75, 76]. The C. elegans endu- 2 and D. melanogaster CG2145 homologs are secreted and reuptaken in other tissues [70, 77], while Xenopus and human EndoU are cytoplasmic and ER- associated [25]. Calcium, a universal eukaryotic second messenger, is near zero in resting cells and rises to 1- 100 \(\mu \mathrm{M}\) concentration during signaling, while extracellular \(\mathrm{Ca}^{2 + }\) can reach mM levels [78]. EndoU may cleave RNAs or alter mRNA expression during thymocyte maturation or apoptosis, consistent with its proposed pro- apoptotic role in B cells [27]. The \(\mathrm{Ca}^{2 + }\) activation mechanism is likely conserved across eukaryotic EndoUs, with varying expression domains and RNA targets across species. If delivered to \(\mathrm{Ca}^{2 + }\) - rich extracellular environments, EndoU may degrade extracellular RNAs, which makes the understanding of the RNA targeting repertoire of EndoU a priority to further illuminate its biological roles.  
+
+Our work provides fundamental insights with potential applications in therapeutic and biotechnological domains. The Nsp15 protein of SARS- CoV- 2 features a nidovirus EndoU- like domain (NendoU) that requires divalent manganese ions to cleave the \(5^{\prime}\) - polyuridine tract of its negative- sense RNA, a crucial process for evading the host immune system [79]. Tipiracil, a uridine analog, effectively binds to the uridine binding site of Nsp15, inhibiting its activity and diminishing Spike protein expression in whole- cell assays, thereby inhibiting SARS- CoV- 2 [80]. However, the structural basis for the divalent metal ion dependence of Nsp15 remains unknown, posing challenges for rational drug design targeting Nsp15 and other members of the EndoU- like superfamily regulated by divalent metal ions. By elucidating the structural basis of the divalent metal ion dependence of human EndoU, we pave the way for developing inhibitors specifically targeting EndoU- like domains that depend on these ions. Furthermore, investigating whether Tipiracil can bind to and inhibit human EndoU could provide valuable insights into the cellular and extracellular functions of EndoU, potentially identifying it as a pharmacological target in certain disease contexts. The
+
+<--- Page Split --->
+
+potential of EndoU as a biotechnological tool for specific RNA sequence cleavage is noteworthy, especially with advancements in self- driving laboratories and machine learning algorithms for enzyme design and optimization [81, 82, 83]. Altering the divalent metal ion dependence of EndoU could yield sensors for detecting environmental contaminants like lead ions in water or enable controlled RNA digestion in biological systems by introducing activating ions, offering significant insights into RNA metabolism and processes.  
+
+In conclusion, we have elucidated the molecular mechanism underlying the activation of EndoU enzymes by divalent metal ions, specifically calcium. This discovery demonstrates a functional role for the N- terminal extension in human EndoU, suggesting a conserved mechanism accross eukaryotic EndoU catalytic domains. Evolution has provided these enzymes with a distinct regulatory segment that interacts with their catalytic core to sense calcium binding and communicate with the catalytic site through allostery. Our findings pave the way for more research into the role of EndoU in biological processes and offer valuable insights for future biotechnological and therapeutic applications.
+
+<--- Page Split --->
+
+## Acknowledgements  
+
+AcknowledgementsWe would like to thank Stephen Smale for sharing the VL3- 3M2 cell line, and the Zhang lab for sharing the pX330 plasmid. This work is funded in part by NIH grant 1R01GM152548. We are grateful to Dr. Mikayel Aznauryan for providing us with access to the fluorescence spectroscopy facilities. We thank INSERM for funding through the ARNA internal call. We acknowledge SOLEIL for provision of synchrotron radiation facilities and we would like to thank PX1 staff members for assistance in using beamline PX1. We would like to thank Dr. Aurélien Thureau from SWING beamline at SOLEIL for assistance during SAXS data collection.  
+
+## Data availability  
+
+Data availabilityThe atomic coordinates of the calcium- activated EndoU structure have been deposited in the PDB under the accession code 9FTW.  
+
+## Author contributions  
+
+Author contributionsS.C., F.V.K., and F.M. designed the research. P.B. and B.V. synthesized the RNAs and 2'F analogs used in the study. K.D. and F.V.K. performed experiments on the mouse thymic lymphoma cell line VL3- 3M2, including CRISPR/Cas9 knockout of EndoU and biochemical characterization of both endogenous and immunoprecipitated EndoU. S.F. and S.T. solved the crystal structure of calcium- activated EndoU. F.M. and S.C. conducted NMR spectroscopy studies. F.M. and M.B. carried out RNA degradation assays with recombinant EndoU. F.M. conducted SAXS experiments, AlphaFold modeling, and Molecular Dynamics simulations. F.M. and F.V.K. wrote the initial draft of the manuscript. F.M., S.C., F.V.K., S.F., and S.T. discussed the initial draft and contributed to the final version of the manuscript.
+
+<--- Page Split --->
+
+## Competing interests  
+
+The authors declare no competing interests.
+
+<--- Page Split --->
+
+## References  
+
+[1] Kushner SR. mRNA decay in prokaryotes and eukaryotes: different approaches to a similar problem. IUBMB life, 56(10):585- 94, 2004. [2] Evguenieva- Hackenberg E and Klug G. RNA degradation in Archaea and Gram- negative bacteria different from Escherichia coli. Progress in molecular biology and translational science, 85:275- 317, 2009. [3] Schmid M and Jensen TH. The exosome: a multipurpose RNA- decay machine. Trends in biochemical sciences, 33(10):501- 10, 2008. [4] Tomecki R and Dziembowski A. Novel endoribonucleases as central players in various pathways of eukaryotic RNA metabolism. RNA (New York, N.Y.), 16(9):1692- 724, 2010. [5] Paysan- Lafosse T, Blum M, Chuguransky S, Grego T, Pinto BL, Salazar GA, Bileschi ML, Bork P, Bridge A, Colwell L, Gough J, Haft DH, Letunic I, Marchler- Bauer A, Mi H, Natale DA, Orengo CA, Pandurangan AP, Rivoire C, Sigrist CJA, Sillitoe I, Thanki N, Thomas PD, Tosatto SCE, Wu CH, and Bateman A. InterPro in 2022. Nucleic acids research, 51(D1):D418- D427, 2023. [6] Mushegian A, Sorokina I, Eroshkin A, and Dlakic M. An ancient evolutionary connection between Ribonuclease A and EndoU families. RNA (New York, N.Y.), 26(7):803- 813, 2020. [7] Raines RT. Ribonuclease A. Chemical reviews, 98(3):1045- 1066, 1998. [8] SMYTH DG, STEIN WH, and MOORE S. The sequence of amino acid residues in bovine pancreatic ribonuclease: revisions and confirmations. The Journal of biological chemistry, 238:227- 34, 1963. [9] Findlay D, Herries DG, Mathias AP, Rabin BR, and Ross CA. The active site and mechanism of action of bovine pancreatic ribonuclease. 7. The catalytic mechanism. The Biochemical journal, 85(1):152- 3, 1962. [10] Avey HP, Boles MO, Carlisle CH, Evans SA, Morris SJ, Palmer RA, Woolhouse BA, and Shall S. Structure of ribonuclease. Nature, 213(5076):557- 62, 1967. [11] Michalska K, Quan Nhan D, Willett JLE, Stols LM, Eschenfeldt WH, Jones AM, Nguyen JY, Koskiniemi S, Low DA, Goulding CW, Joachimiak A, and Hayes CS. Functional plasticity of antibacterial EndoU toxins. Molecular microbiology, 109(4):509- 527, 2018. [12] Caffarelli E, Arese M, Santoro B, Fragapane P, and Bozzoni I. In vitro study of processing of the intron- encoded U16 small nucleolar RNA in Xenopus laevis. Molecular and cellular biology, 14(5):2966- 74, 1994.
+
+<--- Page Split --->
+
+[13] Caffarelli E, Maggi L, Fatica A, Jiricny J, and Bozzoni I. A novel Mn++- dependent ribonuclease that functions in U16 SnoRNA processing in X. laevis. Biochemical and biophysical research communications, 233(2):514- 7, 1997.  
+
+[14] Gioia U, Laneve P, Dlakic M, Arceci M, Bozzoni I, and Caffarelli E. Functional characterization of Xen- doU, the endoribonuclease involved in small nucleolar RNA biosynthesis. The Journal of biological chemistry, 280(19):18996- 9002, 2005.  
+
+[15] Laneve P, Altieri F, Fiori ME, Scaloni A, Bozzoni I, and Caffarelli E. Purification, cloning, and characterization of XenDoU, a novel endoribonuclease involved in processing of intron- encoded small nucleolar RNAs in Xenopus laevis. The Journal of biological chemistry, 278(15):13026- 32, 2003.  
+
+[16] Mistry J, Chuguransky S, Williams L, Qureshi M, Salazar GA, Sonnhammer ELL, Tosatto SCE, Paladin L, Raj S, Richardson LJ, Finn RD, and Bateman A. Pfam: The protein families database in 2021. Nucleic acids research, 49(D1):D412- D419, 2021.  
+
+[17] Bohn H and Winckler W. [Isolation and characterization of a new placental tissue protein (PP11) (author's transl)]. Archives of gynecology, 229(4):293- 301, 1980.  
+
+[18] Inaba N, Renk T, Weinmann E, and Bohn H. Immunohistochemical location of proteins related to pregnancy- specific beta 1- glycoprotein (SP1) and placental tissue proteins (PP5, PP8, PP9, PP10, PP11, PP12) in the after- birth of Cynomolgus monkeys (Macaca fascicularis). Archives of gynecology, 230(3):195- 204, 1981.  
+
+[19] Xu C, Zhang Y, Shen Y, Shi Y, Zhang M, and Zhou L. Integrated Analysis Reveals <i>ENDOU</i> as a Biomarker in Head and Neck Squamous Cell Carcinoma Progression. Frontiers in oncology, 10:522332, 2020.  
+
+[20] Basic V, Zhang B, Domert J, Pellas U, and Tot T. Integrative meta- analysis of gene expression profiles identifies FEN1 and ENDOU as potential diagnostic biomarkers for cervical squamous cell carcinoma. Oncology letters, 22(6):840, 2021.  
+
+[21] Lou Y, Ren L, and Wang L. Hsa_circ_0049396 inhibited oral squamous cell carcinoma progression by regulating the miR- 663b/ENDOU axis. Genomics, 115(1):110553, 2023.  
+
+[22] Inaba N, Ishige H, Ijichi M, Satoh N, Ohkawa R, Sekiya S, Shirotake S, Takamizawa H, Renk T, and Bohn H. Immunohistochemical detection of pregnancy- specific protein (SP1) and placenta- specific tissue proteins (PP5, PP10, PP11 and PP12) in ovarian adenocarcinomas. Oncodevelopmental biology and medicine: the journal of the International Society for Oncodevelopmental Biology and Medicine, 3(5- 6):379- 89, 1982.  
+
+[23] Inaba N, Renk T, Wurster K, Rapp W, and Bohn H. Ectopic synthesis of pregnancy specific beta 1- glycoprotein (SP1) and placental specific tissue proteins (PP5, PP10, PP11, PP12) in nontrophoblastic malignant tumours. Possible markers in oncology. Klinische Wochenschrift, 58(15):789- 91, 1980.
+
+<--- Page Split --->
+
+[24] Inaba N, Renk T, Daume E, and Bohn H. Ectopic production of placenta-"specific" tissue proteins (PP5 and PP11) by malignant breast tumors. Archives of gynecology, 231(1):87- 90, 1981.  
+
+[25] Schwarz DS and Blower MD. The calcium- dependent ribonuclease XendoU promotes ER network formation through local RNA degradation. The Journal of cell biology, 207(1):41- 57, 2014.  
+
+[26] Zhao G and Blackstone C. ER morphology: sculpting with XendoU. Current biology: CB, 24(24):R1170- 2, 2014.  
+
+[27] Poe JC, Kountikov EI, Lykken JM, Natarajan A, Marchuk DA, and Tedder TF. EndoU is a novel regulator of AICD during peripheral B cell selection. The Journal of experimental medicine, 211(1):57- 69, 2014.  
+
+[28] Laneve P, Piacentini L, Casale AM, Capauto D, Gioia U, Cappucci U, Di Carlo V, Bozzoni I, Di Micco P, Morea V, Di Franco CA, and Caffarelli E. Drosophila CG3303 is an essential endoribonuclease linked to TDP- 43- mediated neurodegeneration. Scientific reports, 7:41559, 2017.  
+
+[29] Ujisawa T, Ohta A, Ii T, Minakuchi Y, Toyoda A, Ii M, and Kuhara A. Endoribonuclease ENDU- 2 regulates multiple traits including cold tolerance via cell autonomous and nonautonomous controls in <i>Caeonorhabditis elegans</i>. Proceedings of the National Academy of Sciences of the United States of America, 115(35):8823- 8828, 2018.  
+
+[30] Zhang D, de Souza RF, Anantharaman V, Iyer LM, and Aravind L. Polymorphic toxin systems: Comprehensive characterization of trafficking modes, processing, mechanisms of action, immunity and ecology using comparative genomics. Biology direct, 7:18, 2012.  
+
+[31] Nedialkova DD, Ulferts R, van den Born E, Lauber C, Gorbalenya AE, Ziebuhr J, and Snijder EJ. Biochemical characterization of arterivirus nonstructural protein 11 reveals the nidovirus- wide conservation of a replicative endoribonuclease. Journal of virology, 83(11):5671- 82, 2009.  
+
+[32] Ivanov KA, Hertzig T, Rozanov M, Bayer S, Thiel V, Gorbalenya AE, and Ziebuhr J. Major genetic marker of nidoviruses encodes a replicative endoribonuclease. Proceedings of the National Academy of Sciences of the United States of America, 101(34):12694- 9, 2004.  
+
+[33] Bhardwaj K, Guarino L, and Kao CC. The severe acute respiratory syndrome coronavirus Nsp15 protein is an endoribonuclease that prefers manganese as a cofactor. Journal of virology, 78(22):12218- 24, 2004.  
+
+[34] Renzi F, Caffarelli E, Laneve P, Bozzoni I, Brunori M, and Vallone B. The structure of the endoribonuclease XendoU: From small nucleolar RNA processing to severe acute respiratory syndrome coronavirus replication. Proceedings of the National Academy of Sciences of the United States of America, 103(33):12365- 70, 2006.
+
+<--- Page Split --->
+
+[35] T. Groves, P. Katis, Z. Madden, K. Manickam, D. Ramsden, G. Wu, and C. J. Guidos. In vitro maturation of clonal cd4+cd8+ cell lines in response to tcr engagement. Journal of immunology, 154(10):5011- 22, 1995.  
+
+[36] E.L. Sternburg, K. C. Dias, and F. V. Karginov. Selection- dependent and independent generation of crispr/cas9- mediated gene knockouts in mammalian cells. J. Vis. Exp., (124):e55903, 2017.  
+
+[37] Liu H and Naismith JH. An efficient one- step site- directed deletion, insertion, single and multiple- site plasmid mutagenesis protocol. BMC biotechnology, 8:91, 2008.  
+
+[38] Schanda P, Kupee E, and Brutscher B. SOFAST- HMQC experiments for recording two- dimensional deteronuclear correlation spectra of proteins within a few seconds. Journal of Biomolecular NMR, 33(4):199- 211, 2005.  
+
+[39] Favier A and Brutscher B. Recovering lost magnetization: polarization enhancement in biomolecular NMR. Journal of Biomolecular NMR, 49(1):9- 15, 2011.  
+
+[40] Lescop E, Kern T, and Brutscher B. Guidelines for the use of band- selective radiofrequency pulses in heteronuclear NMR: example of longitudinal- relaxation- enhanced BEST- type 1H- 15N correlation experiments. Journal of Magnetic Resonance, 203(1):190- 8, 2010.  
+
+[41] Keller R and Wuthrich K. Computer- aided resonance assignment (CARA). available at www.nmr.ch.  
+
+[42] Vranken WF, Boucher W, Stevens TJ, Fogh RH, Pajon A, Llinas M, Ulrich EL, Markley JL, Ionides J, and Laue ED. The CCPN data model for NMR spectroscopy: development of a software pipeline. Proteins, 59(4):687- 96, 2005.  
+
+[43] Williamson MP. Using chemical shift perturbation to characterise ligand binding. Progress in Nuclear Magnetic Resonance Spectroscopy, 73:1- 16, 2013.  
+
+[44] Kabsch W. XDS. Acta crystallographica. Section D, Biological crystallography, 66(Pt 2):125- 32, 2010.  
+
+[45] McCoy AJ, Grosse- Kunstleve RW, Adams PD, Winn MD, Storoni LC, and Read RJ. Phaser crystallographic software. Journal of applied crystallography, 40(Pt 4):658- 674, 2007.  
+
+[46] Liebschner D, Afonine PV, Baker ML, Bunkóczi G, Chen VB, Croll TI, Hintze B, Hung LW, Jain S, McCoy AJ, Moriarty NW, Oeffner RD, Poon BK, Prisant MG, Read RJ, Richardson JS, Richardson DC, Sammito MD, Sobolev OV, Stockwell DH, Terwilliger TC, Urzhumtsev AG, Videau LL, Williams CJ, and Adams PD. Macromolecular structure determination using X- rays, neutrons and electrons: recent developments in Phenix. Acta crystallographica. Section D, Structural biology, 75(Pt 10):861- 877, 2019.  
+
+[47] Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, Tunyasuvunakool K, Bates R, Žídek A, Potapenko A, Bridgland A, Meyer C, Kohl SAA, Ballard AJ, Cowie A, Romera- Paredes B, Nikolov S, Jain
+
+<--- Page Split --->
+
+R, Adler J, Back T, Petersen S, Reiman D, Clancy E, Zielinski M, Steinegger M, Pacholska M, Berghammer T, Bodenstein S, Silver D, Vinyals O, Senior AW, Kavukcuoglu K, Kohli P, and Hassabis D. Highly accurate protein structure prediction with AlphaFold. Nature, 596(7873):583–589, 2021.  
+
+[48] Bricogne G, Blanc E, Brandl M, Flensburg C, Keller P, Paciorek W, Roversi P, Sharff A, Smart OS, Vonrhein C, and Womack TO. BUSTER version 2.10.4. Cambridge, United Kingdom: Global Phasing Ltd, 2017.  
+
+[49] Lazar I and Lazar I. GelAnalyzer 23.1.1. available at www.gelanalyzer.com.  
+
+[50] Virtanen P, Gommers R, Oliphant TE, Haberland M, Reddy T, Cournapeau D, Burovski E, Peterson P, Weckesser W, Bright J, van der Walt SJ, Brett M, Wilson J, Millman KJ, Mayorov N, Nelson ARJ, Jones E, Kern R, Larson E, Carey CJ, Polat I, Feng Y, Moore EW, VanderPlas J, Laxalde D, Perktold J, Cimrman R, Henriksen I, Quintero EA, Harris CR, Archibald AM, Ribeiro AH, Pedregosa F, and van Mulbregt P. SciPy 1.0: fundamental algorithms for scientific computing in Python. Nature methods, 17(3):261–272, 2020.  
+
+[51] Van Der Spoel D, Lindahl E, Hess B, Groenhof G, Mark AE, and Berendsen HJ. GROMACS: fast, flexible, and free. Journal of computational chemistry, 26(16):1701–18, 2005.  
+
+[52] Jo S, Kim T, Iyer VG, and Im W. CHARMM-GUI: a web- based graphical user interface for CHARMM. Journal of computational chemistry, 29(11):1859–65, 2008.  
+
+[53] Lee J, Cheng X, Swails JM, Yeom MS, Eastman PK, Lemkul JA, Wei S, Buckner J, Jeong JC, Qi Y, Jo S, Pande VS, Case DA, Brooks CL, MacKerell AD, Klauda JB, and Im W. CHARMM-GUI Input Generator for NAMD, GROMACS, AMBER, OpenMM, and CHARMM/OpenMM Simulations Using the CHARMM36 Additive Force Field. Journal of chemical theory and computation, 12(1):405–13, 2016.  
+
+[54] Abramson J, Adler J, Dunger J, Evans R, Green T, Pritzel A, Ronneberger O, Willmore L, Ballard AJ, Bambrick J, Bodenstein SW, Evans DA, Hung CC, O'Neill M, Reiman D, Tunyasuvunakool K, Wu Z, Žemgulytė A, Arvaniti E, Beattie C, Bertolli O, Bridgland A, Cherepanov A, Congreve M, Cowen- Rivers AI, Cowie A, Figurnov M, Fuchs FB, Gladman H, Jain R, Khan YA, Low CMR, Perlin K, Potapenko A, Savy P, Singh S, Stecula A, Thillaisundaram A, Tong C, Yakneen S, Zhong ED, Zielinski M, Židek A, Bapst V, Kohli P, Jaderberg M, Hassabis D, and Jumper JM. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature, 2024.  
+
+[55] Huang J, Rauscher S, Nawrocki G, Ran T, Feig M, de Groot BL, Grubmüller H, and MacKerell AD. CHARMM36m: an improved force field for folded and intrinsically disordered proteins. Nature methods, 14(1):71–73, 2017.  
+
+[56] A Thureau, P Roblin, and J Perez. BioSAXS on the SWING beamline at Synchrotron SOLEIL. Journal of applied crystallography, 54(6):1698–1710, 2021.
+
+<--- Page Split --->
+
+[57] G David and J Pérez. Combined sampler robot and high- performance liquid chromatography: a fully automated system for biological small- angle X- ray scattering experiments at the Synchrotron SOLEIL SWING beamline. Journal of applied crystallography, 42:892- 900, 2009.  
+
+[58] J Pérez and Y Nishino. Advances in X- ray scattering: from solution SAXS to achievements with coherent beams. Current Opinion in Structural Biology, 22(5):670- 8, 2012.  
+
+[59] Hopkins JB, Gillilan RE, and Skou S. BioXTAS RAW: improvements to a free open- source program for small- angle X- ray scattering data reduction and analysis. Journal of applied crystallography, 50(5):1545- 1553, 2017.  
+
+[60] R P Rambo and J A Tainer. Accurate assessment of mass, models and resolution by small- angle scattering. Nature, 496(7446):477- 81, 2013.  
+
+[61] Grishaev A, Guo L, Irving T, and Bax A. Improved fitting of solution X- ray scattering data to macromolecular structures and structural ensembles by explicit water modeling. Journal of the American Chemical Society, 132(44):15484- 6, 2010.  
+
+[62] D Franke, M V Petoukhov, P V Konarev, A Panjkovich, A Tuukkanen, H D T Mertens, A G Kikhney, N R Hajizadeh, J M Franklin, C M Jeffries, and D I Svergun. ATSAS 2.8: a comprehensive data analysis suite for small- angle scattering from macromolecular solutions. Journal of Applied Crystallography, 50(4):1212- 1225, 2017.  
+
+[63] T. S. Heng and M. W. Painter. The immunological genome project: networks of gene expression in immune cells. Nature immunology, 9(10):1091- 4, 2008.  
+
+[64] M. Uhlen, L. Fagerberg, B. M. Hallstrom, C. Lindskog, P. Oksvold, A. Mardinoglu, A. Sivertsson, C. Kampf, E. Sjostedt, A. Asplund, I. Olsson, K. Edlund, E. Lundberg, S. Navani, C. A. Szigyarto, J. Odeberg, D. Djureinovic, J. O. Takanen, S. Hober, T. Alm, P. H. Edqvist, H. Berling, H. Tegel, J. Mulder, J. Rockberg, P. Nilsson, J. M. Schwenk, M. Hamsten, K. von Feilitzen, M. Forsberg, L. Persson, F. Johansson, M. Zwahlen, G. von Heijne, J. Nielsen, and F. Ponten. Proteomics. tissue- based map of the human proteome. Science, 347(6220):1260419, 2015.  
+
+[65] Laneve P, Gioia U, Ragno R, Altieri F, Di Franco C, Santini T, Arceci M, Bozzoni I, and Caffarelli E. The tumor marker human placental protein 11 is an endoribonuclease. The Journal of biological chemistry, 283(50):34712- 9, 2008.  
+
+[66] Waterhouse A, Bertoni M, Bienert S, Studer G, Tauriello G, Gumienny R, Heer FT, de Beer TAP, Rempfer C, Bordoli L, Lepore R, and Schwede T. SWISS- MODEL: homology modelling of protein structures and complexes. Nucleic acids research, 46(W1):W296- W303, 2018.
+
+<--- Page Split --->
+
+[67] Svergun D, Barberato C, and Koch MHJ. CRYSOL- a program to evaluate X- ray solution scattering of biological macromolecules from atomic coordinates. Journal of applied crystallography, 28:768- 773, 1995.  
+
+[68] C. M. Cuchillo, X. Pares, A. Guasch, T. Barman, F. Travers, and M. V. Nogues. The role of 2',3'- cyclic phosphodiesters in the bovine pancreatic ribonuclease a catalysed cleavage of rna: intermediates or products? FEBS Lett, 333(3):207- 10, 1993.  
+
+[69] W. Yang, J. Y. Lee, and M. Nowotny. Making and breaking nucleic acids: two- mg2+- ion catalysis and substrate specificity. Mol Cell, 22(1):5- 13, 2006.  
+
+[70] Qi W, Gromoff EDV, Xu F, Zhao Q, Yang W, Pfeifer D, Maier W, Long L, and Baumeister R. The secreted endoribonuclease ENDU- 2 from the soma protects germline immortality in C. elegans. Nature communications, 12(1):1262, 2021.  
+
+[71] Bowman AB and Aschner M. Considerations on manganese (Mn) treatments for in vitro studies. Neurotoxicology, 41:141- 2, 2014.  
+
+[72] Rehnberg GL, Hein JF, Carter SD, and Laskey JW. Chronic manganese oxide administration to preweanling rats: manganese accumulation and distribution. Journal of toxicology and environmental health, 6(1):217- 26, 1980.  
+
+[73] Keen CL, Ensunsa JL, and Clegg MS. Manganese metabolism in animals and humans including the toxicity of manganese. Metal ions in biological systems, 37:89- 121, 2000.  
+
+[74] D. Jenne and K. K. Stanley. Nucleotide sequence and organization of the human s- protein gene: repeating peptide motifs in the "pexin" family and a model for their evolution. Biochemistry, 26(21):6735- 42, 1987.  
+
+[75] A. Alquraini, S. Garguilo, G. D'Souza, L. X. Zhang, T. A. Schmidt, G. D. Jay, and K. A. Elsaid. The interaction of lubricin/proteoglycan 4 (prg4) with toll- like receptors 2 and 4: an anti- inflammatory role of prg4 in synovial fluid. Arthritis Res Ther, 17:353, 2015.  
+
+[76] J. W. Goding, B. Grobben, and H. Slegers. Physiological and pathophysiological functions of the ecto- nucleotide pyrophosphatase/phosphodiesterase family. Biochim Biophys Acta, 1638(1):1- 19, 2003.  
+
+[77] I. A. Droujinine, A. S. Meyer, D. Wang, N. D. Udeshi, Y. Hu, D. Rocco, J. A. McMahon, R. Yang, J. Guo, L. Mu, D. K. Carey, T. Svinkina, R. Zeng, T. Branon, A. Tabatabai, J. A. Bosch, J. M. Asara, A. Y. Ting, S. A. Carr, A. P. McMahon, and N. Perrimon. Proteomics of protein trafficking by in vivo tissue- specific labeling. Nat Commun, 12(1):2382, 2021.  
+
+[78] M. D. Bootman and G. Bultynck. Fundamentals of cellular calcium signaling: A primer. Cold Spring Harb Perspect Biol, 12(1), 2020.
+
+<--- Page Split --->
+
+[79] Pillon MC, Frazier MN, Dillard LB, Williams JG, Kocaman S, Krahn JM, Perera L, Hayne CK, Gordon J, Stewart ZD, Sobhany M, Deterding LJ, Hsu AL, Dandey VP, Borgnia MJ, and Stanley RE. Cryo- EM structures of the SARS- CoV- 2 endoribonuclease Nsp15 reveal insight into nuclease specificity and dynamics. Nature communications, 12(1):636, 2021.  
+
+[80] Kim Y, Wower J, Maltseva N, Chang C, Jedrzejczak R, Wilamowski M, Kang S, Nicolaescu V, Randall G, Michalska K, and Joachimiak A. Tipiracil binds to uridine site and inhibits Nsp15 endoribonuclease NendoU from SARS- CoV- 2. Communications biology, 4(1):193, 2021.  
+
+[81] Lapidoth G, Khersonsky O, Lipsh R, Dym O, Albeck S, Rogotner S, and Fleishman SJ. Highly active enzymes by automated combinatorial backbone assembly and sequence design. Nature communications, 9(1):2780, 2018.  
+
+[82] Lipsh- Sokolik R, Khersonsky O, Schröder SP, de Boer C, Hoch SY, Davies GJ, Overkleeft HS, and Fleishman SJ. Combinatorial assembly and design of enzymes. Science (New York, N.Y.), 379(6628):195- 201, 2023.  
+
+[83] Rapp JT, Bremer BJ, and Romero PA. Self- driving laboratories to autonomously navigate the protein fitness landscape. Nature chemical engineering, 1:97- 107, 2024.  
+
+[84] Luscombe NM, Laskowski RA, and Thornton JM. NUCPLOT: a program to generate schematic diagrams of protein- nucleic acid interactions. Nucleic acids research, 25(24):4940- 5, 1997.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 1: Mouse EndoU is expressed during thymocyte development and has a calcium-activated RNase activity. (A) EndoU expression levels in developing thymocyte populations. The color indicates fractional expression of the mRNA across 211 measured cell types. Data from the Immunological Genome Project [63]. (B) RNase activity in WT and EndoU KO cell lysates. Cytoplasmic lysates from the indicated WT, KO or rescue cell lines were incubated +/- 5 mM \(\mathrm{Ca^{2 + }}\) for 15 minutes at \(37^{\circ}\mathrm{C}\) . RNA was extracted, 5 μg were run on an 8 % urea-PAGE gel, and visualized by SYBR Green II. WT-HA denotes a WT EndoU rescue construct with a C-terminal HA tag. (C) Immunoprecipitated EndoU cleavage activity on a defined RNA substrate. (D) Mutation of presumed catalytic site residues abolishes EndoU cleavage activity. (E) EndoU RNase activity is specifically activated by \(\mathrm{Ca^{2 + }}\) ions. (F) Domain structure of EndoU and homologs. </center>
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 2: Crystal structure of calcium-activated EndoU. (A) Overview of calcium-activated EndoU structure. Calcium binding site (1) to (4) as refered in the text are indicated. (B) Close-up view of EndoU calcium binding sites. Residues coordinating calcium directly (pink) or indirectly through water-mediated contacts (cyan) are highlighted. (C) Intramolecular signaling between calcium binding site (1) and remote catalytic residues. Catalytic residue (yellow) are highlighted, along with residues coordinating calcium directly (pink) or indirectly (cyan). (D) Structural change upon EndoU allosteric activation by calcium. </center>
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 3: Enzymatic activity of EndoU and its variants. (A) RNA degradation assays. Comparison of mutants for calcium-binding sites (magenta), the bridging residue E290 (cyan), and catalytic residues (yellow) with wild-type EndoU over a 2 hrs degradation assay. (B) Enzymatic progress curve. Example of fit for a first-order reaction model \(A \times \exp (-k \times t)\) with wild-type EndoU. (C) Relative reaction rates of EndoU variants compared to wild-type. (D) Calcium binding to EndoU monitored through Molecular Dynamics. Each plot displays the distance between the E284 side-chain carboxylate and a calcium ion throughout the simulation. Five calcium ions were introduced in the simulation box. </center>
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Figure 4: NMR mapping of the RNA binding interface on calcium-activated EndoU. Overlay of (A) \(^{15}\mathrm{N}\) or (B) \(^{13}\mathrm{C}\) SOFAST-HMQC spectra from isolated \(^{13}\mathrm{C}^{15}\mathrm{N}\) -EndoU (350 \(\mu \mathrm{M}\) , green) and upon successive additions of 2' fluorinated RNA. (C) Combined \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) chemical shift perturbations between calcium-activated \(^{13}\mathrm{C}^{15}\mathrm{N}\) -EndoU and in complex with 2'F RNA. (D) RNA binding interface mapping on calcium-activated EndoU. (E) Electrostatic surface potential of calcium-activated EndoU. </center>
+
+<--- Page Split --->
+![PLACEHOLDER_40_0]
+
+<center>Figure 5: Schematic representation of eukaryotic EndoU activation upon calcium and substrate binding. </center>
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- Fig3noncroppedgels.pdf- Malardetal2024SI.pdf
+
+<--- Page Split --->

preprint/preprint__7ea7ef9b459a9c05ffe0bbce2c11c10d905560d4f8d0e1620fe90f75c4f3407c/preprint__7ea7ef9b459a9c05ffe0bbce2c11c10d905560d4f8d0e1620fe90f75c4f3407c_det.mmd

@@ -0,0 +1,649 @@
+<|ref|>title<|/ref|><|det|>[[44, 107, 802, 175]]<|/det|>
+# Molecular Basis for the Calcium-Dependent Activation of the Ribonuclease EndoU  
+
+<|ref|>text<|/ref|><|det|>[[44, 196, 192, 235]]<|/det|>
+Florian Malard  INSERM U1212  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 340, 283]]<|/det|>
+Kristen Dias  University of California Riverside  
+
+<|ref|>text<|/ref|><|det|>[[44, 290, 290, 330]]<|/det|>
+Margaux Baudy  INSERM U1212, ARNA unit  
+
+<|ref|>text<|/ref|><|det|>[[44, 337, 290, 376]]<|/det|>
+Stéphane Thore  INSERM U1212, ARNA unit  
+
+<|ref|>text<|/ref|><|det|>[[44, 383, 255, 423]]<|/det|>
+Brune Vialet  University of Bordeaux  
+
+<|ref|>text<|/ref|><|det|>[[44, 429, 400, 469]]<|/det|>
+Philippe Barthélémy  https://orcid.org/0000- 0003- 3917- 0579  
+
+<|ref|>text<|/ref|><|det|>[[44, 475, 560, 515]]<|/det|>
+Sébastien Fribourg  Univ. de Bordeaux, Institut Européen de Chimie et Biologie  
+
+<|ref|>text<|/ref|><|det|>[[44, 521, 340, 561]]<|/det|>
+Fedor Karginov  University of California Riverside  
+
+<|ref|>text<|/ref|><|det|>[[44, 567, 233, 585]]<|/det|>
+Sebastien Campagne  
+
+<|ref|>text<|/ref|><|det|>[[52, 594, 366, 611]]<|/det|>
+sebastien.campagne@inserm.fr  
+
+<|ref|>text<|/ref|><|det|>[[52, 640, 648, 658]]<|/det|>
+INSERM U1212, ARNA unit  https://orcid.org/0000- 0002- 0094- 4760  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 700, 103, 718]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 737, 530, 757]]<|/det|>
+Keywords: EndoU, calcium, RNA, allostery, ribonuclease.  
+
+<|ref|>text<|/ref|><|det|>[[44, 776, 295, 795]]<|/det|>
+Posted Date: July 15th, 2024  
+
+<|ref|>text<|/ref|><|det|>[[44, 814, 475, 833]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 4654759/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 852, 914, 893]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 911, 535, 931]]<|/det|>
+Additional Declarations: There is NO Competing Interest.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 901, 88]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on April 1st, 2025. See the published version at https://doi.org/10.1038/s41467-025-58462-6.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[120, 163, 874, 193]]<|/det|>
+# Molecular Basis for the Calcium-Dependent Activation  
+
+<|ref|>title<|/ref|><|det|>[[309, 216, 686, 245]]<|/det|>
+# of the Ribonuclease EndoU  
+
+<|ref|>text<|/ref|><|det|>[[128, 280, 872, 380]]<|/det|>
+Florian Malard \(^{1,2}\) , Kristen Dias \(^{3}\) , Margaux Baudy \(^{1,2}\) , Stéphane Thore \(^{1}\) , Brune Vialet \(^{1}\) , Philippe Barthélémy \(^{1}\) , Sébastien Fribourg \(^{1,*}\) , Fedor V Karginov \(^{3,*}\) , and Sébastien Campagne \(^{1,2,*}\)  
+
+<|ref|>text<|/ref|><|det|>[[148, 408, 850, 432]]<|/det|>
+\(^{1}\) Univ. Bordeaux, CNRS, INSERM, ARNA, UMR 5320, U1212, F- 33000  
+
+<|ref|>text<|/ref|><|det|>[[413, 448, 584, 469]]<|/det|>
+Bordeaux, France  
+
+<|ref|>text<|/ref|><|det|>[[130, 483, 866, 508]]<|/det|>
+\(^{2}\) Univ. Bordeaux, CNRS, INSERM, IECB, US1, UAR 3033, F- 33600 Pessac,  
+
+<|ref|>text<|/ref|><|det|>[[465, 523, 533, 544]]<|/det|>
+France  
+
+<|ref|>text<|/ref|><|det|>[[128, 558, 868, 620]]<|/det|>
+\(^{3}\) Department of Molecular, Cell and Systems Biology, Institute for Integrative Genome Biology, University of California at Riverside, Riverside, CA, 92521,  
+
+<|ref|>text<|/ref|><|det|>[[473, 635, 523, 655]]<|/det|>
+USA  
+
+<|ref|>text<|/ref|><|det|>[[163, 671, 835, 732]]<|/det|>
+\* Correspondence should be addressed to sebastien.fribourg@inserm.fr, fedor.karginov@ucr.edu and sebastien.campagne@inserm.fr
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 103, 222, 127]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[112, 150, 885, 597]]<|/det|>
+Ribonucleases (RNases) are ubiquitous enzymes that process or degrade RNA, essential for cellular functions and immune responses. The EndoU- like superfamily includes endoribonucleases conserved across bacteria, eukaryotes, and certain viruses, with an ancient evolutionary link to the ribonuclease A- like superfamily. Both bacterial EndoU and animal RNase A share a similar fold and function independently of cofactors. In contrast, the eukaryotic EndoU catalytic domain requires divalent metal ions for catalysis, possibly due to an N- terminal extension near the catalytic core. In this study, we used biophysical and computational techniques along with in vitro assays to investigate the calcium- dependent activation of human EndoU. We determined the crystal structure of EndoU bound to calcium and found that calcium binding remote from the catalytic triad triggers water- mediated intramolecular signaling and structural changes, activating the enzyme through allostery. Calcium- binding involves residues from both the catalytic core and the N- terminal extension, indicating that the N- terminal extension interacts with the catalytic core to modulate activity in response to calcium. Our findings suggest that similar mechanisms may be present across all eukaryotic EndoUs, highlighting a unique evolutionary adaptation that connects endoribonuclease activity to cellular signaling in eukaryotes.  
+
+<|ref|>text<|/ref|><|det|>[[114, 614, 580, 634]]<|/det|>
+Keywords: EndoU, calcium, RNA, allostery, ribonuclease.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[112, 85, 886, 250]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[113, 271, 883, 326]]<|/det|>
+Graphical abstract: Calcium binds at the interface between the catalytic core and N- terminal extension in eukaryotic EndoU catalytic domains, activating the catalytic site at a distance via allostery.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 103, 270, 127]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[112, 150, 885, 597]]<|/det|>
+Ribonucleases (RNases) are nucleases that catalyze the processing or degradation of RNA. Found in all organisms, RNases play vital roles in various cellular processes, including maturing both coding and non- coding RNAs, combating RNA viruses, and contributing to sophisticated immune strategies like RNA interference [1, 2, 3]. For example, RNases catalyze mRNA decay in general pathways (XRN1, exosome/DIS3L) or as part of apoptotic cascades (RNase L, DIS3L2), carry out unconventional splicing or tRNA cleavage during stress (IRE1, angiogenin), or catabolize extracellular RNAs (RNase A). Among RNases, the cellular roles of those that cleave endonucleolytically have been increasingly recognized [4]. RNases can be constitutively active (RNase A, angiogenin), or stimulated by ligand binding (RNase L) or cellular signaling events, such as phosphorylation (IRE1). The ribonuclease A- like domain superfamily (IPR036816 [5]) is the most well- known RNase domain, with many pioneering studies in the \(20^{\mathrm{th}}\) century [6, 7]: it was the first directly sequenced enzyme [8], the first enzyme for which a catalytic mechanism was proposed based on experimental data [9], and one of the first solved three- dimensional structures [10]. Despite its significant impact in enzyme research, it is important to note that the RNase- A- like domain is only found in vertebrates, raising questions about its deeper evolutionary ancestors or relatives [6].  
+
+<|ref|>text<|/ref|><|det|>[[112, 612, 886, 907]]<|/det|>
+The endoribonuclease EndoU- like (Endoribonuclease specific for Uridylate) superfamily (IPR037227 [5]) is a poorly understood group of RNases found in bacteria, eukaryotes and viruses. Notably, a structural similarity between a bacterial EndoU- like toxin and vertebrate RNase A was identified [11]. Furthermore, recent studies uncovered an ancient evolutionary link between the Ribonuclease A and EndoU families, suggesting that the animal RNase A gene could have evolved either through significant alteration of an EndoU gene, or by horizontal acquisition of a prokaryotic ribonuclease [6]. XendoU, the founding member of the EndoU- like superfamily (IPR037227 [5]), was initially identified in Xenopus laevis oocyte extracts as an enzyme that releases small nucleolar RNAs from introns [12, 13, 14]. In vitro studies demonstrated that XendoU is an endonuclease that cleaves single- stranded RNA preferentially at 5' of uridylates [15]. In eukaryotes, XendoU defines
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 87, 885, 412]]<|/det|>
+a distinct EndoU family (IPR018998 [5], PF09412 [16]) that lacks sequence homology with other known RNases, and is broadly conserved across \*Arabidopsis thaliana\*, \*Drosophila melanogaster\*, \*Mus musculus\*, \*Homo sapiens\*, and other species [14, 15]. Human EndoU (hEndoU) was first identified as human placental protein 11 (PP11) due to its prevalence in the placenta [17, 18]. It is also now recognized as a biomarker in various cancers, including squamous cell carcinomas, ovarian adenocarcinoma, non- trophoblastic tumors and breast cancers [19, 20, 21, 22, 23, 24]. In human cells, EndoU has been proposed to be involved in RNA cleavage, ribonucleoprotein particle removal, and endoplasmic reticulum network organization [25, 26]. Across other eukaryotes, EndoU has been implicated in pro- apoptotic processes in mouse B cells, neuron survival in fruit flies, and synaptic remodeling in nematodes [27, 28, 29]. The more distant bacterial EndoU- like ribonucleases are common in microbial warfare as toxins [30].  
+
+<|ref|>text<|/ref|><|det|>[[112, 428, 885, 844]]<|/det|>
+Members of the EndoU- like superfamily (IPR037227 [5]) exhibit notable differences in their activation requirements. For instance, it is well characterized that EndoU- like bacterial toxins and arterial Nsp11 do not need any cofactors for activation, analogous to vertebrate RNase A [11, 31]. In contrast, studies have shown that purified forms of XendoU and coronaviral Nsp15 require millimolar concentrations of \(\mathrm{Ca}^{2 + }\) or \(\mathrm{Mn}^{2 + }\) [15, 25, 32, 33]. The crystal structure of the endoribonuclease XendoU in the absence of divalent metals has been solved [34], suggesting a catalytic site arrangement similar to that of vertebrate RNase A, specifically featuring a catalytic His- His- Lys triad [34]. However, the structural basis for the metal- dependent activation of eukaryotic EndoUs could not be explained by the crystal structure of XendoU, which represents the inactive state of the endonuclease in the absence of a cofactor [34]. Bacterial and metal- independent viral EndoUs share a smaller, C- terminal catalytic domain compared to eukaryotic EndoUs. Because eukaryotic EndoUs contain an N- terminal extension within this catalytic domain that correlates with \(\mathrm{Ca}^{2 + }\) dependence, we hypothesized that it may bind calcium and control the activity of the catalytic core through allostery.  
+
+<|ref|>text<|/ref|><|det|>[[113, 862, 881, 882]]<|/det|>
+In this study, we elucidated the molecular mechanism of EndoU activation by calcium. First, we es
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 352]]<|/det|>
+tablished a thymocyte cell line model to confirm the dependence of EndoU for calcium in both cell extract and recombinant forms. Next, we used biophysical methods to detect an allosteric change upon activation by calcium and to solve the structure of active EndoU. Our structural analysis revealed a calcium- stabilized interaction network involving residues from both the eukaryotic- specific N- terminal extension and the catalytic core of EndoU, ultimately leading to the activation of the catalytic triad. Our findings provide unprecedented atomic- level insights into a metal ion- activated member of the EndoU- like superfamily (IPR037227 [5]), addressing a longstanding question in the study of eukaryotic EndoUs, which are of significant interest due to their switchable endonuclease activity.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 102, 398, 127]]<|/det|>
+## Materials and Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 162, 236, 183]]<|/det|>
+## Cell culture  
+
+<|ref|>text<|/ref|><|det|>[[113, 202, 884, 346]]<|/det|>
+VL3- 3M2 mouse thymic lymphoma cells [35] were cultured in RPMI 1640 (Corning) supplemented with \(10\mathrm{mM}\) HEPES, \(50\mu \mathrm{L}\beta\) - mercaptoethanol, \(1\mathrm{x}\) penicillin/streptomycin, and \(10\%\) fetal bovine serum (FBS). The Platinum- E (Plat- E) retroviral packaging cell line was cultured in DMEM (Corning) supplemented with \(10\%\) FBS (Corning) and \(10\mathrm{units.ml^{- 1}}\) of penicillin/streptomycin (Gibco). All cells were grown at \(37^{\circ}\mathrm{C}\) in an atmosphere containing \(5\%\) \(\mathrm{CO_2}\)  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 381, 376, 403]]<|/det|>
+## VL3-3M2 TCR activation  
+
+<|ref|>text<|/ref|><|det|>[[113, 422, 885, 595]]<|/det|>
+Cell culture 6- well plates were pre- incubated overnight at \(37^{\circ}\mathrm{C}\) with \(1\mathrm{mL}\) of PBS, either with or without \(5\mu \mathrm{g.mL^{- 1}}\) of anti- CD3e/CD28 or anti- CD3/CD4 antibodies. The PBS was then aspirated, and \(5*10^{5}\) cells in \(2\mathrm{mL}\) of media were added. For PMA/ionomycin stimulation, concentrations of \(20\mathrm{ng.mL^{- 1}}\) and \(500\mathrm{ng.mL^{- 1}}\) were used, respectively. Total RNA was extracted using Trizol 24 hours later, and RT- qPCR measurements were conducted for EndoU, Rag1, and CD5, normalized against a \(\beta\) - actin control. Fold changes were calculated relative to an unstimulated control.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 631, 440, 654]]<|/det|>
+## EndoU knockout cell generation  
+
+<|ref|>text<|/ref|><|det|>[[113, 672, 885, 907]]<|/det|>
+EndoU KO VL3- 3M2 cells were generated as described [36]. sgRNAs designed to target intron 1 and exon 11 of the EndoU locus (Table S1) were cloned into the pSpCas9(BB)/pX330 Cas9- sgRNA expression plasmid (Addgene #42230). A neomycin resistance cassette flanked by two 900 bp homology regions to intron 1 and exon 11 were assembled into the pUC- 19 vector as previously described [36]. The Cas9- sgRNA expression plasmids and the homology arm vector were electroporated into \(10^{7}\) VL3- 3M2 cells at 340 V for 47 ms in Opti- MEM (Gibco). Neomycin selection was applied after two days. Clonal cells were subsequently generated and screened via PCR using genomic DNA as the template. This involved primers (Table S1) to detect genomic DNA (positive
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 883, 170]]<|/det|>
+control for WT and KO, gDNA F/R, 800 bp amplicon), primers to verify the presence of the WT allele (EndoU validation F/R, 1079 bp amplicon), and primers to identify the KO allele (EndoU validation F/Resistance R, 999 bp amplicon).  
+
+<|ref|>sub_title<|/ref|><|det|>[[114, 208, 563, 230]]<|/det|>
+## Mouse EndoU tagged and mutant constructs  
+
+<|ref|>text<|/ref|><|det|>[[113, 249, 884, 392]]<|/det|>
+The mouse EndoU cDNA (NM_001168693) was PCR amplified from VL3- 3M2 cDNA with primers containing XhoI (forward) and BglII (reverse) restriction sites and ligated into the pMSCV- PIG or pRL- TK vectors. The Q5 site- directed mutagenesis kit (NEB Cat. E0554S) was used to add a C- terminal FLAG- HA tag, or to create the E285A;H286A catalytically dead mutant version, in pMSCV- PIG.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 429, 789, 452]]<|/det|>
+## Viral production and stable integration of EndoU rescue constructs  
+
+<|ref|>text<|/ref|><|det|>[[114, 470, 884, 551]]<|/det|>
+VL3- 3M2 clonal EndoU knockout cells were rescued through viral integration of the above EndoU constructs. Plat- E cells were calcium- phosphate transfected with \(10 \mu \mathrm{g}\) of pMSCV- PIG and \(2.5 \mu \mathrm{g}\) VSVG to produce amphitropic VSVG- pseudotyped retrovirus.  
+
+<|ref|>sub_title<|/ref|><|det|>[[114, 588, 216, 610]]<|/det|>
+## RT-qPCR  
+
+<|ref|>text<|/ref|><|det|>[[114, 630, 883, 711]]<|/det|>
+RNA was extracted from whole cells using ribozol followed by two phenol chloroform (pH 5.2) extractions. Superscript II reverse transcriptase was used for cDNA synthesis with \(1 \mu \mathrm{g}\) of total RNA as template. TaqMan probes against EndoU (Cat. 4351372) were used in the RT- qPCR.  
+
+<|ref|>sub_title<|/ref|><|det|>[[114, 748, 207, 769]]<|/det|>
+## Cell lysis  
+
+<|ref|>text<|/ref|><|det|>[[113, 789, 884, 901]]<|/det|>
+Cell lysis was carried out by first washing the cells once with PBS buffer and then resuspending them in hypotonic lysis buffer (10 mM Tris- HCl pH 7.5, 10 mM KCl, 5 mM DTT, protease inhibitor). The cells were subsequently incubated on ice for 20 minutes. Isotonicity was restored by adjusting the KCl concentration to \(100 \mathrm{mM}\) using a \(5 \mathrm{X}\) supplemental buffer (450 mM KCl,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 882, 140]]<|/det|>
+0.08 U. \(\mu \mathrm{l}^{- 1}\) RNaseIN). In certain experiments, lysates were centrifuged at \(17 000 \mathrm{g}\) for 20 minutes to separate the cytoplasmic fraction and collect the supernatant.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 178, 342, 200]]<|/det|>
+## Immunoprecipitations  
+
+<|ref|>text<|/ref|><|det|>[[113, 219, 883, 361]]<|/det|>
+Immunoprecipitations were carried out using protein A Dynabeads. Beads were prepared by incubation with \(16.7 \mu \mathrm{g}.\mathrm{ml}^{- 1}\) anti- mouse \(\mathrm{Fc}\gamma\) bridging antibody and \(16.7 \mu \mathrm{g}.\mathrm{ml}^{- 1}\) mouse anti- HA.11 antibody, sequentially. Cell lysates were incubated with prepared beads for 1 hour at room temperature. To equalize the amount of EndoU across reactions, an excess of cell lysate over bead capacity was used, and saturation of EndoU binding was verified by western blot.  
+
+<|ref|>sub_title<|/ref|><|det|>[[114, 398, 490, 420]]<|/det|>
+## On-bead mouse EndoU RNase assays  
+
+<|ref|>text<|/ref|><|det|>[[113, 438, 884, 641]]<|/det|>
+In a total volume of \(10 \mu \mathrm{L}\) , reactions consisted of (unless used as a variable) \(2 \mathrm{mM}\) calcium, \(100 \mathrm{mM}\) Tris- HCl (pH 7.5), \(10 \mathrm{mM}\) NaCl, \(5 \mu \mathrm{g}\) of total cytoplasmic RNA, or \(1 \mu \mathrm{M}\) of specific RNA oligo (Table S2, typically 50 mer 1), and the immunoprecipitated EndoU. Reactions were incubated at \(37^{\circ} \mathrm{C}\) , RNA was extracted and run on an \(8 \%\) urea- PAGE gel, and visualized by SYBR Green II. Densitometry was used to quantify substrate degradation using Quantity One (BioRad). Experiments were done in triplicate from distinct samples, with central tendencies expressed as means and variations as standard deviation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 679, 407, 700]]<|/det|>
+## Production of human EndoU  
+
+<|ref|>text<|/ref|><|det|>[[113, 720, 884, 893]]<|/det|>
+The open reading frame (ORF) encoding the catalytic domain of human EndoU (135- 410) was sub- cloned into the pET24b(+) plasmid \((\mathrm{Kan}^{\mathrm{R}})\) downstream of the GB1 protein ORF followed by a hexahistidine tag and a TEV protease cleavage site. Expression of EndoU was achieved in Escherichia coli BL21 Rosetta (DE3) pLysS. The bacteria were grown in rich LB medium or in M9 minimal medium supplemented with \(^{15}\mathrm{N}\) - labeled \(\mathrm{NH_4Cl}\) (1 g. \(\mathrm{L}^{- 1}\) ) and \(^{13}\mathrm{C}\) - labeled glucose (2 g. \(\mathrm{L}^{- 1}\) ) to achieve uniform isotope labeling. The cultures were grown at \(37^{\circ} \mathrm{C}\) until reaching an \(\mathrm{OD}_{600}\) of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 80, 886, 719]]<|/det|>
+approximately 0.6. Subsequently, protein expression was induced using \(0.25 \mathrm{mM}\) IPTG at \(15^{\circ} \mathrm{C}\) over 16 hours. The bacteria were harvested by centrifugation (5000 g, 10 min, \(4^{\circ} \mathrm{C}\) ), and the resulting pellets were resuspended in ice- cold lysis buffer (20 mM Tris pH 8, 500 mM NaCl, 250 \(\mu \mathrm{L} \cdot \mathrm{L}^{- 1} \beta\) - mercaptoethanol). This buffer was further supplemented with \(1 \mathrm{mg} \cdot \mathrm{mL}^{- 1}\) lysozyme and \(10 \mu \mathrm{L} \cdot \mathrm{L}^{- 1} \mathrm{DNase}\) (NEB). Cell lysis was achieved by sonication, running three cycles of 5 minutes each at \(20 \%\) amplitude, with 20- second on/off intervals. The lysate was clarified by centrifugation (20000 g, 30 min, \(4^{\circ} \mathrm{C}\) ) and the supernatant was loaded onto a gravity- flow histidine affinity chromatography column equilibrated with loading buffer (20 mM Tris pH 8, 500 mM NaCl, 250 \(\mu \mathrm{L} \cdot \mathrm{L}^{- 1} \beta\) - mercaptoethanol). The column was washed with \(15 \mathrm{mM}\) imidazole (10 CV), and the protein was eluted with \(300 \mathrm{mM}\) imidazole (5 CV). The eluted protein was then dialyzed against TEV digestion buffer (10 mM Tris pH 8, 250 mM NaCl, \(125 \mu \mathrm{L} \cdot \mathrm{L}^{- 1} \beta\) - mercaptoethanol) over 16 hours, in the presence of His \(_{6}\) - TEV protease (1:10 w/w ratio) to digest the GB1- His \(_{6}\) tag. Post- digestion, EndoU was isolated from the flow- through fraction following its loading onto a gravity- flow histidine affinity chromatography column, and washing with the loading buffer (5 CV). The resulting protein was concentrated, and a large excess of EDTA (250 mM) was added to chelate potential divalent cations. Further purification was achieved using a Superdex 75 column pre- equilibrated with storage buffer (10 mM Tris pH 7, 50 mM NaCl, \(1 \mathrm{mM}\) TCEP). Finally, EndoU was concentrated to a concentration of \(500 \mu \mathrm{M}\) . It was used immediately for enzymatic assays, while it was stored at \(- 80^{\circ} \mathrm{C}\) for other experiments. Point mutants were generated using the QuickChange protocol [37] and EndoU variants were purified using the same protocol as the wild type protein. The sequences of the oligonucleotides are given (Table S3).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 751, 380, 773]]<|/det|>
+## Oligonucleotides synthesis  
+
+<|ref|>text<|/ref|><|det|>[[114, 792, 884, 905]]<|/det|>
+Oligonucleotides were synthesized using the \(\beta\) - phosphoramidite method with an H8 automated synthesizer (K&A Labs, Germany) on a micromolar scale. For the synthesis of \(2^{\circ} \mathrm{F}\) RNA analogs, sequences started with a Unylinker solid support (Glen Research), and nucleotides were added sequentially using \(2^{\circ} \mathrm{F}\) phosphoramidites. For the synthesis of \(3^{\circ}\) labeled Cyanine 5 RNA, the dye was
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 291]]<|/det|>
+directly attached to the support, and RNA monomers were used. All phosphoramidites and the Cyanine 5 solid support were purchased from LINK (Scotland). Deprotection of the oligonucleotides was performed according to the suppliers protocols. The concentrated crude oligonucleotides were then resuspended in water. The sample concentration was determined from the absorbance at 260 nm and the molar extinction coefficient of the oligonucleotide. This value was calculated using the Integrated DNA Technology online oligo analyzer tool, which uses the standard nearest neighbor method.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 328, 410, 350]]<|/det|>
+## Nuclear Magnetic Resonance  
+
+<|ref|>text<|/ref|><|det|>[[112, 368, 886, 844]]<|/det|>
+Nuclear Magnetic Resonance (NMR) spectroscopy was used to analyze protein structure and dynamics. Experiments were performed using either a Bruker AVIll NMR spectrometer at 700 MHz with a room- temperature probe, or a Bruker Avance NEO spectrometer at 800 MHz with a cryogenic \(5\mathrm{mm}\) TCI \(^{1}\mathrm{H - }^{13}\mathrm{C / }^{15}\mathrm{N / }^{2}\mathrm{H}\) Z- gradient probe. These experiments were carried out at \(35^{\circ}\mathrm{C}\) in a minimal buffer composed of \(10\mathrm{mM}\) Tris (pH 7), \(50\mathrm{mM}\) NaCl, \(1\mathrm{mM}\) TCEP, and \(10\%\) \(\mathrm{D}_2\mathrm{O}\) for field frequency lock. We acquired 2D \(^{1}\mathrm{H - }^{15}\mathrm{N}\) and \(^{1}\mathrm{H - }^{13}\mathrm{C}\) correlation spectra using the SOFAST- HMQC experiment scheme [38]. Sequence- specific backbone assignments of \(^{15}\mathrm{N}^{13}\mathrm{C}\) - labeled calcium- activated EndoU were achieved via classical 3D triple resonance experiments based on the BEST- TROSY principle [39, 40]. The same approach was applied to EndoU bound to RNA targets. Spectra processing was conducted with Topspin 4 (Bruker) and analyzed using CARA [41] and CCPNMR software 2.4 [42]. Combined \(^{1}\mathrm{H - }^{15}\mathrm{N}\) chemical shift perturbations \((\Delta \delta_{\mathrm{comb}})\) were calculated as \(\Delta \delta_{comb} = \sqrt{\Delta\delta^{1}H + 0.14} * \Delta \delta^{15}N\) , where \(\Delta \delta^{1}H\) and \(\Delta \delta^{15}N\) are the chemical shift perturbations (in ppm) for \(^{1}\mathrm{H}\) and \(^{15}\mathrm{N}\) resonances, respectively [43]. NMR titrations to map the RNA binding surface on calcium- bound EndoU were performed using a non- cleavable, \(2^{\prime}\) - fluorinated RNA obtained in house via solid- phase synthesis with the following sequence: \(5^{\prime}\) - AAGUCC- 3'.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 100, 365, 121]]<|/det|>
+## Structure Determination  
+
+<|ref|>text<|/ref|><|det|>[[113, 140, 885, 494]]<|/det|>
+Structure DeterminationA sample of the catalytic domain of human EndoU, spanning residues 135 to 410, was prepared at a concentration of \(12 \mathrm{mg}.\mathrm{mL}^{- 1}\) in a buffer containing \(10 \mathrm{mM}\) Tris pH 7, \(50 \mathrm{mM}\) NaCl, \(1 \mathrm{mM}\) TCEP, and \(20 \mathrm{mM}\) \(\mathrm{CaCl}_2\) . The crystallization of EndoU was carried out at \(20^{\circ}\mathrm{C}\) using the MCSG4 matrix screen, specifically condition F6, which comprises \(0.1 \mathrm{M}\) sodium acetate, \(0.1 \mathrm{M}\) HEPES pH 7.5, and \(22 \%\) PEG 4k. The resulting crystals were flash- frozen in liquid nitrogen using a cryoprotectant solution identical to the crystallization condition but supplemented with \(20 \%\) ethylene glycol. Diffraction data were collected at the SOLEIL synchrotron on the PX1 beamline and processed using XDS [44]. Molecular replacement was conducted with Phaser [45] from the Phenix suite [46], using the AlphaFold 2 [47] predicted structure of the human EndoU protein as the model. This process identified two molecules per asymmetric unit, which were subsequently refined using Phenix and BUSTER [48]. Detailed crystallographic data and refinement statistics are presented in the supplementary materials (Table S4).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 530, 485, 552]]<|/det|>
+## Enzymatic RNA Degradation Assays  
+
+<|ref|>text<|/ref|><|det|>[[113, 571, 885, 896]]<|/det|>
+Enzymatic RNA Degradation AssaysRNA degradation assays on the human catalytic domain were carried out to assess the relative activity of EndoU and its variants. An RNA sequence, 5'- CAGGUUUCCCCAACGAAAAAAAAAA- 3', was obtained in- house via solid- phase synthesis. The RNA was labeled at the 3' end with a Cyanine- 5 (Cy5) fluorescent probe for detection purposes. In each assay, EndoU or one of its variants was prepared at a final concentration of \(1 \mathrm{nM}\) in presence of \(1 \mu \mathrm{M}\) of the RNA. The enzymatic reaction was initiated by introducing \(2 \mathrm{mM}\) \(\mathrm{CaCl}_2\) into the mixture. Samples were collected at 15 time points: 0, 1, 3, 5, 10, 15, 20, 25, 30, 40, 50, 60, 80, 100, and 120 minutes. The reaction was terminated at each time point with an excess of EDTA to chelate calcium ions in order to prevent EndoU activation and further RNA degradation. RNA degradation was monitored by resolving the samples on a polyacrylamide gel containing \(6 \mathrm{M}\) urea, followed by electrophoresis at \(250 \mathrm{V}\) for 50 minutes. The gel was then scanned with a fluorescence scanner. We used the GelAnalyzer
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 352]]<|/det|>
+software [49] to integrate band intensities, which were normalized relative to the zero time point. Each assay was conducted in triplicate to ensure the reproducibility of the results. Measurements were taken from distinct samples for each replicate; central tendencies are expressed as means, and variations as standard deviations. Data were processed and analyzed using custom Python scripts. A first- order reaction model, \(A * exp(- k * t)\) , was used to fit the enzymatic progress curve, using the curve_fit function from the scipy.optimize module for regression [50]. The fitted reaction rate \(k\) characterizes the activity of each EndoU variant. To enable comparison across different variants, this reaction rate was subsequently expressed in relative terms with respect to the wild- type EndoU, yielding a dimensionless parameter.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 389, 325, 410]]<|/det|>
+## Molecular Dynamics  
+
+<|ref|>text<|/ref|><|det|>[[112, 427, 885, 907]]<|/det|>
+Molecular dynamics simulations were performed using the GROMACS software package (version 2022.1) [51]. System preparation was achieved through CHARMM- GUI and the Input Generator module [52, 53], which was also used to apply single amino- acid substitutions for EndoU variants. To monitor the stability of the calcium binding sites, we used the crystal structure of EndoU bound to calcium ions as input. To monitor the binding of calcium to apo- EndoU, we removed calcium ions from the crystal structure and used the resulting structure as input. Calcium ions were then reintroduced into the system as salt ions. To propose an ensemble of models of calcium- activated EndoU in complex with RNA, we used the AlphaFold (AF) 3 [54] webserver with the human EndoU sequence (135- 410), a \((\mathrm{U})_6\) RNA, and four calcium ions as inputs. The top- ranked model accurately reproduced each of the calcium binding sites, with an RMSD of 0.313 Å between the crystal structure of calcium- activated EndoU and the corresponding part of the AF3 model. Therefore, we created a hybrid model comprising the experimental structure of calcium- activated EndoU in complex with the AF3- modeled bound RNA. This resulting model was used as input for MD simulations. For all simulations, we used the CHARMM36m force field [55] and the TIP3P water model. Each system was solvated in a cubic box with a 1.0 nm buffer zone between the protein and the box edge, and 50 mM NaCl was added with adjustments to neutralize the system. After down
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 291]]<|/det|>
+loading the generated inputs, energy minimization was executed in GROMACS using the steepest descent algorithm until the maximum force was below \(1000 \mathrm{kJ} \cdot \mathrm{mol}^{- 1} \cdot \mathrm{nm}^{- 1}\) , and then equilibration was done under NVT conditions for \(125 \mathrm{ps}\) . Particle Mesh Ewald was used for long- range electrostatics, with a cutoff of \(1.0 \mathrm{nm}\) for van der Waals interactions, and a time step of \(2 \mathrm{fs}\) was applied. The production phase of the simulations was carried out for \(1 \mu \mathrm{s}\) under NPT conditions at a temperature of \(35^{\circ} \mathrm{C}\) and a pressure of \(1 \mathrm{bar}\) . The output was then analyzed for various parameters using the built- in tools of GROMACS.  
+
+<|ref|>sub_title<|/ref|><|det|>[[114, 327, 360, 350]]<|/det|>
+## SEC-SAXS experiments  
+
+<|ref|>text<|/ref|><|det|>[[112, 368, 885, 846]]<|/det|>
+SEC- SAXS experimentsSEC- SAXS experiments were conducted on the SWING beamline at the SOLEIL synchrotron (Saint- Aubin, France). All procedures were carried out at a temperature of \(35^{\circ} \mathrm{C}\) using a buffer composed of \(10 \mathrm{mM}\) Tris pH 7, \(50 \mathrm{mM}\) NaCl, and \(1 \mathrm{mM}\) TCEP. EndoU was prepared to a concentration of \(500 \mu \mathrm{M}\) . A volume of \(75 \mu \mathrm{L}\) was injected onto a size exclusion column (Bio- SEC 3 Agilent \(100 \AA\) ), and was then eluted directly into the SAXS flow- through capillary cell at a flow rate of \(0.3 \mathrm{mL} \cdot \mathrm{min}^{- 1}\) . SAXS data were collected using an EigerX 4M detector situated \(2 \mathrm{m}\) away, using the definition of the momentum transfer \(q: q = 4 \pi \sin (\theta) / \lambda\) , where \(2 \theta\) represents the scattering angle and \(\lambda\) the X- ray wavelength (1.033 \(\AA\) for these experiments). The overall SEC- SAXS setup has been described in earlier publications [56, 57, 58]. A total of 900 SAXS frames were continuously recorded during elution, each with a duration of \(1.99 \mathrm{s}\) and a \(0.01 \mathrm{s}\) dead time between frames. 180 frames were collected before the dead volume to account for buffer scattering. Data reduction to absolute units, buffer subtraction, and averaging of identical frames corresponding to the elution peak were performed using the in- house SWING software FOXTROT [57] and BioXTAS [59]. BioXTAS was also employed to compute the gyration ratio and to estimate the molecular weight based on the volume of correlation [60]. The fitting of the EndoU homology model to the experimental SAXS data were accomplished through the Crysol software, part of the ATSAS Suite [61, 62].
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 100, 339, 121]]<|/det|>
+## Intrinsic Fluorescence  
+
+<|ref|>text<|/ref|><|det|>[[113, 140, 884, 342]]<|/det|>
+To assess the impact of calcium binding on the tertiary structure of EndoU, we measured the intrinsic fluorescence of the protein with a temperature- controlled spectrofluorometer (FS5, Edinburgh Instruments). Protein samples were prepared at \(10~\mu \mathrm{M}\) in a buffer of \(10~\mathrm{mM}\) Tris pH 7, \(50~\mathrm{mM}\) NaCl, and \(1\mathrm{mM}\) TCEP, and their fluorescence emission spectra were recorded at \(35^{\circ}\mathrm{C}\) . Emissions from \(300\) to \(525~\mathrm{nm}\) were recorded to detect fluorescence from tryptophan, tyrosine, and phenylalanine. Slit widths for excitation and emission were set at \(5\mathrm{nm}\) , and each spectrum was an average of three scans, corrected for buffer baseline.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 104, 205, 127]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 161, 735, 185]]<|/det|>
+## EndoU Expression and RNase Activity in a Thymocyte Model  
+
+<|ref|>text<|/ref|><|det|>[[113, 203, 885, 435]]<|/det|>
+In mammals, EndoU expression is limited to specific cell types. Analysis of Immunological Genome Project data [63] on mRNA from 211 mouse hematopoietic cell types revealed strong EndoU expression in developing thymocytes, starting at the double negative (DN) 2- 3 transition and progressing through the double positive (DP) stages (Fig. 1 A). EndoU expression is absent in the later stages: single positive thymocytes that survive selection and circulating T cells. Outside the hematopoietic system, EndoU protein staining in human samples [64] showed cytoplasmic expression in stratified squamous epithelia (e.g., skin, esophagus, cervix) and the trophoblast layer in the placenta.  
+
+<|ref|>text<|/ref|><|det|>[[112, 454, 886, 806]]<|/det|>
+For molecular and biochemical analysis of EndoU, we used the mouse thymic lymphoma cell line VL3- 3M2 [35], which resembles double positive thymocytes with high EndoU expression (Fig. S1 A). Upon stimulation with PMA/ionomycin, anti- CD3/CD28, or anti- CD3/CD4, the cell line shows further maturation, including downregulation of Rag1 and EndoU and upregulation of CD5 (Fig. S1 B). EndoU was knocked out in VL3- 3M2 cells using CRISPR/Cas9. We confirmed the deletion of the genomic region (Fig. S1 C) and the loss of EndoU mRNA (Fig. S1 D). We assayed endogenous ribonuclease activity in WT and EndoU KO VL3- 3M2 extracts, based on experiments in Xenopus laevis egg extracts [25]. Incubation of cytoplasmic extracts at \(37^{\circ}\mathrm{C}\) for 15 minutes without divalent metals caused no RNA degradation (Fig. 1 B). However, WT extracts with \(5\mathrm{mM}\) \(\mathrm{Ca}^{2 + }\) showed robust RNA cleavage, which was absent in EndoU KO extracts and rescued by expressing WT or HA- tagged EndoU in KO cells (Fig. 1 B). Thus, VL3- 3M2 extracts have strong EndoU- and \(\mathrm{Ca}^{2 + }\) - dependent ribonuclease activity.  
+
+<|ref|>text<|/ref|><|det|>[[114, 824, 884, 905]]<|/det|>
+To confirm the role and enzymatic properties of EndoU, we used HA- tagged EndoU rescue cells for on- bead in vitro cleavage assays with immunoprecipitated EndoU. Since mouse EndoU cleaved various RNA sequences (Fig. S2 A), we used an arbitrary substrate (50 mer 1) for subsequent ex
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 262]]<|/det|>
+periments (Table S2). Time course measurements (Fig. S1 B) were used to calculate initial reaction rates (Fig. S2 C). Mutation of two critical residues [65] abolished activity (Fig. 1 D), confirming the role of EndoU in RNA cleavage. We showed that only \(\mathrm{Ca^{2 + }}\) stimulated cleavage, unlike \(\mathrm{Mn^{2 + }}\) or other divalent metals (Fig. 1 E, S2 B), with optimal activity at \(1 - 2 \mathrm{mM} \mathrm{Ca^{2 + }}\) . EndoU showed little dependence on \(\mathrm{Na^{+}}\) or \(\mathrm{K^{+}}\) and sustained activity across pH 4- 8 (Fig. S2 D, E, F). These results indicate EndoU is a calcium- activated ribonuclease targeting a large repertoire of RNAs.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 297, 671, 320]]<|/det|>
+## Structural Basis for EndoU Activation by Calcium Ions  
+
+<|ref|>text<|/ref|><|det|>[[113, 338, 885, 632]]<|/det|>
+To examine structural changes in EndoU upon \(\mathrm{Ca^{2 + }}\) activation, we first computed a homology model of human EndoU using XendoU crystal structure [34] as template via SWISS- MODEL [66]. The apo- EndoU structure is globular, with a predominantly \(\beta\) - sheet catalytic core and an \(\alpha\) - helical bundle N- terminal extension (Fig. S3 A). We expressed the XendoU catalytic domain (Fig. 1 D) of human EndoU with \(^{15}\mathrm{N}\) labeling for NMR spectroscopy. The \(^{15}\mathrm{N}\) SOFAST- HMQC spectrum of apo- EndoU showed well- dispersed signals but fewer than expected, suggesting conformational exchange in the \(\mu \mathrm{s}\) - ms range (Fig. S3 B). Size Exclusion Chromatography with Small- Angle X- ray Scattering (SEC- SAXS) validated the correct folding of recombinant EndoU, matching theoretical predictions (Fig. S3 C). Furthermore, the crystal program [67] showed strong correlation between experimental and theoretical SAXS data, validating the structural model (Fig. S3 D).  
+
+<|ref|>text<|/ref|><|det|>[[113, 650, 885, 914]]<|/det|>
+We then studied the effect of divalent metal ions on EndoU structure and dynamics by comparing \(^{15}\mathrm{N}\) SOFAST- HMQC spectra of apo- EndoU to metal- bound states. Saturating concentrations of magnesium, nickel, or strontium caused signal loss in the NMR spectra, suggesting either protein aggregation or increased conformational exchange (Fig. S4 A, B, C). In contrast, saturating calcium restored a set of well- dispersed peaks in the NMR spectrum (Fig. S4 D). At sub- saturating calcium levels, we observed chemical shift perturbations and a shift in the fluorescence spectrum of the protein (Fig. S5 A, B). Intriguingly, the addition of a \(2'\) - fluorinated nonhydrolyzable RNA in the presence of sub- saturating calcium produced effects similar to those observed with saturating calcium, including the restoration of a set of well- dispersed peaks (Fig. S5 C). This was not
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 884, 201]]<|/det|>
+observed in the absence of calcium, where the substrate analog did not significantly alter the NMR spectrum (Fig. S5 D). These findings suggests a two- step activation process involving local structural changes at lower calcium concentrations and the abrogation of conformational exchange in the \(\mu \mathrm{s}\) - ms range at higher calcium concentrations or upon binding of a substrate analog.  
+
+<|ref|>text<|/ref|><|det|>[[112, 217, 885, 664]]<|/det|>
+To elucidate the effect of calcium, we crystallized EndoU with an excess of calcium and solved its structure at 1.7 Å resolution. The structure revealed five calcium ions, with one aiding crystal packing and four potentially activating the protein (Fig. 2 A, S6). Each calcium ion is coordinated by seven oxygen atoms from acidic side- chains, backbone carbonyl groups, or protein- stabilized water molecules (Fig. 2 B). Sites (1) and (3) include residues from both the catalytic core and the eukaryotic- specific N- terminal extension (Fig. S7), with site (1) located 12.8 Å away from the catalytic triad (H285, H300, K343). Comparing apo- XendoU, apo- EndoU, and calcium- activated EndoU structures revealed conformational changes upon calcium binding that mainly cluster nearby protein loops (Fig. 2 C, D). The side- chain of E290, located midway between site (1) and the catalytic triad, flips to engage with a water molecule in the calcium coordination network. This correlates with the side- chain rotation of catalytic H285, then locked by an electrostatic interaction with E290 (Fig. 2 C). The bonding correlates with a disorder- to- order transition in the loop carrying E290, forming a \(\beta\) - hairpin and stabilizing the catalytic site (Fig. 2 D). Our data show a model for the calcium- dependent regulation of EndoU through allostery, with site (1) and residue E290 as key mediators in the intramolecular signaling leading to EndoU activation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 699, 644, 722]]<|/det|>
+## Experimental Validation of EndoU Activation Model  
+
+<|ref|>text<|/ref|><|det|>[[114, 741, 885, 913]]<|/det|>
+To validate our structure- based model for calcium- mediated EndoU activation, we first designed variants with altered calcium- binding sites. RNA degradation assays were conducted for each variant (Fig. 3 A). Without calcium or wild- type EndoU, no RNA cleavage was detected, whereas their presence led to almost complete RNA degradation over time. The degradation data fitted a first- order reaction model, providing a kinetic parameter describing the reaction (Fig. 3 B). Disrupting calcium binding sites (2) or (4) with mutations E226A or D330A led to RNA degradation rates
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 883, 201]]<|/det|>
+similar to the wild- type (Fig. 3 A, C). In contrast, disruption of calcium binding sites (1) or (3) with mutations E284A or D179A abolished enzymatic activity. Interestingly, calcium binding sites (1) and (3) are defined by residues from both the catalytic core and the eukaryote- specific N- terminal extension, while this is not the case for sites (2) and (4) (Fig. S7).  
+
+<|ref|>text<|/ref|><|det|>[[113, 219, 885, 483]]<|/det|>
+Our results clearly indicate that the eukaryote- specific N- terminal extension of EndoU contributes to calcium sensing, thereby enabling allosteric regulation. Even though the crystal structure of calcium- activated EndoU could explain the role of calcium binding site (1) in this process, it was not the case for site (3). We hypothesized that calcium binding to site (1) could be promoted by a prior binding event at site (3) and relied on Molecular Dynamics (MD) experiments to test this hypothesis. With the wild- type EndoU, we observed calcium binding to site (1) within less than 100 ns simulation time (Fig. 3 D). Disrupting site (3) with mutation D179A resulted in no stable binding at site (1) for any of the five calcium ions added in the \(1 \mu \mathrm{s}\) simulation. This suggests that that cooperative binding of calcium at sites (1) and (3) leads to EndoU activation.  
+
+<|ref|>text<|/ref|><|det|>[[113, 500, 885, 883]]<|/det|>
+We further proposed that calcium sensing information at site (1) was communicated to the remote catalytic site through water- mediated intramolecular signaling events enabled by key residue E290 (Fig. 2 C). Disruption of the intramolecular signaling cascade with mutation E290A completely abrogated activity, while charge- conservative mutation E290D resulted in a 4- fold increase in the enzymatic reaction rate (Fig. 3A, C). Consistent with our structural model, a negatively charged side- chain at position 290 is required for the allosteric activation of EndoU by calcium. In this model, E290 locks the catalytic H285 side- chain in an active conformation. Accordingly, substitution of catalytic H285 with alanine completely abrogated enzymatic activity, as observed for catalytic mutant H300A, underscoring the importance of the histidine pair in catalysis. Substitution of catalytic residue K343 by alanine resulted in nearly half reduction of enzymatic activity, consistent with the role of K343 as a stabilizer of reaction intermediates. Overall, mutagenesis experiments corroborate the residue assignments proposed in our structure- based model for calcium- mediated activation of EndoU.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 99, 719, 122]]<|/det|>
+## Calcium-Activated EndoU in Complex with an RNA Analog  
+
+<|ref|>text<|/ref|><|det|>[[112, 139, 886, 616]]<|/det|>
+To experimentally determine the RNA- binding surface of calcium- activated EndoU, we first performed 3D triple resonance NMR experiments for backbone chemical shift assignment, successfully assigning \(89.3\%\) (251 out of 281 residues) of the backbone resonances (Fig. S8). Excluding the N- terminal GGSEFA sequence and nine proline residues, the assignment coverage increased to \(94.3\%\) . Next, we recorded \(^{15}\mathrm{N}\) and \(^{13}\mathrm{C}\) SOFAST- HMQC spectra of calcium- activated \(^{13}\mathrm{C}^{15}\mathrm{N}\) - labeled EndoU with a \(2^{\prime}\) - fluorinated RNA substrate analog (Fig. 4 A, B). Severe line broadening in a subset of crosspeaks and additional spectral changes in the rapid exchange regime were observed (Fig. 4 A). We calculated \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) chemical shift perturbations between RNA- bound and unbound EndoU, noting particularly the residues that disappeared in the bound state (Fig. 4 C). All disappearing residues are located in the C- terminal catalytic core, covering the \(\beta\) - sheet surface composed of two independent \(\beta\) - sheets and a short \(\alpha\) - helix (Fig. 4 D). The strongest chemical shift perturbations also correspond to residues in this area. Electrostatic analysis indicated that the \(\beta\) - sheet is highly basic, suitable for RNA binding (Fig. 4 E). In contrast, the N- terminal \(\alpha\) - helical region showed minimal changes upon RNA binding, with no disappearing resonances, and displayed a neutral or acidic surface charge. These data support that RNA binds to the conserved catalytic core of the protein, involving an extended, basic \(\beta\) - sheet- rich groove for RNA binding.  
+
+<|ref|>text<|/ref|><|det|>[[113, 632, 886, 896]]<|/det|>
+To model the complex between calcium- activated EndoU and RNA, we used AlphaFold (AF) 3 [54] with the primary sequence of EndoU, a \((\mathrm{U})_6\) RNA, and four calcium ions as inputs. We controlled that a \(2^{\prime}\mathrm{F}(\mathrm{U})_6\) RNA analog interacts with calcium- activated EndoU (Fig. S9). Using the top- ranked AF model as input, we then conducted a \(1 \mu \mathrm{s}\) Molecular Dynamics (MD) simulation using the top- ranked AF model as input. Over the trajectory, we clustered structures based on RNA conformations to obtain a final ensemble of RNA- bound calcium- activated EndoU models (Fig. S10 A). All models consistently reproduced the crystal structure of calcium- activated EndoU, with a mean RMSD of \(1.18 \pm 0.29 \mathrm{\AA}\) . The intermolecular interface with the \((\mathrm{U})_6\) RNA is defined by the cleft between the two front \(\beta\) - sheets, where nucleotide \(\mathrm{U}_2\) anchors in close proximity to the catalytic
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 261]]<|/det|>
+triad, exposing its sugar 2'OH for nucleophilic attack by residue H300 (Fig. S10 B, S11). Conformational heterogeneity is observed at the interface with EndoU across the models for the rest of the RNA sequence, consistent with the experimental NMR data showing line broadening for residues in the RNA- binding region due to conformational sampling in the \(\mu \mathrm{s}\) - ms range. Overall, the ensemble of models aligns well with our experimental data, providing a robust structural hypothesis for calcium- activated, RNA- bound EndoU.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 103, 244, 127]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[113, 150, 884, 265]]<|/det|>
+In this work, we elucidated the molecular basis for the \(\mathrm{Ca^{2 + }}\) - dependent activation of human EndoU, with implications for the entire eukaryotic EndoU family due to the conserved sequence and structure of its catalytic domain across eukarya. Our findings, in conjunction with existing data, suggest an allosteric rather than catalytic requirement for a divalent metal in EndoU cleavage (Fig. 5).  
+
+<|ref|>text<|/ref|><|det|>[[112, 279, 886, 757]]<|/det|>
+Indeed, a common evolutionary origin with the RNase A family has been proposed based on structural and distant sequence similarities [6]. RNase A enzymes use a catalytic triad of two histidines and a lysine [68], and the Xenopus EndoU structure suggested a mechanistic similarity, where conserved His and Lys residues mark the proposed catalytic site [34]. Mutation of the corresponding residues in human EndoU supported these assignments (Fig. 3 C). Then, both EndoU and RNase A leaves a \(5^{\prime}\) - OH product, which is characteristic of metal- independent endonuclease catalytic catalysis [69]. Also, bacterial and some viral EndoU homologs do not require divalent metal ions [31, 11]. In the context of eukaryotic EndoU, our experimental data identified that calcium binding was necessary for cleavage activity in both mouse and human EndoU (Fig. 2, 3 C), with coordinating residues from both the conserved catalytic core and the eukaryote- specific N- terminal extension. We explained how the binding of calcium at the interface between the two regions could trigger water- mediated intramolecular signaling, ultimately leading to local structural changes that result in the positioning of catalytic residues in an active conformation. Therefore, our work supports a model where calcium is not directly involved in catalysis but rather activates the catalytic His- His- Lys triad through allostery. At the same time, we assign a role to the eukaryotic- specific N- terminal extension in sensing calcium at relevant sites, enabling allosteric regulation by calcium ions.  
+
+<|ref|>text<|/ref|><|det|>[[113, 772, 884, 886]]<|/det|>
+Eukaryotic EndoU proteins primarily depend on calcium for activation, as our data for mouse and human EndoU indicate exclusive activation by \(\mathrm{Ca^{2 + }}\) , consistent with Xenopus EndoU findings [25]. Although early Xenopus studies also reported \(\mathrm{Mn^{2 + }}\) requirements, \(\mathrm{Ca^{2 + }}\) - dependent cleavage was observed [13], and the C. elegans endu- 2 homolog is activated by both \(\mathrm{Ca^{2 + }}\) and \(\mathrm{Mn^{2 + }}\) [70]. The
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 87, 884, 504]]<|/det|>
+physiological relevance of \(\mathrm{Mn}^{2 + }\) activation is unclear due to its low concentration in mammalian tissues [71, 72, 73], whereas \(\mathrm{Ca}^{2 + }\) concentrations vary significantly, making EndoU localization crucial for understanding \(\mathrm{Ca}^{2 + }\) - mediated activation. Animal EndoUs contain an N- terminal signal peptide, and mammalian EndoUs have two somatomedin B (SmB) domains rich in disulfide bonds, suggesting secretion or ER association [74, 75, 76]. The C. elegans endu- 2 and D. melanogaster CG2145 homologs are secreted and reuptaken in other tissues [70, 77], while Xenopus and human EndoU are cytoplasmic and ER- associated [25]. Calcium, a universal eukaryotic second messenger, is near zero in resting cells and rises to 1- 100 \(\mu \mathrm{M}\) concentration during signaling, while extracellular \(\mathrm{Ca}^{2 + }\) can reach mM levels [78]. EndoU may cleave RNAs or alter mRNA expression during thymocyte maturation or apoptosis, consistent with its proposed pro- apoptotic role in B cells [27]. The \(\mathrm{Ca}^{2 + }\) activation mechanism is likely conserved across eukaryotic EndoUs, with varying expression domains and RNA targets across species. If delivered to \(\mathrm{Ca}^{2 + }\) - rich extracellular environments, EndoU may degrade extracellular RNAs, which makes the understanding of the RNA targeting repertoire of EndoU a priority to further illuminate its biological roles.  
+
+<|ref|>text<|/ref|><|det|>[[113, 520, 885, 905]]<|/det|>
+Our work provides fundamental insights with potential applications in therapeutic and biotechnological domains. The Nsp15 protein of SARS- CoV- 2 features a nidovirus EndoU- like domain (NendoU) that requires divalent manganese ions to cleave the \(5^{\prime}\) - polyuridine tract of its negative- sense RNA, a crucial process for evading the host immune system [79]. Tipiracil, a uridine analog, effectively binds to the uridine binding site of Nsp15, inhibiting its activity and diminishing Spike protein expression in whole- cell assays, thereby inhibiting SARS- CoV- 2 [80]. However, the structural basis for the divalent metal ion dependence of Nsp15 remains unknown, posing challenges for rational drug design targeting Nsp15 and other members of the EndoU- like superfamily regulated by divalent metal ions. By elucidating the structural basis of the divalent metal ion dependence of human EndoU, we pave the way for developing inhibitors specifically targeting EndoU- like domains that depend on these ions. Furthermore, investigating whether Tipiracil can bind to and inhibit human EndoU could provide valuable insights into the cellular and extracellular functions of EndoU, potentially identifying it as a pharmacological target in certain disease contexts. The
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 261]]<|/det|>
+potential of EndoU as a biotechnological tool for specific RNA sequence cleavage is noteworthy, especially with advancements in self- driving laboratories and machine learning algorithms for enzyme design and optimization [81, 82, 83]. Altering the divalent metal ion dependence of EndoU could yield sensors for detecting environmental contaminants like lead ions in water or enable controlled RNA digestion in biological systems by introducing activating ions, offering significant insights into RNA metabolism and processes.  
+
+<|ref|>text<|/ref|><|det|>[[113, 279, 884, 512]]<|/det|>
+In conclusion, we have elucidated the molecular mechanism underlying the activation of EndoU enzymes by divalent metal ions, specifically calcium. This discovery demonstrates a functional role for the N- terminal extension in human EndoU, suggesting a conserved mechanism accross eukaryotic EndoU catalytic domains. Evolution has provided these enzymes with a distinct regulatory segment that interacts with their catalytic core to sense calcium binding and communicate with the catalytic site through allostery. Our findings pave the way for more research into the role of EndoU in biological processes and offer valuable insights for future biotechnological and therapeutic applications.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 103, 350, 128]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[113, 150, 884, 354]]<|/det|>
+AcknowledgementsWe would like to thank Stephen Smale for sharing the VL3- 3M2 cell line, and the Zhang lab for sharing the pX330 plasmid. This work is funded in part by NIH grant 1R01GM152548. We are grateful to Dr. Mikayel Aznauryan for providing us with access to the fluorescence spectroscopy facilities. We thank INSERM for funding through the ARNA internal call. We acknowledge SOLEIL for provision of synchrotron radiation facilities and we would like to thank PX1 staff members for assistance in using beamline PX1. We would like to thank Dr. Aurélien Thureau from SWING beamline at SOLEIL for assistance during SAXS data collection.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 398, 316, 423]]<|/det|>
+## Data availability  
+
+<|ref|>text<|/ref|><|det|>[[115, 447, 882, 499]]<|/det|>
+Data availabilityThe atomic coordinates of the calcium- activated EndoU structure have been deposited in the PDB under the accession code 9FTW.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 544, 373, 568]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[113, 591, 884, 856]]<|/det|>
+Author contributionsS.C., F.V.K., and F.M. designed the research. P.B. and B.V. synthesized the RNAs and 2'F analogs used in the study. K.D. and F.V.K. performed experiments on the mouse thymic lymphoma cell line VL3- 3M2, including CRISPR/Cas9 knockout of EndoU and biochemical characterization of both endogenous and immunoprecipitated EndoU. S.F. and S.T. solved the crystal structure of calcium- activated EndoU. F.M. and S.C. conducted NMR spectroscopy studies. F.M. and M.B. carried out RNA degradation assays with recombinant EndoU. F.M. conducted SAXS experiments, AlphaFold modeling, and Molecular Dynamics simulations. F.M. and F.V.K. wrote the initial draft of the manuscript. F.M., S.C., F.V.K., S.F., and S.T. discussed the initial draft and contributed to the final version of the manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 104, 359, 129]]<|/det|>
+## Competing interests  
+
+<|ref|>text<|/ref|><|det|>[[115, 154, 457, 172]]<|/det|>
+The authors declare no competing interests.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 103, 249, 127]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[115, 145, 886, 911]]<|/det|>
+[1] Kushner SR. mRNA decay in prokaryotes and eukaryotes: different approaches to a similar problem. IUBMB life, 56(10):585- 94, 2004. [2] Evguenieva- Hackenberg E and Klug G. RNA degradation in Archaea and Gram- negative bacteria different from Escherichia coli. Progress in molecular biology and translational science, 85:275- 317, 2009. [3] Schmid M and Jensen TH. The exosome: a multipurpose RNA- decay machine. Trends in biochemical sciences, 33(10):501- 10, 2008. [4] Tomecki R and Dziembowski A. Novel endoribonucleases as central players in various pathways of eukaryotic RNA metabolism. RNA (New York, N.Y.), 16(9):1692- 724, 2010. [5] Paysan- Lafosse T, Blum M, Chuguransky S, Grego T, Pinto BL, Salazar GA, Bileschi ML, Bork P, Bridge A, Colwell L, Gough J, Haft DH, Letunic I, Marchler- Bauer A, Mi H, Natale DA, Orengo CA, Pandurangan AP, Rivoire C, Sigrist CJA, Sillitoe I, Thanki N, Thomas PD, Tosatto SCE, Wu CH, and Bateman A. InterPro in 2022. Nucleic acids research, 51(D1):D418- D427, 2023. [6] Mushegian A, Sorokina I, Eroshkin A, and Dlakic M. An ancient evolutionary connection between Ribonuclease A and EndoU families. RNA (New York, N.Y.), 26(7):803- 813, 2020. [7] Raines RT. Ribonuclease A. Chemical reviews, 98(3):1045- 1066, 1998. [8] SMYTH DG, STEIN WH, and MOORE S. The sequence of amino acid residues in bovine pancreatic ribonuclease: revisions and confirmations. The Journal of biological chemistry, 238:227- 34, 1963. [9] Findlay D, Herries DG, Mathias AP, Rabin BR, and Ross CA. The active site and mechanism of action of bovine pancreatic ribonuclease. 7. The catalytic mechanism. The Biochemical journal, 85(1):152- 3, 1962. [10] Avey HP, Boles MO, Carlisle CH, Evans SA, Morris SJ, Palmer RA, Woolhouse BA, and Shall S. Structure of ribonuclease. Nature, 213(5076):557- 62, 1967. [11] Michalska K, Quan Nhan D, Willett JLE, Stols LM, Eschenfeldt WH, Jones AM, Nguyen JY, Koskiniemi S, Low DA, Goulding CW, Joachimiak A, and Hayes CS. Functional plasticity of antibacterial EndoU toxins. Molecular microbiology, 109(4):509- 527, 2018. [12] Caffarelli E, Arese M, Santoro B, Fragapane P, and Bozzoni I. In vitro study of processing of the intron- encoded U16 small nucleolar RNA in Xenopus laevis. Molecular and cellular biology, 14(5):2966- 74, 1994.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 90, 884, 159]]<|/det|>
+[13] Caffarelli E, Maggi L, Fatica A, Jiricny J, and Bozzoni I. A novel Mn++- dependent ribonuclease that functions in U16 SnoRNA processing in X. laevis. Biochemical and biophysical research communications, 233(2):514- 7, 1997.  
+
+<|ref|>text<|/ref|><|det|>[[112, 171, 884, 239]]<|/det|>
+[14] Gioia U, Laneve P, Dlakic M, Arceci M, Bozzoni I, and Caffarelli E. Functional characterization of Xen- doU, the endoribonuclease involved in small nucleolar RNA biosynthesis. The Journal of biological chemistry, 280(19):18996- 9002, 2005.  
+
+<|ref|>text<|/ref|><|det|>[[112, 251, 884, 320]]<|/det|>
+[15] Laneve P, Altieri F, Fiori ME, Scaloni A, Bozzoni I, and Caffarelli E. Purification, cloning, and characterization of XenDoU, a novel endoribonuclease involved in processing of intron- encoded small nucleolar RNAs in Xenopus laevis. The Journal of biological chemistry, 278(15):13026- 32, 2003.  
+
+<|ref|>text<|/ref|><|det|>[[112, 332, 884, 400]]<|/det|>
+[16] Mistry J, Chuguransky S, Williams L, Qureshi M, Salazar GA, Sonnhammer ELL, Tosatto SCE, Paladin L, Raj S, Richardson LJ, Finn RD, and Bateman A. Pfam: The protein families database in 2021. Nucleic acids research, 49(D1):D412- D419, 2021.  
+
+<|ref|>text<|/ref|><|det|>[[112, 412, 882, 456]]<|/det|>
+[17] Bohn H and Winckler W. [Isolation and characterization of a new placental tissue protein (PP11) (author's transl)]. Archives of gynecology, 229(4):293- 301, 1980.  
+
+<|ref|>text<|/ref|><|det|>[[112, 467, 884, 538]]<|/det|>
+[18] Inaba N, Renk T, Weinmann E, and Bohn H. Immunohistochemical location of proteins related to pregnancy- specific beta 1- glycoprotein (SP1) and placental tissue proteins (PP5, PP8, PP9, PP10, PP11, PP12) in the after- birth of Cynomolgus monkeys (Macaca fascicularis). Archives of gynecology, 230(3):195- 204, 1981.  
+
+<|ref|>text<|/ref|><|det|>[[112, 549, 884, 593]]<|/det|>
+[19] Xu C, Zhang Y, Shen Y, Shi Y, Zhang M, and Zhou L. Integrated Analysis Reveals <i>ENDOU</i> as a Biomarker in Head and Neck Squamous Cell Carcinoma Progression. Frontiers in oncology, 10:522332, 2020.  
+
+<|ref|>text<|/ref|><|det|>[[112, 604, 884, 671]]<|/det|>
+[20] Basic V, Zhang B, Domert J, Pellas U, and Tot T. Integrative meta- analysis of gene expression profiles identifies FEN1 and ENDOU as potential diagnostic biomarkers for cervical squamous cell carcinoma. Oncology letters, 22(6):840, 2021.  
+
+<|ref|>text<|/ref|><|det|>[[112, 683, 884, 727]]<|/det|>
+[21] Lou Y, Ren L, and Wang L. Hsa_circ_0049396 inhibited oral squamous cell carcinoma progression by regulating the miR- 663b/ENDOU axis. Genomics, 115(1):110553, 2023.  
+
+<|ref|>text<|/ref|><|det|>[[112, 739, 884, 832]]<|/det|>
+[22] Inaba N, Ishige H, Ijichi M, Satoh N, Ohkawa R, Sekiya S, Shirotake S, Takamizawa H, Renk T, and Bohn H. Immunohistochemical detection of pregnancy- specific protein (SP1) and placenta- specific tissue proteins (PP5, PP10, PP11 and PP12) in ovarian adenocarcinomas. Oncodevelopmental biology and medicine: the journal of the International Society for Oncodevelopmental Biology and Medicine, 3(5- 6):379- 89, 1982.  
+
+<|ref|>text<|/ref|><|det|>[[112, 844, 884, 912]]<|/det|>
+[23] Inaba N, Renk T, Wurster K, Rapp W, and Bohn H. Ectopic synthesis of pregnancy specific beta 1- glycoprotein (SP1) and placental specific tissue proteins (PP5, PP10, PP11, PP12) in nontrophoblastic malignant tumours. Possible markers in oncology. Klinische Wochenschrift, 58(15):789- 91, 1980.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 90, 884, 135]]<|/det|>
+[24] Inaba N, Renk T, Daume E, and Bohn H. Ectopic production of placenta-"specific" tissue proteins (PP5 and PP11) by malignant breast tumors. Archives of gynecology, 231(1):87- 90, 1981.  
+
+<|ref|>text<|/ref|><|det|>[[112, 149, 884, 193]]<|/det|>
+[25] Schwarz DS and Blower MD. The calcium- dependent ribonuclease XendoU promotes ER network formation through local RNA degradation. The Journal of cell biology, 207(1):41- 57, 2014.  
+
+<|ref|>text<|/ref|><|det|>[[112, 207, 884, 250]]<|/det|>
+[26] Zhao G and Blackstone C. ER morphology: sculpting with XendoU. Current biology: CB, 24(24):R1170- 2, 2014.  
+
+<|ref|>text<|/ref|><|det|>[[112, 264, 884, 308]]<|/det|>
+[27] Poe JC, Kountikov EI, Lykken JM, Natarajan A, Marchuk DA, and Tedder TF. EndoU is a novel regulator of AICD during peripheral B cell selection. The Journal of experimental medicine, 211(1):57- 69, 2014.  
+
+<|ref|>text<|/ref|><|det|>[[112, 321, 884, 390]]<|/det|>
+[28] Laneve P, Piacentini L, Casale AM, Capauto D, Gioia U, Cappucci U, Di Carlo V, Bozzoni I, Di Micco P, Morea V, Di Franco CA, and Caffarelli E. Drosophila CG3303 is an essential endoribonuclease linked to TDP- 43- mediated neurodegeneration. Scientific reports, 7:41559, 2017.  
+
+<|ref|>text<|/ref|><|det|>[[112, 404, 884, 498]]<|/det|>
+[29] Ujisawa T, Ohta A, Ii T, Minakuchi Y, Toyoda A, Ii M, and Kuhara A. Endoribonuclease ENDU- 2 regulates multiple traits including cold tolerance via cell autonomous and nonautonomous controls in <i>Caeonorhabditis elegans</i>. Proceedings of the National Academy of Sciences of the United States of America, 115(35):8823- 8828, 2018.  
+
+<|ref|>text<|/ref|><|det|>[[112, 511, 884, 580]]<|/det|>
+[30] Zhang D, de Souza RF, Anantharaman V, Iyer LM, and Aravind L. Polymorphic toxin systems: Comprehensive characterization of trafficking modes, processing, mechanisms of action, immunity and ecology using comparative genomics. Biology direct, 7:18, 2012.  
+
+<|ref|>text<|/ref|><|det|>[[112, 594, 884, 662]]<|/det|>
+[31] Nedialkova DD, Ulferts R, van den Born E, Lauber C, Gorbalenya AE, Ziebuhr J, and Snijder EJ. Biochemical characterization of arterivirus nonstructural protein 11 reveals the nidovirus- wide conservation of a replicative endoribonuclease. Journal of virology, 83(11):5671- 82, 2009.  
+
+<|ref|>text<|/ref|><|det|>[[112, 676, 884, 745]]<|/det|>
+[32] Ivanov KA, Hertzig T, Rozanov M, Bayer S, Thiel V, Gorbalenya AE, and Ziebuhr J. Major genetic marker of nidoviruses encodes a replicative endoribonuclease. Proceedings of the National Academy of Sciences of the United States of America, 101(34):12694- 9, 2004.  
+
+<|ref|>text<|/ref|><|det|>[[112, 759, 884, 803]]<|/det|>
+[33] Bhardwaj K, Guarino L, and Kao CC. The severe acute respiratory syndrome coronavirus Nsp15 protein is an endoribonuclease that prefers manganese as a cofactor. Journal of virology, 78(22):12218- 24, 2004.  
+
+<|ref|>text<|/ref|><|det|>[[112, 816, 884, 885]]<|/det|>
+[34] Renzi F, Caffarelli E, Laneve P, Bozzoni I, Brunori M, and Vallone B. The structure of the endoribonuclease XendoU: From small nucleolar RNA processing to severe acute respiratory syndrome coronavirus replication. Proceedings of the National Academy of Sciences of the United States of America, 103(33):12365- 70, 2006.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 90, 886, 137]]<|/det|>
+[35] T. Groves, P. Katis, Z. Madden, K. Manickam, D. Ramsden, G. Wu, and C. J. Guidos. In vitro maturation of clonal cd4+cd8+ cell lines in response to tcr engagement. Journal of immunology, 154(10):5011- 22, 1995.  
+
+<|ref|>text<|/ref|><|det|>[[112, 150, 883, 194]]<|/det|>
+[36] E.L. Sternburg, K. C. Dias, and F. V. Karginov. Selection- dependent and independent generation of crispr/cas9- mediated gene knockouts in mammalian cells. J. Vis. Exp., (124):e55903, 2017.  
+
+<|ref|>text<|/ref|><|det|>[[112, 207, 883, 252]]<|/det|>
+[37] Liu H and Naismith JH. An efficient one- step site- directed deletion, insertion, single and multiple- site plasmid mutagenesis protocol. BMC biotechnology, 8:91, 2008.  
+
+<|ref|>text<|/ref|><|det|>[[112, 264, 883, 309]]<|/det|>
+[38] Schanda P, Kupee E, and Brutscher B. SOFAST- HMQC experiments for recording two- dimensional deteronuclear correlation spectra of proteins within a few seconds. Journal of Biomolecular NMR, 33(4):199- 211, 2005.  
+
+<|ref|>text<|/ref|><|det|>[[112, 321, 883, 366]]<|/det|>
+[39] Favier A and Brutscher B. Recovering lost magnetization: polarization enhancement in biomolecular NMR. Journal of Biomolecular NMR, 49(1):9- 15, 2011.  
+
+<|ref|>text<|/ref|><|det|>[[112, 378, 883, 450]]<|/det|>
+[40] Lescop E, Kern T, and Brutscher B. Guidelines for the use of band- selective radiofrequency pulses in heteronuclear NMR: example of longitudinal- relaxation- enhanced BEST- type 1H- 15N correlation experiments. Journal of Magnetic Resonance, 203(1):190- 8, 2010.  
+
+<|ref|>text<|/ref|><|det|>[[112, 462, 815, 481]]<|/det|>
+[41] Keller R and Wuthrich K. Computer- aided resonance assignment (CARA). available at www.nmr.ch.  
+
+<|ref|>text<|/ref|><|det|>[[112, 494, 883, 563]]<|/det|>
+[42] Vranken WF, Boucher W, Stevens TJ, Fogh RH, Pajon A, Llinas M, Ulrich EL, Markley JL, Ionides J, and Laue ED. The CCPN data model for NMR spectroscopy: development of a software pipeline. Proteins, 59(4):687- 96, 2005.  
+
+<|ref|>text<|/ref|><|det|>[[112, 576, 883, 620]]<|/det|>
+[43] Williamson MP. Using chemical shift perturbation to characterise ligand binding. Progress in Nuclear Magnetic Resonance Spectroscopy, 73:1- 16, 2013.  
+
+<|ref|>text<|/ref|><|det|>[[112, 633, 836, 653]]<|/det|>
+[44] Kabsch W. XDS. Acta crystallographica. Section D, Biological crystallography, 66(Pt 2):125- 32, 2010.  
+
+<|ref|>text<|/ref|><|det|>[[112, 666, 883, 710]]<|/det|>
+[45] McCoy AJ, Grosse- Kunstleve RW, Adams PD, Winn MD, Storoni LC, and Read RJ. Phaser crystallographic software. Journal of applied crystallography, 40(Pt 4):658- 674, 2007.  
+
+<|ref|>text<|/ref|><|det|>[[112, 723, 884, 842]]<|/det|>
+[46] Liebschner D, Afonine PV, Baker ML, Bunkóczi G, Chen VB, Croll TI, Hintze B, Hung LW, Jain S, McCoy AJ, Moriarty NW, Oeffner RD, Poon BK, Prisant MG, Read RJ, Richardson JS, Richardson DC, Sammito MD, Sobolev OV, Stockwell DH, Terwilliger TC, Urzhumtsev AG, Videau LL, Williams CJ, and Adams PD. Macromolecular structure determination using X- rays, neutrons and electrons: recent developments in Phenix. Acta crystallographica. Section D, Structural biology, 75(Pt 10):861- 877, 2019.  
+
+<|ref|>text<|/ref|><|det|>[[112, 856, 883, 901]]<|/det|>
+[47] Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, Tunyasuvunakool K, Bates R, Žídek A, Potapenko A, Bridgland A, Meyer C, Kohl SAA, Ballard AJ, Cowie A, Romera- Paredes B, Nikolov S, Jain
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[149, 90, 884, 160]]<|/det|>
+R, Adler J, Back T, Petersen S, Reiman D, Clancy E, Zielinski M, Steinegger M, Pacholska M, Berghammer T, Bodenstein S, Silver D, Vinyals O, Senior AW, Kavukcuoglu K, Kohli P, and Hassabis D. Highly accurate protein structure prediction with AlphaFold. Nature, 596(7873):583–589, 2021.  
+
+<|ref|>text<|/ref|><|det|>[[113, 172, 884, 216]]<|/det|>
+[48] Bricogne G, Blanc E, Brandl M, Flensburg C, Keller P, Paciorek W, Roversi P, Sharff A, Smart OS, Vonrhein C, and Womack TO. BUSTER version 2.10.4. Cambridge, United Kingdom: Global Phasing Ltd, 2017.  
+
+<|ref|>text<|/ref|><|det|>[[113, 228, 653, 246]]<|/det|>
+[49] Lazar I and Lazar I. GelAnalyzer 23.1.1. available at www.gelanalyzer.com.  
+
+<|ref|>text<|/ref|><|det|>[[113, 258, 884, 378]]<|/det|>
+[50] Virtanen P, Gommers R, Oliphant TE, Haberland M, Reddy T, Cournapeau D, Burovski E, Peterson P, Weckesser W, Bright J, van der Walt SJ, Brett M, Wilson J, Millman KJ, Mayorov N, Nelson ARJ, Jones E, Kern R, Larson E, Carey CJ, Polat I, Feng Y, Moore EW, VanderPlas J, Laxalde D, Perktold J, Cimrman R, Henriksen I, Quintero EA, Harris CR, Archibald AM, Ribeiro AH, Pedregosa F, and van Mulbregt P. SciPy 1.0: fundamental algorithms for scientific computing in Python. Nature methods, 17(3):261–272, 2020.  
+
+<|ref|>text<|/ref|><|det|>[[113, 389, 884, 433]]<|/det|>
+[51] Van Der Spoel D, Lindahl E, Hess B, Groenhof G, Mark AE, and Berendsen HJ. GROMACS: fast, flexible, and free. Journal of computational chemistry, 26(16):1701–18, 2005.  
+
+<|ref|>text<|/ref|><|det|>[[113, 445, 884, 489]]<|/det|>
+[52] Jo S, Kim T, Iyer VG, and Im W. CHARMM-GUI: a web- based graphical user interface for CHARMM. Journal of computational chemistry, 29(11):1859–65, 2008.  
+
+<|ref|>text<|/ref|><|det|>[[113, 500, 884, 595]]<|/det|>
+[53] Lee J, Cheng X, Swails JM, Yeom MS, Eastman PK, Lemkul JA, Wei S, Buckner J, Jeong JC, Qi Y, Jo S, Pande VS, Case DA, Brooks CL, MacKerell AD, Klauda JB, and Im W. CHARMM-GUI Input Generator for NAMD, GROMACS, AMBER, OpenMM, and CHARMM/OpenMM Simulations Using the CHARMM36 Additive Force Field. Journal of chemical theory and computation, 12(1):405–13, 2016.  
+
+<|ref|>text<|/ref|><|det|>[[113, 607, 884, 775]]<|/det|>
+[54] Abramson J, Adler J, Dunger J, Evans R, Green T, Pritzel A, Ronneberger O, Willmore L, Ballard AJ, Bambrick J, Bodenstein SW, Evans DA, Hung CC, O'Neill M, Reiman D, Tunyasuvunakool K, Wu Z, Žemgulytė A, Arvaniti E, Beattie C, Bertolli O, Bridgland A, Cherepanov A, Congreve M, Cowen- Rivers AI, Cowie A, Figurnov M, Fuchs FB, Gladman H, Jain R, Khan YA, Low CMR, Perlin K, Potapenko A, Savy P, Singh S, Stecula A, Thillaisundaram A, Tong C, Yakneen S, Zhong ED, Zielinski M, Židek A, Bapst V, Kohli P, Jaderberg M, Hassabis D, and Jumper JM. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature, 2024.  
+
+<|ref|>text<|/ref|><|det|>[[113, 788, 884, 856]]<|/det|>
+[55] Huang J, Rauscher S, Nawrocki G, Ran T, Feig M, de Groot BL, Grubmüller H, and MacKerell AD. CHARMM36m: an improved force field for folded and intrinsically disordered proteins. Nature methods, 14(1):71–73, 2017.  
+
+<|ref|>text<|/ref|><|det|>[[113, 869, 884, 912]]<|/det|>
+[56] A Thureau, P Roblin, and J Perez. BioSAXS on the SWING beamline at Synchrotron SOLEIL. Journal of applied crystallography, 54(6):1698–1710, 2021.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 90, 884, 161]]<|/det|>
+[57] G David and J Pérez. Combined sampler robot and high- performance liquid chromatography: a fully automated system for biological small- angle X- ray scattering experiments at the Synchrotron SOLEIL SWING beamline. Journal of applied crystallography, 42:892- 900, 2009.  
+
+<|ref|>text<|/ref|><|det|>[[112, 174, 883, 219]]<|/det|>
+[58] J Pérez and Y Nishino. Advances in X- ray scattering: from solution SAXS to achievements with coherent beams. Current Opinion in Structural Biology, 22(5):670- 8, 2012.  
+
+<|ref|>text<|/ref|><|det|>[[112, 231, 883, 277]]<|/det|>
+[59] Hopkins JB, Gillilan RE, and Skou S. BioXTAS RAW: improvements to a free open- source program for small- angle X- ray scattering data reduction and analysis. Journal of applied crystallography, 50(5):1545- 1553, 2017.  
+
+<|ref|>text<|/ref|><|det|>[[112, 289, 883, 334]]<|/det|>
+[60] R P Rambo and J A Tainer. Accurate assessment of mass, models and resolution by small- angle scattering. Nature, 496(7446):477- 81, 2013.  
+
+<|ref|>text<|/ref|><|det|>[[112, 346, 884, 416]]<|/det|>
+[61] Grishaev A, Guo L, Irving T, and Bax A. Improved fitting of solution X- ray scattering data to macromolecular structures and structural ensembles by explicit water modeling. Journal of the American Chemical Society, 132(44):15484- 6, 2010.  
+
+<|ref|>text<|/ref|><|det|>[[112, 428, 884, 522]]<|/det|>
+[62] D Franke, M V Petoukhov, P V Konarev, A Panjkovich, A Tuukkanen, H D T Mertens, A G Kikhney, N R Hajizadeh, J M Franklin, C M Jeffries, and D I Svergun. ATSAS 2.8: a comprehensive data analysis suite for small- angle scattering from macromolecular solutions. Journal of Applied Crystallography, 50(4):1212- 1225, 2017.  
+
+<|ref|>text<|/ref|><|det|>[[112, 535, 883, 580]]<|/det|>
+[63] T. S. Heng and M. W. Painter. The immunological genome project: networks of gene expression in immune cells. Nature immunology, 9(10):1091- 4, 2008.  
+
+<|ref|>text<|/ref|><|det|>[[112, 592, 884, 738]]<|/det|>
+[64] M. Uhlen, L. Fagerberg, B. M. Hallstrom, C. Lindskog, P. Oksvold, A. Mardinoglu, A. Sivertsson, C. Kampf, E. Sjostedt, A. Asplund, I. Olsson, K. Edlund, E. Lundberg, S. Navani, C. A. Szigyarto, J. Odeberg, D. Djureinovic, J. O. Takanen, S. Hober, T. Alm, P. H. Edqvist, H. Berling, H. Tegel, J. Mulder, J. Rockberg, P. Nilsson, J. M. Schwenk, M. Hamsten, K. von Feilitzen, M. Forsberg, L. Persson, F. Johansson, M. Zwahlen, G. von Heijne, J. Nielsen, and F. Ponten. Proteomics. tissue- based map of the human proteome. Science, 347(6220):1260419, 2015.  
+
+<|ref|>text<|/ref|><|det|>[[112, 750, 884, 819]]<|/det|>
+[65] Laneve P, Gioia U, Ragno R, Altieri F, Di Franco C, Santini T, Arceci M, Bozzoni I, and Caffarelli E. The tumor marker human placental protein 11 is an endoribonuclease. The Journal of biological chemistry, 283(50):34712- 9, 2008.  
+
+<|ref|>text<|/ref|><|det|>[[112, 832, 884, 903]]<|/det|>
+[66] Waterhouse A, Bertoni M, Bienert S, Studer G, Tauriello G, Gumienny R, Heer FT, de Beer TAP, Rempfer C, Bordoli L, Lepore R, and Schwede T. SWISS- MODEL: homology modelling of protein structures and complexes. Nucleic acids research, 46(W1):W296- W303, 2018.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 90, 884, 135]]<|/det|>
+[67] Svergun D, Barberato C, and Koch MHJ. CRYSOL- a program to evaluate X- ray solution scattering of biological macromolecules from atomic coordinates. Journal of applied crystallography, 28:768- 773, 1995.  
+
+<|ref|>text<|/ref|><|det|>[[112, 148, 884, 217]]<|/det|>
+[68] C. M. Cuchillo, X. Pares, A. Guasch, T. Barman, F. Travers, and M. V. Nogues. The role of 2',3'- cyclic phosphodiesters in the bovine pancreatic ribonuclease a catalysed cleavage of rna: intermediates or products? FEBS Lett, 333(3):207- 10, 1993.  
+
+<|ref|>text<|/ref|><|det|>[[112, 230, 884, 275]]<|/det|>
+[69] W. Yang, J. Y. Lee, and M. Nowotny. Making and breaking nucleic acids: two- mg2+- ion catalysis and substrate specificity. Mol Cell, 22(1):5- 13, 2006.  
+
+<|ref|>text<|/ref|><|det|>[[112, 288, 884, 357]]<|/det|>
+[70] Qi W, Gromoff EDV, Xu F, Zhao Q, Yang W, Pfeifer D, Maier W, Long L, and Baumeister R. The secreted endoribonuclease ENDU- 2 from the soma protects germline immortality in C. elegans. Nature communications, 12(1):1262, 2021.  
+
+<|ref|>text<|/ref|><|det|>[[112, 370, 883, 415]]<|/det|>
+[71] Bowman AB and Aschner M. Considerations on manganese (Mn) treatments for in vitro studies. Neurotoxicology, 41:141- 2, 2014.  
+
+<|ref|>text<|/ref|><|det|>[[112, 428, 883, 474]]<|/det|>
+[72] Rehnberg GL, Hein JF, Carter SD, and Laskey JW. Chronic manganese oxide administration to preweanling rats: manganese accumulation and distribution. Journal of toxicology and environmental health, 6(1):217- 26, 1980.  
+
+<|ref|>text<|/ref|><|det|>[[112, 486, 884, 531]]<|/det|>
+[73] Keen CL, Ensunsa JL, and Clegg MS. Manganese metabolism in animals and humans including the toxicity of manganese. Metal ions in biological systems, 37:89- 121, 2000.  
+
+<|ref|>text<|/ref|><|det|>[[112, 544, 884, 589]]<|/det|>
+[74] D. Jenne and K. K. Stanley. Nucleotide sequence and organization of the human s- protein gene: repeating peptide motifs in the "pexin" family and a model for their evolution. Biochemistry, 26(21):6735- 42, 1987.  
+
+<|ref|>text<|/ref|><|det|>[[112, 601, 884, 671]]<|/det|>
+[75] A. Alquraini, S. Garguilo, G. D'Souza, L. X. Zhang, T. A. Schmidt, G. D. Jay, and K. A. Elsaid. The interaction of lubricin/proteoglycan 4 (prg4) with toll- like receptors 2 and 4: an anti- inflammatory role of prg4 in synovial fluid. Arthritis Res Ther, 17:353, 2015.  
+
+<|ref|>text<|/ref|><|det|>[[112, 684, 884, 729]]<|/det|>
+[76] J. W. Goding, B. Grobben, and H. Slegers. Physiological and pathophysiological functions of the ecto- nucleotide pyrophosphatase/phosphodiesterase family. Biochim Biophys Acta, 1638(1):1- 19, 2003.  
+
+<|ref|>text<|/ref|><|det|>[[112, 741, 884, 835]]<|/det|>
+[77] I. A. Droujinine, A. S. Meyer, D. Wang, N. D. Udeshi, Y. Hu, D. Rocco, J. A. McMahon, R. Yang, J. Guo, L. Mu, D. K. Carey, T. Svinkina, R. Zeng, T. Branon, A. Tabatabai, J. A. Bosch, J. M. Asara, A. Y. Ting, S. A. Carr, A. P. McMahon, and N. Perrimon. Proteomics of protein trafficking by in vivo tissue- specific labeling. Nat Commun, 12(1):2382, 2021.  
+
+<|ref|>text<|/ref|><|det|>[[112, 848, 884, 893]]<|/det|>
+[78] M. D. Bootman and G. Bultynck. Fundamentals of cellular calcium signaling: A primer. Cold Spring Harb Perspect Biol, 12(1), 2020.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 90, 884, 186]]<|/det|>
+[79] Pillon MC, Frazier MN, Dillard LB, Williams JG, Kocaman S, Krahn JM, Perera L, Hayne CK, Gordon J, Stewart ZD, Sobhany M, Deterding LJ, Hsu AL, Dandey VP, Borgnia MJ, and Stanley RE. Cryo- EM structures of the SARS- CoV- 2 endoribonuclease Nsp15 reveal insight into nuclease specificity and dynamics. Nature communications, 12(1):636, 2021.  
+
+<|ref|>text<|/ref|><|det|>[[112, 199, 884, 268]]<|/det|>
+[80] Kim Y, Wower J, Maltseva N, Chang C, Jedrzejczak R, Wilamowski M, Kang S, Nicolaescu V, Randall G, Michalska K, and Joachimiak A. Tipiracil binds to uridine site and inhibits Nsp15 endoribonuclease NendoU from SARS- CoV- 2. Communications biology, 4(1):193, 2021.  
+
+<|ref|>text<|/ref|><|det|>[[112, 281, 884, 325]]<|/det|>
+[81] Lapidoth G, Khersonsky O, Lipsh R, Dym O, Albeck S, Rogotner S, and Fleishman SJ. Highly active enzymes by automated combinatorial backbone assembly and sequence design. Nature communications, 9(1):2780, 2018.  
+
+<|ref|>text<|/ref|><|det|>[[112, 339, 884, 382]]<|/det|>
+[82] Lipsh- Sokolik R, Khersonsky O, Schröder SP, de Boer C, Hoch SY, Davies GJ, Overkleeft HS, and Fleishman SJ. Combinatorial assembly and design of enzymes. Science (New York, N.Y.), 379(6628):195- 201, 2023.  
+
+<|ref|>text<|/ref|><|det|>[[112, 396, 884, 440]]<|/det|>
+[83] Rapp JT, Bremer BJ, and Romero PA. Self- driving laboratories to autonomously navigate the protein fitness landscape. Nature chemical engineering, 1:97- 107, 2024.  
+
+<|ref|>text<|/ref|><|det|>[[112, 453, 884, 497]]<|/det|>
+[84] Luscombe NM, Laskowski RA, and Thornton JM. NUCPLOT: a program to generate schematic diagrams of protein- nucleic acid interactions. Nucleic acids research, 25(24):4940- 5, 1997.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[112, 87, 880, 450]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 462, 882, 645]]<|/det|>
+<center>Figure 1: Mouse EndoU is expressed during thymocyte development and has a calcium-activated RNase activity. (A) EndoU expression levels in developing thymocyte populations. The color indicates fractional expression of the mRNA across 211 measured cell types. Data from the Immunological Genome Project [63]. (B) RNase activity in WT and EndoU KO cell lysates. Cytoplasmic lysates from the indicated WT, KO or rescue cell lines were incubated +/- 5 mM \(\mathrm{Ca^{2 + }}\) for 15 minutes at \(37^{\circ}\mathrm{C}\) . RNA was extracted, 5 μg were run on an 8 % urea-PAGE gel, and visualized by SYBR Green II. WT-HA denotes a WT EndoU rescue construct with a C-terminal HA tag. (C) Immunoprecipitated EndoU cleavage activity on a defined RNA substrate. (D) Mutation of presumed catalytic site residues abolishes EndoU cleavage activity. (E) EndoU RNase activity is specifically activated by \(\mathrm{Ca^{2 + }}\) ions. (F) Domain structure of EndoU and homologs. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[118, 87, 884, 830]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[114, 841, 883, 970]]<|/det|>
+<center>Figure 2: Crystal structure of calcium-activated EndoU. (A) Overview of calcium-activated EndoU structure. Calcium binding site (1) to (4) as refered in the text are indicated. (B) Close-up view of EndoU calcium binding sites. Residues coordinating calcium directly (pink) or indirectly through water-mediated contacts (cyan) are highlighted. (C) Intramolecular signaling between calcium binding site (1) and remote catalytic residues. Catalytic residue (yellow) are highlighted, along with residues coordinating calcium directly (pink) or indirectly (cyan). (D) Structural change upon EndoU allosteric activation by calcium. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[113, 213, 883, 626]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 636, 883, 782]]<|/det|>
+<center>Figure 3: Enzymatic activity of EndoU and its variants. (A) RNA degradation assays. Comparison of mutants for calcium-binding sites (magenta), the bridging residue E290 (cyan), and catalytic residues (yellow) with wild-type EndoU over a 2 hrs degradation assay. (B) Enzymatic progress curve. Example of fit for a first-order reaction model \(A \times \exp (-k \times t)\) with wild-type EndoU. (C) Relative reaction rates of EndoU variants compared to wild-type. (D) Calcium binding to EndoU monitored through Molecular Dynamics. Each plot displays the distance between the E284 side-chain carboxylate and a calcium ion throughout the simulation. Five calcium ions were introduced in the simulation box. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 88, 875, 850]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 861, 884, 971]]<|/det|>
+<center>Figure 4: NMR mapping of the RNA binding interface on calcium-activated EndoU. Overlay of (A) \(^{15}\mathrm{N}\) or (B) \(^{13}\mathrm{C}\) SOFAST-HMQC spectra from isolated \(^{13}\mathrm{C}^{15}\mathrm{N}\) -EndoU (350 \(\mu \mathrm{M}\) , green) and upon successive additions of 2' fluorinated RNA. (C) Combined \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) chemical shift perturbations between calcium-activated \(^{13}\mathrm{C}^{15}\mathrm{N}\) -EndoU and in complex with 2'F RNA. (D) RNA binding interface mapping on calcium-activated EndoU. (E) Electrostatic surface potential of calcium-activated EndoU. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 265, 888, 675]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 689, 882, 727]]<|/det|>
+<center>Figure 5: Schematic representation of eukaryotic EndoU activation upon calcium and substrate binding. </center>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[43, 43, 312, 71]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[42, 93, 768, 114]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 131, 298, 178]]<|/det|>
+- Fig3noncroppedgels.pdf- Malardetal2024SI.pdf
+
+<--- Page Split --->

preprint/preprint__7eb5a732be038ff17c0e485a80b7f3520a7fb9da7418a97bea42bb9db384224f/images_list.json

@@ -0,0 +1,77 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 | Phenomenon description and test setup. a, Stages of a typical contact electrification cycle between insulators in a homogeneous gas with constant pressure: 1) Electrically neutral surfaces forced into intimate",
+    "footnote": [],
+    "bbox": [
+      [
+        90,
+        42,
+        926,
+        320
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2 | Test strategies. a, Pseudo-constant-pressure tests. b, Pseudo-constant-distance tests. The contact electrification (state 1) practically involves multiple contact cycles to deposit an adequate amount of surface charge, and state 1 represents the contact phase in the final cycle. In a, from state 1 to 2 the gap is first raised to overcome Van der Waals adhesion and then reduced, which may also apply to b depending on the target gap size at state 2. Two paths are implemented in each case, starting with contact electrification under low and high gas pressures, respectively, where partial breakdown discharge is inevitably present in contact cycles at high pressure.",
+    "footnote": [],
+    "bbox": [
+      [
+        88,
+        45,
+        510,
+        636
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 | Measured breakdown voltage of nitrogen between electrified PDMS and acrylic surfaces. a,",
+    "footnote": [],
+    "bbox": [
+      [
+        87,
+        40,
+        512,
+        670
+      ]
+    ],
+    "page_idx": 14
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 | Effective secondary-electron-emission coefficient of PDMS surface under nitrogen cation bombardments at room temperature \\(20^{\\circ}\\mathrm{C}\\) , estimated from test results of nitrogen breakdown voltage between a PDMS-acrylic contact pair.",
+    "footnote": [],
+    "bbox": [
+      [
+        92,
+        49,
+        506,
+        243
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 | Surface charge accumulation in PDMS-acrylic contact electrification cycles. a, Time history of",
+    "footnote": [],
+    "bbox": [
+      [
+        80,
+        42,
+        510,
+        525
+      ]
+    ],
+    "page_idx": 16
+  }
+]

preprint/preprint__7eb5a732be038ff17c0e485a80b7f3520a7fb9da7418a97bea42bb9db384224f/preprint__7eb5a732be038ff17c0e485a80b7f3520a7fb9da7418a97bea42bb9db384224f.mmd

@@ -0,0 +1,291 @@
+
+# Measuring gas discharge in contact electrification  
+
+Hongcheng Tao  https://orcid.org/0000- 0002- 0730- 669X  James Gibert ( jgibert@purdue.edu )  Purdue University  https://orcid.org/0000- 0002- 1429- 5378  
+
+Physical Sciences - Article  
+
+Keywords:  
+
+Posted Date: June 19th, 2023  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 2973930/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 7th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 43721- 1.
+
+<--- Page Split --->
+
+## Measuring gas discharge in contact electrification  
+
+Hongcheng Tao \(^{1}\) , James Gibert \(^{1}\)  
+
+\(^{1}\) School of Mechanical Engineering, Purdue University, West Lafayette, IN, USA.  
+
+Contact electrification in a gas medium is usually followed by partial surface charge dissipation caused by gas breakdown triggered during separation. It is widely assumed that such discharge obeys the classical Paschen's law, which describes the general dependence of breakdown voltage on the product of gas pressure and gap distance. However, quantification of this relationship in contact electrification involving insulators is impeded by challenges in nondestructive in situ measurement of the gap voltage. The present work proposes and implements an electrode- free strategy for capturing discrete discharge events by monitoring the gap voltage via Coulomb force, providing experimental evidence for a Paschen- type behavior for nitrogen breakdown between a silicone- acrylic contact pair. The method offers an alternative approach for characterizing either the ionization energies of gases or the secondary- electron- emission properties of surfaces without the requirement of an external voltage source, which can potentially benefit applications ranging from the design of insulative materials to the development of triboelectric sensors and generators.  
+
+Contact electrification, the intriguing natural phenomenon of electric charge transfer between touching surfaces, has been studied for centuries. The underlying charging mechanism, however, remains under debate partly due to challenges in quantifying the resultant surface charge density \(^{1 - 4}\) which is potentially impeded by a stage of discharge during surface separation (Fig. 1a). At an infinitesimal gap immediately after disengaging, the surfaces possess a raw amount of opposite charge. When they continue to separate, the surface charge forms an electric field across the gap which is subsequently filled by any gaseous or liquid medium that flows in from the surroundings. As the gap voltage increases with distance, it may trigger dielectric breakdown of the medium and thus partially dissipate the surface charge \(^{5 - 8}\) . In atmospheric air, the first breakdown events usually happen within a few micrometers, thus concealing the initial charge density. In real life, while most often noticed as little shocks from a winter laundry, sparks generated by surface charge may pose fire and explosion hazards in dairy farms as well as in industrial processes involving powders and
+
+<--- Page Split --->
+
+27 fabrics \(^{9 - 11}\) . On the contrary, the lack of gas discharge in space instead causes insulative parts in satellites to break down from heavy surface charge buildup \(^{12,13}\) . Succeeding research in electrostatic generators dating back to the 1700s \(^{14}\) , the significance of gas breakdown is also acknowledged recently in energy harvesters that employ contact electrification, namely triboelectric generators \(^{15,16}\) , where it can be either a limiting factor of output performance \(^{17 - 19}\) or instead exploited as a mechanism of current \(^{20,21}\) . A comprehensive model of the gas breakdown discharge process in contact electrification is therefore desired in these scenarios and has conventionally been based on Paschen's law \(^{22,23}\) which describes the dependence of breakdown voltage \(V_{\mathrm{b}}\) on the product of gas pressure \(p\) and gap distance \(d\) as  
+
+\[V_{\mathrm{b}} = \frac{Bpd}{\ln(Apd) - \ln[\ln(1 + \gamma_{\mathrm{se}}^{-1})]}\]  
+
+where constants \(A\) and \(B\) are decided by the gas constituents, and the secondary- electron- emission coefficient \(\gamma_{\mathrm{se}}\) is also dependent on the surface materials. While Paschen's law has been widely assessed for gas discharge between electrodes with a voltage supply, its applicability to gas breakdown triggered by finite surface charge due to contact electrification, especially between insulators, lacks experimental validation. Difficulties lie in monitoring the gap voltage in situ during surface separation since the placement of electrodes connected to an external circuit may disturb the electric field by induced charge, while the electrode geometry and location may affect the accuracy of voltage measurement, regardless of surface conductivity. At the same time, the measurement of gas breakdown voltage also requires both a range typically exceeding \(1\mathrm{kV}\) and a high input impedance. The present work therefore proposes an alternative nondestructive approach similar to the setup reported in a prior work \(^{7}\) which uses Coulomb force measurements to monitor surface charge variations and thus quantify the breakdown voltage of a gas medium between electrified surfaces with respect to its pressure and the gap distance.  
+
+## Experimental approach  
+
+The test apparatus (Fig. 1b) performs contact electrification in a vacuum chamber and then measures the attractive Coulomb force between the charged surfaces when they are separated. A load cell with \(25\mathrm{N}\) capacity is mounted above the top sample surface to monitor the contact force as well as any strong adhesion
+
+<--- Page Split --->
+
+when the surfaces disengage, the majority of which is attributed to Van der Waals interactions<sup>24</sup>. A second load cell with 1 N capacity is placed beneath the lower sample surface to measure the Coulomb force, which is overloaded during the contacts as a compromise. The top and bottom sample surfaces are planar and made of PDMS (Fig. 1c) and acrylic, respectively, with an effective circular contact area of \(45.6 \mathrm{cm}^2\) (76.2 mm diameter) which is relatively large to ensure sufficient load cell resolution for capturing low- voltage gas breakdown. Two test strategies, namely pseudo- constant- pressure and pseudo- constant- distance tests, are implemented to reconstruct the presumed Paschen curve by detecting gas breakdown events when gap distance and gas pressure are varied, respectively. A pseudo- constant- pressure test simulates the general contact electrification process by separating charged surfaces at different controlled gas pressures (Fig. 2a). The vacuum chamber is first flushed with the operating gas, where the surfaces are brought to a significant gap distance of around \(30 \mathrm{mm}\) while the gas pressure is swept between 10 Pa and \(100 \mathrm{kPa}\) 3 times. It is assumed that the majority of any residual surface charge is dissipated during this stage, at which point the load cells are zeroed. The gap is then slowly closed until a contact force is detected and the corresponding displacement is recorded as the zero point for gap distance. The gas pressure in the chamber is thereafter lowered and kept around \(10 \mathrm{Pa}\) , where the surfaces are pressed into several quasi- static contact cycles with a controlled peak contact force of \(24 \mathrm{N}\) (5.2 kPa) until a certain amount of surface charge is deposited. The surface charge density is in general not saturated but assumed uniform, while the maximum gap distance at the separation stage of each contact cycle is kept small (less than \(2 \mathrm{mm}\) ) to avoid triggering gas breakdown, albeit ideal disengaging at exactly zero gap distance is not feasible since extra tension is required to overcome the Van der Waals adhesion. After the surfaces fully disengage at the end of the last contact cycle, they are brought back to a near- zero gap (around \(0.02 \mathrm{mm}\) ) and the gas pressure is raised and kept closely around a target value. The gap is then increased quasi- statically, during which the Coulomb attraction is monitored and post- processed to calculate the surface charge density and gap voltage. Each discontinuity (drop) in the measured voltage represents a breakdown event and is recorded as an intersection with the hypothetical Paschen curve. The first breakdown event in tests with comparatively high initial raw charge density reveals earlier sections (at smaller gaps) of the Paschen curve, and ideally the further parts (at larger gaps) can later
+
+<--- Page Split --->
+
+be covered by breakdown events that follow. In practice, the lowest measurable voltage is limited by the resolution of the load cell so that tests at various target gas pressures are performed to obtain different sections of the Paschen curve. When the target gas pressure is high, massive discharge may already happen while the pressure is being raised to the target value, in which case an alternative strategy (path 2, Fig. 2a) is employed where the contact cycles are performed at a higher gas pressure (e.g., atmospheric, around \(100\mathrm{kPa}\) ) instead. The separation stage in the contact cycles is no longer free of discharge but an adequate amount of residual surface charge can still be deposited \(^{25}\) . Similarly, the surfaces are brought to a near-zero gap after the final contact cycle and gas pressure is then lowered to the target value, followed by the same separation and Coulomb force measurement procedure.  
+
+A pseudo- constant- distance test instead fixes the gap length and records the breakdown events as the gas pressure varies (Fig. 2b). Similarly, contact electrification cycles are first performed at a low pressure around \(10\mathrm{Pa}\) . At the end of the final contact cycle, the surfaces are moved to the target gap distance and the operating gas is slowly released into the chamber while the Coulomb force is monitored so that breakdown events are indicated by discontinuities in the measurement. However, discharge detection by increasing the gas pressure may only reveal the first half (left of the minimum voltage point) of the Paschen curve since intersections with the second half is impossible once the gap voltage falls below the minimum breakdown threshold. An alternative strategy (path 2, Fig. 2b) starts with contact electrification cycles at a high gas pressure (e.g., atmospheric, around \(100\mathrm{kPa}\) ) and then slowly pumps gas out of the chamber while the surfaces are kept at the target gap distance. Breakdown events represented by intersections with the second half of the Paschen curve can hence be obtained.  
+
+## Results and discussion  
+
+Test results of nitrogen at room temperature \(20^{\circ}\mathrm{C}\) are depicted in Fig. 3, where Figs. 3a and 3b demonstrate the gap voltage monitored during multiple test runs and the detected breakdown events are collected in Fig. 3c. The test conditions are described as pseudo- constantly controlled since in each pseudo- constant- pressure test the gas pressure has minor fluctuations around the target value, as labeled, while in both test strategies the true gap distance is subject to deflections of the load cells and as a consequence any discharge event
+
+<--- Page Split --->
+
+causes a reduction of Coulomb force and therefore an increase of gap distance, which is compensated for in postprocessing. In the same plots the theoretical Paschen curves for nitrogen \(^{22}\) with coefficients \(A = 8.85\) \(\mathrm{Pa}^{- 1}\mathrm{m}^{- 1}\) and \(B = 243.77 \mathrm{VPa}^{- 1}\mathrm{m}^{- 1}\) are displayed assuming two reference values 0.3 and 0.005 for \(\gamma_{\mathrm{se}}\) , since secondary- electron- emission properties of the tested PDMS surface (PDMS gains negative charge against acrylic - the polarity is determined by direct surface charge collection in room air using a brush electrode grounded through an electrometer, explained in Methods) under bombardments of nitrogen cations remain to be characterized. The assumption that \(\gamma_{\mathrm{se}}\) is invariant is also challenged since the probability of secondary electron emission from the insulator surface generally increases with the energy possessed by the incident cations \(^{26,27}\) . Assuming \(A\) and \(B\) are constant, effective values of \(\gamma_{\mathrm{se}}\) can be calculated by plugging gap voltage, gas pressure and gap distance measurements at each breakdown event in Fig. 3c into Paschen's law, as plotted in Fig. 4 with respect to the reduced electric field \(V_{\mathrm{b}} / (pd)\) which is theoretically proportional to the average energy of incident cations. It shows that \(\gamma_{\mathrm{se}}\) increases roughly with the reduced electric field beyond around \(200 \mathrm{VPa}^{- 1}\mathrm{m}^{- 1}\) , while below this level high \(\gamma_{\mathrm{se}}\) values are again observed which matches reports in literature \(^{28 - 31}\) . The increase of \(\gamma_{\mathrm{se}}\) at low incident cation energies is attributed to secondary electron emission caused by agents other than gas cations, such as photons and metastable gas molecules resulting from non- ionizing collisions between electrons and gas molecules, as illustrated in Extended Data Fig. 5. At low reduced electric fields, a greater portion of the kinetic energy that each electron gains from the electric field is devoted to the creation of metastable gas molecules and photons instead of cations, thus adding to the effective secondary electron emission for an incident cation on the negatively charged surface (PDMS). Meanwhile, the \(\gamma_{\mathrm{se}}\) values at low reduced electric fields are mostly calculated from gas breakdown events at small gaps, which reduces the diffusion of such metastable molecules and photons into the surroundings since unlike cations they are not accelerated along the electric field. Besides the deviations in \(\gamma_{\mathrm{se}}\) , discrepancies are observed in the overlapping of pseudo- constant- pressure test results where the minimum breakdown voltage appears higher in tests at low target gas pressures, which is attributed to the reduced validity of the infinite- parallel- plate assumption in calculating the gap voltage as the distance increases. At the same time, it is also expected that voltage evaluation upon the infinite- parallel- plate and uniform- surface- charge assumptions result in the
+
+<--- Page Split --->
+
+recorded breakdown voltages being lower than the actual Paschen curve since, unlike between electrodes, gas breakdown is always initiated at locations on the insulator surfaces with the highest voltage (and probably thereafter propagated over the entire area) which is above the calculated average. Besides, since all breakdown events between surfaces with finite charge occur as pulses, they are theoretically triggered at a voltage lower than what sustains a continuous current across the gap as in the case of how Paschen curves are obtained for gases between electrodes (Extended Data Fig. 4). Moreover, in pseudo- constant- pressure tests at low target pressures, minor discharge events are observed (before the first significant breakdown, labeled in Fig. 3a) at conditions relatively distant from predictions by Paschen's law, where successive drops of Coulomb force are detected at increasing voltages so that it does not indicate intersections with the first half of the Paschen curve. Discharge in contact electrification at similar conditions on the left of the Paschen curve has been reported<sup>7</sup>, as postprocessed in a separate work<sup>32</sup>. This is hypothetically attributed to the increased significance of neutral particles being sputtered or evaporated<sup>33-36</sup> from the charged surfaces and participating in the avalanches of ionizations, since an increased gap distance at a pseudo- constant (reduced) electric field promotes the population of high- energy impinging electrons or cations, while the observation that each successive minor discharge event happens at a higher voltage may again be attributed to the increased diffusion of such neutral particles into the surroundings at larger gap distances.  
+
+The above results illustrate the characterization of gas breakdown in the contact electrification of insulators. The same test strategies are applicable when one or both surfaces are conductive if the effect of induced surface charge redistribution is assumed minimal under the infinite- parallel- plate assumption. Given a surface with known secondary- electron- emission behaviors, e.g., metal electrodes, the test setup can be used to estimate the ionization energy of an operating gas, while given a known gas it can be used to quantify the secondary electron emission from a test surface if it gains negative charge upon contact with the other surface, which reduces the risk of dielectric breakdown of the surface itself (if nonconductive) as compared to a dielectric barrier discharge setup. In the presented tests, PDMS is selected as one of the surface materials for its high tendency to gain negative charge in contact, inferred by experimentally established triboelectric series<sup>37,38</sup>, as well as its appropriate stiffness which is low enough to guarantee intimate contact with the other
+
+<--- Page Split --->
+
+reasonable deformation so that gas breakdown is not triggered during the separation phase of contact electrification cycles. Moreover, the charging of PDMS is consistently efficient where fewer than 10 contact cycles against acrylic are sufficient to deposit a significant surface charge density, while it has been observed that the charging of PDMS against PTFE, which has an even higher negative charge affinity, becomes less efficient after breakdown discharge at low gas pressures. This may be attributed to surface erosion caused either by mechanical compression and friction during contacts or more likely by sputtering under the cation bombardment in the gas breakdown process, which is a potential limitation in the application of the presented test strategies. Meanwhile, effective charging of harder surfaces using the same test setup may require extra contact force, finer surface topography as well as the introduction of friction, presenting more severe challenges in the overload protection of the load cell measuring Coulomb force, especially after aforementioned surface erosion occurs. In the presented results, the selection of sample dimensions, load cell capacities as well as the target gas pressures and gap distances aims at proving the engineering feasibility of revealing the complete Paschen curve including the minimum voltage point, i.e., to exhibit a gap voltage that survives all combinations of pressure and distance, as well as an evident connection between results from paths 1 and 2 in the pseudo-constant- distance tests, starting from low and high gas pressures, respectively, without loss of data resolution. Furthermore, the amount of surface charge loss in each breakdown event is generally random, so that practically the survival gap voltage is always lower than the theoretical minimum and results from multiple test runs need to be combined to approximate its value asymptotically. Breakdown events at higher voltages in further regions of the Paschen curve can be obtained via pseudo-constant- distance tests with a larger target gap distance, or by increasing the number of contact cycles to deposit more surface charge. In this case, variations of \(\gamma_{\mathrm{se}}\) with respect to the surface charge density cannot be excluded since it is anticipated that secondary electron emission from an insulator surface with filled bands is more probable. Typically, the saturation level of surface charge density in a contact pair of materials distant on the triboelectric series can generate a Coulomb force with an order of magnitude comparable to the contact force and the Van der Waals adhesion so that practically they can be measured using the same load cell. The
+
+<--- Page Split --->
+
+presented test apparatus can therefore be used to quantify the buildup of surface charge by monitoring all surface interaction forces during contact cycles under low gas pressures with confidence that gas breakdown of Paschen (Townsend) type has not been triggered. As a brief demonstration, load cell readings in continuous contact cycles between PDMS and acrylic surfaces with a reduced effective contact area of \(15.6 \mathrm{cm}^2\) (circular, \(44.5 \mathrm{mm}\) diameter) are shown in Fig. 5. In each cycle the Coulomb force at an infinitesimal gap immediately before the surfaces engage (state 2 in Figs. 5d and 5e) is used to calculate the real- time surface charge density under the infinite- parallel- plate assumption. This may facilitate investigations of charge transfer mechanisms in contact electrification of different materials based on either a measured saturation level of charge density or the trend of accumulative charge deposition by repeated contacts.  
+
+## References  
+
+McCarty, L. S. & Whitesides, G. M. Electrostatic charging due to separation of ions at interfaces: contact electrification of ionic electrets. Angewandte Chemie International Edition 47, 2188- 2207 (2008).  
+
+Lacks, D. J. & Shinbrot, T. Long- standing and unresolved issues in triboelectric charging. Nature Reviews Chemistry 3, 465- 476 (2019).  
+
+Baytekin, H. et al. The mosaic of surface charge in contact electrification. Science 333, 308- 312 (2011).  
+
+Matsuyama, T. & Yamamoto, H. Charge- relaxation process dominates contact charging of a particle in atmospheric condition: II. The general model. Journal of Physics D: Applied Physics 30, 2170 (1997).  
+
+Lowell, J. & Rose- Innes, A. Contact electrification. Advances in Physics 29, 947- 1023 (1980).  
+
+Kwetkus, B., Sattler, K. & Siegmann, H.- C. Gas breakdown in contact electrification. Journal of Physics D: Applied Physics 25, 139 (1992).  
+
+Horn, R. G. & Smith, D. T. Contact electrification and adhesion between dissimilar materials. Science 256, 362- 364 (1992).  
+
+Camara, C. G., Escobar, J. V., Hird, J. R. & Putterman, S. J. Correlation between nanosecond X- ray flashes and stick- slip friction in peeling tape. Nature 455, 1089- 1092 (2008).  
+
+Glor, M. Ignition hazard due to static electricity in particulate processes. Powder Technology 135, 223- 233 (2003).
+
+<--- Page Split --->
+
+10 Chongqi, M., Shulin, Z. & Gu, H. Anti- static charge character of the plasma treated polyester filter fabric. Journal of Electrostatics 68, 111- 115 (2010).  
+
+11 Matsusaka, S., Maruyama, H., Matsuyama, T. & Ghadiri, M. Triboelectric charging of powders: A review. Chemical Engineering Science 65, 5781- 5807 (2010).  
+
+12 Garrett, H. B. The charging of spacecraft surfaces. Reviews of Geophysics 19, 577- 616 (1981).  
+
+13 Garrett, H. B. & Whittlesey, A. C. Spacecraft charging, an update. IEEE transactions on plasma science 28, 2017- 2028 (2000).  
+
+14 Heilbron, J. L. Electricity in the 17th and 18th centuries: A study of early modern physics. (Univ of California Press, 2022).  
+
+15 Wu, C., Wang, A. C., Ding, W., Guo, H. & Wang, Z. L. Triboelectric nanogenerator: a foundation of the energy for the new era. Advanced Energy Materials 9, 1802906 (2019).  
+
+16 Xu, W. et al. A droplet- based electricity generator with high instantaneous power density. Nature 578, 392- 396 (2020).  
+
+17 Liu, Y. et al. Quantifying contact status and the air- breakdown model of charge- excitation triboelectric nanogenerators to maximize charge density. Nature communications 11, 1- 8 (2020).  
+
+18 Wang, J. et al. Achieving ultrahigh triboelectric charge density for efficient energy harvesting. Nature communications 8, 1- 8 (2017).  
+
+19 Yang, B., Tao, X.- m. & Peng, Z.- h. Upper limits for output performance of contact- mode triboelectric nanogenerator systems. Nano Energy 57, 66- 73 (2019).  
+
+20 Liu, D. et al. A constant current triboelectric nanogenerator arising from electrostatic breakdown. Science advances 5, eaav6437 (2019).  
+
+21 Liu, D. et al. Hugely enhanced output power of direct- current triboelectric nanogenerators by using electrostatic breakdown effect. Advanced Materials Technologies 5, 2000289 (2020).  
+
+22 Lieberman, M. A. & Lichtenberg, A. J. Principles of plasma discharges and materials processing. (John Wiley & Sons, 2005).  
+
+23 Wadhwa, C. High voltage engineering. (New Age International, 2006).
+
+<--- Page Split --->
+
+24 Roberts, A. Surface charge contribution in rubber adhesion and friction. Journal of Physics D: Applied Physics 10, 1801 (1977).  
+
+25 Fu, J. et al. Achieving ultrahigh output energy density of triboelectric nanogenerators in high-pressure gas environment. Advanced Science 7, 2001757 (2020).  
+
+26 Moon, K. S., Lee, J. & Whang, K.- W. Electron ejection from MgO thin films by low energy noble gas ions: Energy dependence and initial instability of the secondary electron emission coefficient. Journal of applied physics 86, 4049- 4051 (1999).  
+
+27 Auday, G., Guillot, P. & Galy, J. Secondary emission of dielectrics used in plasma display panels. Journal of Applied Physics 88, 4871- 4874 (2000).  
+
+28 Druyvesteyn, M. & Penning, F. M. The mechanism of electrical discharges in gases of low pressure. Reviews of Modern Physics 12, 87 (1940).  
+
+29 Little, P. Secondary effects. Handbuch der Physik 4, 574- 662 (1956).  
+
+30 Phelps, A. & Petrovic, Z. L. Cold- cathode discharges and breakdown in argon: surface and gas phase production of secondary electrons. Plasma Sources Science and Technology 8, R21 (1999).  
+
+31 Chiang, C.- L. et al. Secondary electron emission characteristics of oxide electrodes in flat electron emission lamp. Aip Advances 6, 015317 (2016).  
+
+32 Matsuyama, T. in AIP Conference Proceedings. 020001 (AIP Publishing LLC).  
+
+33 Davies, D. K. & Biondi, M. A. Detection of electrode vapor between plane parallel copper electrodes prior to current amplification and breakdown in vacuum. Journal of Applied Physics 41, 88- 93 (1970).  
+
+34 Davies, D. K. & Biondi, M. A. Mechanism of dc electrical breakdown between extended electrodes in vacuum. Journal of Applied Physics 42, 3089- 3107 (1971).  
+
+35 Farrall, G. Electrical breakdown in vacuum. IEEE transactions on electrical insulation, 815- 841 (1985).  
+
+36 McCracken, G. The behaviour of surfaces under ion bombardment. Reports on Progress in Physics 38, 241 (1975).  
+
+37 Diaz, A. & Felix- Navarro, R. A semi- quantitative tribo- electric series for polymeric materials: the influence of chemical structure and properties. Journal of Electrostatics 62, 277- 290 (2004).
+
+<--- Page Split --->
+
+260 38 Zou, H. et al. Quantifying the triboelectric series. Nature communications 10, 1427 (2019).
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Fig. 1 | Phenomenon description and test setup. a, Stages of a typical contact electrification cycle between insulators in a homogeneous gas with constant pressure: 1) Electrically neutral surfaces forced into intimate </center>  
+
+Fig. 1 | Phenomenon description and test setup. a, Stages of a typical contact electrification cycle between insulators in a homogeneous gas with constant pressure: 1) Electrically neutral surfaces forced into intimate contact. 2) Surface separation immediately after disengaging, where the raw amount of surface charge is maintained while the gap is filled with surrounding gas. 3) The first gas breakdown event as the gap voltage approaches the breakdown threshold. Discharge occurs in the form of self-sustaining cascades of Townsend avalanches, partially reducing the surface charge density and thus the gap voltage. More breakdown events follow as the gap increases. b, Test apparatus in an acrylic vacuum chamber. c, Fabrication of the PDMS surface, where liquid PDMS is cast in a 3D- printed plastic mold pressed against a clear polyester sheet on a flat surface. The PDMS sample has a larger diameter (95 mm) than the acrylic disc (76.2 mm).
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Fig. 2 | Test strategies. a, Pseudo-constant-pressure tests. b, Pseudo-constant-distance tests. The contact electrification (state 1) practically involves multiple contact cycles to deposit an adequate amount of surface charge, and state 1 represents the contact phase in the final cycle. In a, from state 1 to 2 the gap is first raised to overcome Van der Waals adhesion and then reduced, which may also apply to b depending on the target gap size at state 2. Two paths are implemented in each case, starting with contact electrification under low and high gas pressures, respectively, where partial breakdown discharge is inevitably present in contact cycles at high pressure. </center>
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Fig. 3 | Measured breakdown voltage of nitrogen between electrified PDMS and acrylic surfaces. a, </center>  
+
+Pseudo- constant- pressure test results at multiple target gas pressures as labeled with tolerances. b, Pseudo- constant- distance test results at multiple target gap distances labeled with their nominal values at zero load cell deflection. c, Collected gap voltage readings at breakdown events detected from test runs in both strategies.
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Fig. 4 | Effective secondary-electron-emission coefficient of PDMS surface under nitrogen cation bombardments at room temperature \(20^{\circ}\mathrm{C}\) , estimated from test results of nitrogen breakdown voltage between a PDMS-acrylic contact pair. </center>
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Fig. 5 | Surface charge accumulation in PDMS-acrylic contact electrification cycles. a, Time history of </center>  
+
+the bottom load cell readings, positive being attraction. b, Time history of the top load cell readings, positive being attraction. c, Surface charge density estimated via Coulomb force readings at time instants of infinitesimal gap distance immediately before the surfaces engage in each cycle. d, Details of the top load cell reading in a labeled time window, with cyclic states demonstrated in e, including 1) surfaces separated to maximum distance, 2) surfaces at an infinitesimal gap immediately before engaging, 3) surfaces compressed to peak contact force and 4) surfaces experiencing peak Van der Waals adhesion immediately before disengaging.
+
+<--- Page Split --->
+
+## Sample fabrication and test setup  
+
+The bottom sample (acrylic disc) is \(76.2\mathrm{mm}\) (3 inches) in diameter and \(6.4\mathrm{mm}\) (1/4 inches) thick (commercially available, McMaster- Carr). The PDMS sample is fabricated following steps shown in Fig. 1c with details displayed in Extended Data Fig. 1a. The 3D- printed mold (with ASA filament, on Prusa MK3S) is pressed against a clear polyester sheet (Grafix, 0.18 mm thickness, cleaned with isopropyl alcohol) on a flat surface, and then degassed liquid PDMS (Sylgard 184) is poured inside via an array of channels on its bottom. The PDMS is then cured in atmosphere under room temperature for 48 hours, while the channels ventilate extra bubbles generated during the casting and curing processes. Overhang structures printed on the floor of the mold serve as buried mechanical locks that seize the cured PDMS to prevent it from peeling off the mold floor or walls under strong Van der Waals adhesion (Fig. 5) during surface separation in contact electrification cycles. The overhangs are \(3.4\mathrm{mm}\) in height and the bulk PDMS above is \(6.6\mathrm{mm}\) thick.  
+
+The test setup is illustrated in Fig. 1b along with a picture shown in Extended Data Fig. 1b. The acrylic vacuum chamber is customized (Sanatron) to be \(305\mathrm{mm}\) (12 inches) cubic with inlet and outlet connected to the gas cylinder (Indiana Oxygen, nitrogen, \(>99.998\%\) purity) and a 2- stage mechanical pump (Across International SuperVac- 5C, 5.6 cfm, mounted on a separate table for vibration isolation), respectively. Gas pressure in the chamber is measured with a Pirani gauge (Instructech Stinger CVM 211) with a log- linear output calibrated for nitrogen. The PDMS sample is mounted on the top load cell (Mark- 10, MR03- 5, 25 N capacity), whose vertical motion is driven by 2 synchronized stepper motors with a displacement resolution of \(0.02\mathrm{mm}\) , where motor heat is dissipated by conduction through the chamber body. The acrylic sample is fixed on the bottom load cell (Futek, LRF 400, 1.2 N capacity) which is mounted on the base by 4 leveling screws. The overload of the bottom load cell during the contact cycles is kept within its safety level of 200 N. The surface alignment error is calibrated to under \(0.02\mathrm{mm}\) using the leveling screws by sliding a thin strip of paper from several directions into the gap and then comparing the friction when the paper is being pulled out after the gap is closed (Extended Data Fig. 1c). Drift and hysteresis of the bottom load cell readings under gas pressure variations is shown in Extended Data Fig. 1e where the load cell is zeroed when the
+
+<--- Page Split --->
+
+surfaces are separated to a gap of \(30\mathrm{mm}\) and the gas pressure is swept from \(20\mathrm{Pa}\) to \(100\mathrm{kPa}\) for 2 cycles.  
+
+This error is not compensated for in the results reported in Fig. 3 since in pseudo- constant- distance tests the deviation is trivial at high voltages while at low voltages the gas pressure is in general lower than \(10\mathrm{kPa}\) where the load cell error is within \(0.1\mathrm{mN}\) , and in pseudo- constant- pressure tests the breakdown events under pressures higher than \(10\mathrm{kPa}\) generally occur at small gap distances where load cell error \((< 1\mathrm{mN})\) is trivial compared to the magnitude of the Coulomb forces corresponding to the breakdown voltages. Readings of 3 typical contact cycles in the charging phase of a test run are shown in Extended Data Fig. 2 for the bottom load cell, from which a linear stiffness of \(4.57\mathrm{N / mm}\) (Extended Data Fig. 2c) is estimated for deflections of the load cells to be compensated in the recorded gap distance and the calculation of gap voltage thereafter. All structural parts in the test apparatus are 3D- printed in ASA and PLA, CAD files available upon request, and the samples are kept away from any grounded conductor to prevent any induced charge from disturbing the electric field. The load cell output and stepper motor input are transmitted via a wire feedthrough on the back of the vacuum chamber, for which disturbance and noise are minimized by disabling the stepper motors 0.4 seconds before whenever a Coulomb force reading is taken. Data acquisition, visualization and test programming are integrated in a user interface on the Qt framework with its serial communication module.  
+
+## Supplementary tests in atmospheric air  
+
+Preliminary results of a supplementary test conducted in room air provide evidence for the accuracy of surface charge density estimation via Coulomb force measurement while determining the polarity of contact electrification, which is hardly feasible in the vacuum chamber. Two sample surfaces are first brought into controlled contact cycles and then transferred onto a low- capacity load cell where the Coulomb force is measured at a fixed destination gap distance, as shown in Extended Data Fig. 3. A sharp- tip electrode (brush) grounded through an electrometer (Keithley 6514) is then swept over each surface successively so that the majority of the surface charge is collected via tip breakdown by the brush and measured by the electrometer. Extended Data Fig. 3c depicts the time histories of load cell and electrometer readings for a typical test run using a PDMS- PTFE contact pair with a \(35\mathrm{- mm}\) - diameter circular effective contact area. The samples undergo 5 contact cycles with a peak force of \(4\mathrm{N}\) (4.16 kPa) and then separate to a significant gap around
+
+<--- Page Split --->
+
+50 mm to induce adequate air breakdown discharge so that the surface charge remains quasi- constant in later steps when the sample is transferred to the low- capacity load cell and the gap is closed. The surfaces are brought to a destination gap distance of \(1.5 \mathrm{mm}\) where a \(6 \mathrm{mN}\) Coulomb force is measured to give an estimated surface charge density of \(11.70 \mu \mathrm{C} / \mathrm{m}^2\) , while direct surface charge collection by the brush electrode reads \(10.7 \mathrm{nC}\) for the PDMS surface and \(- 11 \mathrm{nC}\) for the PTFE surface, yielding surface charge densities of \(11.12 \mu \mathrm{C} / \mathrm{m}^2\) and \(- 11.43 \mu \mathrm{C} / \mathrm{m}^2\) , respectively.  
+
+## Townsend process of gas breakdown between electrodes  
+
+While the methodology in the present work is ideally applicable to characterizing gas breakdown between either conductive or insulative surfaces, Paschen's law is typically examined between electrodes (conductive) connected to a voltage source. Extended Data Fig. 4 displays the classical Townsend's theory for this process to clarify corresponding discussions in the main text.  
+
+## Evaluation of gap voltage by Coulomb force  
+
+In the demonstrated tests of nitrogen breakdown between PDMS and acrylic surfaces, assuming a uniform surface charge density \(\pm \sigma\) on both samples (circular), the magnitude of the attractive Coulomb force at gap distance \(d\) is given by  
+
+\[F_{\mathrm{c}} = \frac{\sigma^{2}d}{2\epsilon_{\mathrm{gas}}}\int_{0}^{2\pi}\int_{0}^{R}\int_{0}^{R}\frac{r_{1}r_{2}}{(r_{1}^{2} + r_{2}^{2} - 2r_{1}r_{2}\cos\theta + d^{2})^{3 / 2}}\mathrm{d}r_{1}\mathrm{d}r_{2}\mathrm{d}\theta\]  
+
+where \(\epsilon_{\mathrm{gas}} \approx \epsilon_{\mathrm{vacuum}} \approx 8.85 \times 10^{- 12} \mathrm{F} \cdot \mathrm{m}^{- 1}\) is the permittivity of the operating gas and \(R\) the sample radius. Once the surface charge density is derived from the Coulomb force reading, the corresponding gap voltage is evaluated using the infinite- parallel- plate assumption so that \(V = \sigma d / \epsilon_{\mathrm{gas}}\) . This is based on the assumption that gas breakdown events are generally triggered at locations with the highest voltage, in this case the center of the discs, where the voltage across the gap along a linear path connecting the center of the two surfaces is  
+
+\[V = \frac{\sigma}{4\pi\epsilon_{\mathrm{gas}}}\int_{0}^{d}\int_{0}^{2\pi}\int_{0}^{R}\frac{hr}{(r^{2} + h^{2})^{3 / 2}}\mathrm{d}r\mathrm{d}\theta \mathrm{d}h\]  
+
+so that, for example, in the presented test runs the error of voltage estimation is kept below \(5\%\) when the gap distance is lower than \(4 \mathrm{mm}\) .
+
+<--- Page Split --->
+
+## Mechanisms of secondary electron emission  
+
+Extended Data Fig. 5 describes theoretical mechanisms of secondary electron emission from the negatively charged surface to clarify discussions in the main text regarding variations in the estimated secondary- electron- emission coefficients shown in Fig. 4.  
+
+## Data Availability  
+
+The datasets analyzed in Figs. 3 and 4 are available from the figshare repository linked to this manuscript.  
+
+## Code Availability  
+
+Source code for control, data acquisition and interface of the customized test apparatus is available at https://github.com/adamsPurdue/customized- contact- electrification- ender- futek- mark10  
+
+## Acknowledgements  
+
+The authors thank Professors Chelsea Davis, Arvind Raman and Jefferey Rhoads from Purdue University for discussions and advice. The authors thank the faculty and staff members of the Ray W. Herrick Laboratories for facilitating the experimental setup. This work is supported by the National Science Foundation under grant CMMI 1662925 and CAREER Award CMMI 2145803.  
+
+## Author Contributions  
+
+JG led the proposal of the methodology and analysis. HT implemented the experimental setup. HT and JG processed the data and prepared the manuscript.  
+
+## Competing Interest Declaration  
+
+The authors declare that they have no conflicts of interest.  
+
+## Additional Information  
+
+Correspondence and requests for materials should be addressed to JG (email: jgibert@purdue.edu).  
+
+Reprints and permissions information is available at www.nature.com/reprints
+
+<--- Page Split --->
+![PLACEHOLDER_21_0]
+  
+
+Extended Data Fig. 1 | Test setup and configuration. a, Structure of the ASA plastic mold for the PDMS sample showing overhang locks to be buried in cured PDMS. b, Picture of the test apparatus excluding the gas cylinder, vacuum pump, load cell amplifiers and data acquisition terminal. c, Calibration of surface alignment using a paper strip. d, Leveling screws connecting the bottom load cell and the base. e, Drift of bottom load cell reading with respect to gas pressure.
+
+<--- Page Split --->
+![PLACEHOLDER_22_0]
+  
+
+Extended Data Fig. 2 | Load cell deflection. a, Displacement time history of the top sample- load cell assembly. b, Time history of the bottom load cell readings. c, Estimation of load cell deflection stiffness from the linear operational region of the bottom load cell.
+
+<--- Page Split --->
+![PLACEHOLDER_23_0]
+  
+
+Extended Data Fig. 3 | Preliminary tests in room air. a, Test apparatus. b, Test steps including 1) contact electrification cycles, 2) Coulomb force measurement and 3) surface charge collection using a copper brush grounded through an electrometer. c, A typical test with time histories of load cell and electrometer readings.
+
+<--- Page Split --->
+![PLACEHOLDER_24_0]
+  
+
+Extended Data Fig. 4 | Gas breakdown between electrodes. a, Test setup with electrodes submerged in the examined gas, connected to a voltage source with continuous charge supply. b, Typical relation between the current across the gap and the voltage applied, for a fixed combination of gap distance and gas pressure, with definition of the corresponding breakdown voltage. c, General shape of Paschen's law (the Paschen curve) for a fixed combination of gas type and electrode materials, showing the dependence of breakdown voltage on the product of gap distance and gas pressure. d, Mechanism of gas breakdown explained as self- sustaining cascades of Townsend avalanches.
+
+<--- Page Split --->
+![PLACEHOLDER_25_0]
+  
+
+Extended Data Fig. 5 | Theoretical mechanisms of secondary electron emission from the negatively charged surface that contribute to sustaining Townsend avalanches in the gas breakdown during contact electrification. These include but are not limited to that by gas cations, excited or metastable gas molecules as well as photons emitted from them, fast gas molecules created by neutralization of accelerated cations near the negatively charged surface, and photons released from the positively charged surface by incident electrons, along with potential secondary electron emission from ionizing collisions between gas cations and molecules. Primary electron emission mechanisms include background radiation and potential field emission.
+
+<--- Page Split --->

preprint/preprint__7eb5a732be038ff17c0e485a80b7f3520a7fb9da7418a97bea42bb9db384224f/preprint__7eb5a732be038ff17c0e485a80b7f3520a7fb9da7418a97bea42bb9db384224f_det.mmd

@@ -0,0 +1,377 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 920, 144]]<|/det|>
+# Measuring gas discharge in contact electrification  
+
+<|ref|>text<|/ref|><|det|>[[44, 163, 567, 250]]<|/det|>
+Hongcheng Tao  https://orcid.org/0000- 0002- 0730- 669X  James Gibert ( jgibert@purdue.edu )  Purdue University  https://orcid.org/0000- 0002- 1429- 5378  
+
+<|ref|>text<|/ref|><|det|>[[44, 291, 272, 310]]<|/det|>
+Physical Sciences - Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 329, 136, 348]]<|/det|>
+Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 366, 301, 385]]<|/det|>
+Posted Date: June 19th, 2023  
+
+<|ref|>text<|/ref|><|det|>[[44, 404, 473, 423]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 2973930/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 441, 910, 484]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License.  Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 503, 530, 522]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 558, 945, 601]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on December 7th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 43721- 1.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[88, 45, 744, 75]]<|/det|>
+## Measuring gas discharge in contact electrification  
+
+<|ref|>text<|/ref|><|det|>[[90, 86, 345, 106]]<|/det|>
+Hongcheng Tao \(^{1}\) , James Gibert \(^{1}\)  
+
+<|ref|>text<|/ref|><|det|>[[90, 121, 740, 143]]<|/det|>
+\(^{1}\) School of Mechanical Engineering, Purdue University, West Lafayette, IN, USA.  
+
+<|ref|>text<|/ref|><|det|>[[85, 155, 945, 563]]<|/det|>
+Contact electrification in a gas medium is usually followed by partial surface charge dissipation caused by gas breakdown triggered during separation. It is widely assumed that such discharge obeys the classical Paschen's law, which describes the general dependence of breakdown voltage on the product of gas pressure and gap distance. However, quantification of this relationship in contact electrification involving insulators is impeded by challenges in nondestructive in situ measurement of the gap voltage. The present work proposes and implements an electrode- free strategy for capturing discrete discharge events by monitoring the gap voltage via Coulomb force, providing experimental evidence for a Paschen- type behavior for nitrogen breakdown between a silicone- acrylic contact pair. The method offers an alternative approach for characterizing either the ionization energies of gases or the secondary- electron- emission properties of surfaces without the requirement of an external voltage source, which can potentially benefit applications ranging from the design of insulative materials to the development of triboelectric sensors and generators.  
+
+<|ref|>text<|/ref|><|det|>[[85, 576, 945, 945]]<|/det|>
+Contact electrification, the intriguing natural phenomenon of electric charge transfer between touching surfaces, has been studied for centuries. The underlying charging mechanism, however, remains under debate partly due to challenges in quantifying the resultant surface charge density \(^{1 - 4}\) which is potentially impeded by a stage of discharge during surface separation (Fig. 1a). At an infinitesimal gap immediately after disengaging, the surfaces possess a raw amount of opposite charge. When they continue to separate, the surface charge forms an electric field across the gap which is subsequently filled by any gaseous or liquid medium that flows in from the surroundings. As the gap voltage increases with distance, it may trigger dielectric breakdown of the medium and thus partially dissipate the surface charge \(^{5 - 8}\) . In atmospheric air, the first breakdown events usually happen within a few micrometers, thus concealing the initial charge density. In real life, while most often noticed as little shocks from a winter laundry, sparks generated by surface charge may pose fire and explosion hazards in dairy farms as well as in industrial processes involving powders and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[88, 42, 944, 310]]<|/det|>
+27 fabrics \(^{9 - 11}\) . On the contrary, the lack of gas discharge in space instead causes insulative parts in satellites to break down from heavy surface charge buildup \(^{12,13}\) . Succeeding research in electrostatic generators dating back to the 1700s \(^{14}\) , the significance of gas breakdown is also acknowledged recently in energy harvesters that employ contact electrification, namely triboelectric generators \(^{15,16}\) , where it can be either a limiting factor of output performance \(^{17 - 19}\) or instead exploited as a mechanism of current \(^{20,21}\) . A comprehensive model of the gas breakdown discharge process in contact electrification is therefore desired in these scenarios and has conventionally been based on Paschen's law \(^{22,23}\) which describes the dependence of breakdown voltage \(V_{\mathrm{b}}\) on the product of gas pressure \(p\) and gap distance \(d\) as  
+
+<|ref|>equation<|/ref|><|det|>[[381, 325, 652, 365]]<|/det|>
+\[V_{\mathrm{b}} = \frac{Bpd}{\ln(Apd) - \ln[\ln(1 + \gamma_{\mathrm{se}}^{-1})]}\]  
+
+<|ref|>text<|/ref|><|det|>[[88, 378, 944, 784]]<|/det|>
+where constants \(A\) and \(B\) are decided by the gas constituents, and the secondary- electron- emission coefficient \(\gamma_{\mathrm{se}}\) is also dependent on the surface materials. While Paschen's law has been widely assessed for gas discharge between electrodes with a voltage supply, its applicability to gas breakdown triggered by finite surface charge due to contact electrification, especially between insulators, lacks experimental validation. Difficulties lie in monitoring the gap voltage in situ during surface separation since the placement of electrodes connected to an external circuit may disturb the electric field by induced charge, while the electrode geometry and location may affect the accuracy of voltage measurement, regardless of surface conductivity. At the same time, the measurement of gas breakdown voltage also requires both a range typically exceeding \(1\mathrm{kV}\) and a high input impedance. The present work therefore proposes an alternative nondestructive approach similar to the setup reported in a prior work \(^{7}\) which uses Coulomb force measurements to monitor surface charge variations and thus quantify the breakdown voltage of a gas medium between electrified surfaces with respect to its pressure and the gap distance.  
+
+<|ref|>sub_title<|/ref|><|det|>[[90, 799, 294, 819]]<|/det|>
+## Experimental approach  
+
+<|ref|>text<|/ref|><|det|>[[88, 832, 944, 923]]<|/det|>
+The test apparatus (Fig. 1b) performs contact electrification in a vacuum chamber and then measures the attractive Coulomb force between the charged surfaces when they are separated. A load cell with \(25\mathrm{N}\) capacity is mounted above the top sample surface to monitor the contact force as well as any strong adhesion
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[87, 46, 945, 940]]<|/det|>
+when the surfaces disengage, the majority of which is attributed to Van der Waals interactions<sup>24</sup>. A second load cell with 1 N capacity is placed beneath the lower sample surface to measure the Coulomb force, which is overloaded during the contacts as a compromise. The top and bottom sample surfaces are planar and made of PDMS (Fig. 1c) and acrylic, respectively, with an effective circular contact area of \(45.6 \mathrm{cm}^2\) (76.2 mm diameter) which is relatively large to ensure sufficient load cell resolution for capturing low- voltage gas breakdown. Two test strategies, namely pseudo- constant- pressure and pseudo- constant- distance tests, are implemented to reconstruct the presumed Paschen curve by detecting gas breakdown events when gap distance and gas pressure are varied, respectively. A pseudo- constant- pressure test simulates the general contact electrification process by separating charged surfaces at different controlled gas pressures (Fig. 2a). The vacuum chamber is first flushed with the operating gas, where the surfaces are brought to a significant gap distance of around \(30 \mathrm{mm}\) while the gas pressure is swept between 10 Pa and \(100 \mathrm{kPa}\) 3 times. It is assumed that the majority of any residual surface charge is dissipated during this stage, at which point the load cells are zeroed. The gap is then slowly closed until a contact force is detected and the corresponding displacement is recorded as the zero point for gap distance. The gas pressure in the chamber is thereafter lowered and kept around \(10 \mathrm{Pa}\) , where the surfaces are pressed into several quasi- static contact cycles with a controlled peak contact force of \(24 \mathrm{N}\) (5.2 kPa) until a certain amount of surface charge is deposited. The surface charge density is in general not saturated but assumed uniform, while the maximum gap distance at the separation stage of each contact cycle is kept small (less than \(2 \mathrm{mm}\) ) to avoid triggering gas breakdown, albeit ideal disengaging at exactly zero gap distance is not feasible since extra tension is required to overcome the Van der Waals adhesion. After the surfaces fully disengage at the end of the last contact cycle, they are brought back to a near- zero gap (around \(0.02 \mathrm{mm}\) ) and the gas pressure is raised and kept closely around a target value. The gap is then increased quasi- statically, during which the Coulomb attraction is monitored and post- processed to calculate the surface charge density and gap voltage. Each discontinuity (drop) in the measured voltage represents a breakdown event and is recorded as an intersection with the hypothetical Paschen curve. The first breakdown event in tests with comparatively high initial raw charge density reveals earlier sections (at smaller gaps) of the Paschen curve, and ideally the further parts (at larger gaps) can later
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[87, 42, 944, 344]]<|/det|>
+be covered by breakdown events that follow. In practice, the lowest measurable voltage is limited by the resolution of the load cell so that tests at various target gas pressures are performed to obtain different sections of the Paschen curve. When the target gas pressure is high, massive discharge may already happen while the pressure is being raised to the target value, in which case an alternative strategy (path 2, Fig. 2a) is employed where the contact cycles are performed at a higher gas pressure (e.g., atmospheric, around \(100\mathrm{kPa}\) ) instead. The separation stage in the contact cycles is no longer free of discharge but an adequate amount of residual surface charge can still be deposited \(^{25}\) . Similarly, the surfaces are brought to a near-zero gap after the final contact cycle and gas pressure is then lowered to the target value, followed by the same separation and Coulomb force measurement procedure.  
+
+<|ref|>text<|/ref|><|det|>[[87, 356, 945, 725]]<|/det|>
+A pseudo- constant- distance test instead fixes the gap length and records the breakdown events as the gas pressure varies (Fig. 2b). Similarly, contact electrification cycles are first performed at a low pressure around \(10\mathrm{Pa}\) . At the end of the final contact cycle, the surfaces are moved to the target gap distance and the operating gas is slowly released into the chamber while the Coulomb force is monitored so that breakdown events are indicated by discontinuities in the measurement. However, discharge detection by increasing the gas pressure may only reveal the first half (left of the minimum voltage point) of the Paschen curve since intersections with the second half is impossible once the gap voltage falls below the minimum breakdown threshold. An alternative strategy (path 2, Fig. 2b) starts with contact electrification cycles at a high gas pressure (e.g., atmospheric, around \(100\mathrm{kPa}\) ) and then slowly pumps gas out of the chamber while the surfaces are kept at the target gap distance. Breakdown events represented by intersections with the second half of the Paschen curve can hence be obtained.  
+
+<|ref|>sub_title<|/ref|><|det|>[[90, 740, 283, 758]]<|/det|>
+## Results and discussion  
+
+<|ref|>text<|/ref|><|det|>[[86, 774, 944, 936]]<|/det|>
+Test results of nitrogen at room temperature \(20^{\circ}\mathrm{C}\) are depicted in Fig. 3, where Figs. 3a and 3b demonstrate the gap voltage monitored during multiple test runs and the detected breakdown events are collected in Fig. 3c. The test conditions are described as pseudo- constantly controlled since in each pseudo- constant- pressure test the gas pressure has minor fluctuations around the target value, as labeled, while in both test strategies the true gap distance is subject to deflections of the load cells and as a consequence any discharge event
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[88, 45, 944, 949]]<|/det|>
+causes a reduction of Coulomb force and therefore an increase of gap distance, which is compensated for in postprocessing. In the same plots the theoretical Paschen curves for nitrogen \(^{22}\) with coefficients \(A = 8.85\) \(\mathrm{Pa}^{- 1}\mathrm{m}^{- 1}\) and \(B = 243.77 \mathrm{VPa}^{- 1}\mathrm{m}^{- 1}\) are displayed assuming two reference values 0.3 and 0.005 for \(\gamma_{\mathrm{se}}\) , since secondary- electron- emission properties of the tested PDMS surface (PDMS gains negative charge against acrylic - the polarity is determined by direct surface charge collection in room air using a brush electrode grounded through an electrometer, explained in Methods) under bombardments of nitrogen cations remain to be characterized. The assumption that \(\gamma_{\mathrm{se}}\) is invariant is also challenged since the probability of secondary electron emission from the insulator surface generally increases with the energy possessed by the incident cations \(^{26,27}\) . Assuming \(A\) and \(B\) are constant, effective values of \(\gamma_{\mathrm{se}}\) can be calculated by plugging gap voltage, gas pressure and gap distance measurements at each breakdown event in Fig. 3c into Paschen's law, as plotted in Fig. 4 with respect to the reduced electric field \(V_{\mathrm{b}} / (pd)\) which is theoretically proportional to the average energy of incident cations. It shows that \(\gamma_{\mathrm{se}}\) increases roughly with the reduced electric field beyond around \(200 \mathrm{VPa}^{- 1}\mathrm{m}^{- 1}\) , while below this level high \(\gamma_{\mathrm{se}}\) values are again observed which matches reports in literature \(^{28 - 31}\) . The increase of \(\gamma_{\mathrm{se}}\) at low incident cation energies is attributed to secondary electron emission caused by agents other than gas cations, such as photons and metastable gas molecules resulting from non- ionizing collisions between electrons and gas molecules, as illustrated in Extended Data Fig. 5. At low reduced electric fields, a greater portion of the kinetic energy that each electron gains from the electric field is devoted to the creation of metastable gas molecules and photons instead of cations, thus adding to the effective secondary electron emission for an incident cation on the negatively charged surface (PDMS). Meanwhile, the \(\gamma_{\mathrm{se}}\) values at low reduced electric fields are mostly calculated from gas breakdown events at small gaps, which reduces the diffusion of such metastable molecules and photons into the surroundings since unlike cations they are not accelerated along the electric field. Besides the deviations in \(\gamma_{\mathrm{se}}\) , discrepancies are observed in the overlapping of pseudo- constant- pressure test results where the minimum breakdown voltage appears higher in tests at low target gas pressures, which is attributed to the reduced validity of the infinite- parallel- plate assumption in calculating the gap voltage as the distance increases. At the same time, it is also expected that voltage evaluation upon the infinite- parallel- plate and uniform- surface- charge assumptions result in the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[88, 42, 945, 590]]<|/det|>
+recorded breakdown voltages being lower than the actual Paschen curve since, unlike between electrodes, gas breakdown is always initiated at locations on the insulator surfaces with the highest voltage (and probably thereafter propagated over the entire area) which is above the calculated average. Besides, since all breakdown events between surfaces with finite charge occur as pulses, they are theoretically triggered at a voltage lower than what sustains a continuous current across the gap as in the case of how Paschen curves are obtained for gases between electrodes (Extended Data Fig. 4). Moreover, in pseudo- constant- pressure tests at low target pressures, minor discharge events are observed (before the first significant breakdown, labeled in Fig. 3a) at conditions relatively distant from predictions by Paschen's law, where successive drops of Coulomb force are detected at increasing voltages so that it does not indicate intersections with the first half of the Paschen curve. Discharge in contact electrification at similar conditions on the left of the Paschen curve has been reported<sup>7</sup>, as postprocessed in a separate work<sup>32</sup>. This is hypothetically attributed to the increased significance of neutral particles being sputtered or evaporated<sup>33-36</sup> from the charged surfaces and participating in the avalanches of ionizations, since an increased gap distance at a pseudo- constant (reduced) electric field promotes the population of high- energy impinging electrons or cations, while the observation that each successive minor discharge event happens at a higher voltage may again be attributed to the increased diffusion of such neutral particles into the surroundings at larger gap distances.  
+
+<|ref|>text<|/ref|><|det|>[[88, 600, 945, 936]]<|/det|>
+The above results illustrate the characterization of gas breakdown in the contact electrification of insulators. The same test strategies are applicable when one or both surfaces are conductive if the effect of induced surface charge redistribution is assumed minimal under the infinite- parallel- plate assumption. Given a surface with known secondary- electron- emission behaviors, e.g., metal electrodes, the test setup can be used to estimate the ionization energy of an operating gas, while given a known gas it can be used to quantify the secondary electron emission from a test surface if it gains negative charge upon contact with the other surface, which reduces the risk of dielectric breakdown of the surface itself (if nonconductive) as compared to a dielectric barrier discharge setup. In the presented tests, PDMS is selected as one of the surface materials for its high tendency to gain negative charge in contact, inferred by experimentally established triboelectric series<sup>37,38</sup>, as well as its appropriate stiffness which is low enough to guarantee intimate contact with the other
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[88, 78, 945, 940]]<|/det|>
+reasonable deformation so that gas breakdown is not triggered during the separation phase of contact electrification cycles. Moreover, the charging of PDMS is consistently efficient where fewer than 10 contact cycles against acrylic are sufficient to deposit a significant surface charge density, while it has been observed that the charging of PDMS against PTFE, which has an even higher negative charge affinity, becomes less efficient after breakdown discharge at low gas pressures. This may be attributed to surface erosion caused either by mechanical compression and friction during contacts or more likely by sputtering under the cation bombardment in the gas breakdown process, which is a potential limitation in the application of the presented test strategies. Meanwhile, effective charging of harder surfaces using the same test setup may require extra contact force, finer surface topography as well as the introduction of friction, presenting more severe challenges in the overload protection of the load cell measuring Coulomb force, especially after aforementioned surface erosion occurs. In the presented results, the selection of sample dimensions, load cell capacities as well as the target gas pressures and gap distances aims at proving the engineering feasibility of revealing the complete Paschen curve including the minimum voltage point, i.e., to exhibit a gap voltage that survives all combinations of pressure and distance, as well as an evident connection between results from paths 1 and 2 in the pseudo-constant- distance tests, starting from low and high gas pressures, respectively, without loss of data resolution. Furthermore, the amount of surface charge loss in each breakdown event is generally random, so that practically the survival gap voltage is always lower than the theoretical minimum and results from multiple test runs need to be combined to approximate its value asymptotically. Breakdown events at higher voltages in further regions of the Paschen curve can be obtained via pseudo-constant- distance tests with a larger target gap distance, or by increasing the number of contact cycles to deposit more surface charge. In this case, variations of \(\gamma_{\mathrm{se}}\) with respect to the surface charge density cannot be excluded since it is anticipated that secondary electron emission from an insulator surface with filled bands is more probable. Typically, the saturation level of surface charge density in a contact pair of materials distant on the triboelectric series can generate a Coulomb force with an order of magnitude comparable to the contact force and the Van der Waals adhesion so that practically they can be measured using the same load cell. The
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[88, 42, 945, 344]]<|/det|>
+presented test apparatus can therefore be used to quantify the buildup of surface charge by monitoring all surface interaction forces during contact cycles under low gas pressures with confidence that gas breakdown of Paschen (Townsend) type has not been triggered. As a brief demonstration, load cell readings in continuous contact cycles between PDMS and acrylic surfaces with a reduced effective contact area of \(15.6 \mathrm{cm}^2\) (circular, \(44.5 \mathrm{mm}\) diameter) are shown in Fig. 5. In each cycle the Coulomb force at an infinitesimal gap immediately before the surfaces engage (state 2 in Figs. 5d and 5e) is used to calculate the real- time surface charge density under the infinite- parallel- plate assumption. This may facilitate investigations of charge transfer mechanisms in contact electrification of different materials based on either a measured saturation level of charge density or the trend of accumulative charge deposition by repeated contacts.  
+
+<|ref|>sub_title<|/ref|><|det|>[[90, 358, 187, 376]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[88, 390, 945, 450]]<|/det|>
+McCarty, L. S. & Whitesides, G. M. Electrostatic charging due to separation of ions at interfaces: contact electrification of ionic electrets. Angewandte Chemie International Edition 47, 2188- 2207 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[88, 459, 944, 518]]<|/det|>
+Lacks, D. J. & Shinbrot, T. Long- standing and unresolved issues in triboelectric charging. Nature Reviews Chemistry 3, 465- 476 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[88, 529, 943, 551]]<|/det|>
+Baytekin, H. et al. The mosaic of surface charge in contact electrification. Science 333, 308- 312 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[88, 561, 943, 620]]<|/det|>
+Matsuyama, T. & Yamamoto, H. Charge- relaxation process dominates contact charging of a particle in atmospheric condition: II. The general model. Journal of Physics D: Applied Physics 30, 2170 (1997).  
+
+<|ref|>text<|/ref|><|det|>[[88, 630, 872, 652]]<|/det|>
+Lowell, J. & Rose- Innes, A. Contact electrification. Advances in Physics 29, 947- 1023 (1980).  
+
+<|ref|>text<|/ref|><|det|>[[88, 663, 943, 722]]<|/det|>
+Kwetkus, B., Sattler, K. & Siegmann, H.- C. Gas breakdown in contact electrification. Journal of Physics D: Applied Physics 25, 139 (1992).  
+
+<|ref|>text<|/ref|><|det|>[[88, 733, 943, 792]]<|/det|>
+Horn, R. G. & Smith, D. T. Contact electrification and adhesion between dissimilar materials. Science 256, 362- 364 (1992).  
+
+<|ref|>text<|/ref|><|det|>[[88, 805, 943, 864]]<|/det|>
+Camara, C. G., Escobar, J. V., Hird, J. R. & Putterman, S. J. Correlation between nanosecond X- ray flashes and stick- slip friction in peeling tape. Nature 455, 1089- 1092 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[88, 875, 943, 934]]<|/det|>
+Glor, M. Ignition hazard due to static electricity in particulate processes. Powder Technology 135, 223- 233 (2003).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[30, 42, 944, 95]]<|/det|>
+10 Chongqi, M., Shulin, Z. & Gu, H. Anti- static charge character of the plasma treated polyester filter fabric. Journal of Electrostatics 68, 111- 115 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[30, 110, 944, 164]]<|/det|>
+11 Matsusaka, S., Maruyama, H., Matsuyama, T. & Ghadiri, M. Triboelectric charging of powders: A review. Chemical Engineering Science 65, 5781- 5807 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[30, 180, 875, 202]]<|/det|>
+12 Garrett, H. B. The charging of spacecraft surfaces. Reviews of Geophysics 19, 577- 616 (1981).  
+
+<|ref|>text<|/ref|><|det|>[[30, 216, 944, 270]]<|/det|>
+13 Garrett, H. B. & Whittlesey, A. C. Spacecraft charging, an update. IEEE transactions on plasma science 28, 2017- 2028 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[30, 285, 944, 340]]<|/det|>
+14 Heilbron, J. L. Electricity in the 17th and 18th centuries: A study of early modern physics. (Univ of California Press, 2022).  
+
+<|ref|>text<|/ref|><|det|>[[30, 355, 944, 410]]<|/det|>
+15 Wu, C., Wang, A. C., Ding, W., Guo, H. & Wang, Z. L. Triboelectric nanogenerator: a foundation of the energy for the new era. Advanced Energy Materials 9, 1802906 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[30, 425, 944, 479]]<|/det|>
+16 Xu, W. et al. A droplet- based electricity generator with high instantaneous power density. Nature 578, 392- 396 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[30, 494, 944, 549]]<|/det|>
+17 Liu, Y. et al. Quantifying contact status and the air- breakdown model of charge- excitation triboelectric nanogenerators to maximize charge density. Nature communications 11, 1- 8 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[30, 564, 944, 619]]<|/det|>
+18 Wang, J. et al. Achieving ultrahigh triboelectric charge density for efficient energy harvesting. Nature communications 8, 1- 8 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[30, 634, 944, 688]]<|/det|>
+19 Yang, B., Tao, X.- m. & Peng, Z.- h. Upper limits for output performance of contact- mode triboelectric nanogenerator systems. Nano Energy 57, 66- 73 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[30, 703, 944, 757]]<|/det|>
+20 Liu, D. et al. A constant current triboelectric nanogenerator arising from electrostatic breakdown. Science advances 5, eaav6437 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[30, 773, 944, 827]]<|/det|>
+21 Liu, D. et al. Hugely enhanced output power of direct- current triboelectric nanogenerators by using electrostatic breakdown effect. Advanced Materials Technologies 5, 2000289 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[30, 842, 944, 896]]<|/det|>
+22 Lieberman, M. A. & Lichtenberg, A. J. Principles of plasma discharges and materials processing. (John Wiley & Sons, 2005).  
+
+<|ref|>text<|/ref|><|det|>[[30, 911, 694, 933]]<|/det|>
+23 Wadhwa, C. High voltage engineering. (New Age International, 2006).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[77, 42, 944, 100]]<|/det|>
+24 Roberts, A. Surface charge contribution in rubber adhesion and friction. Journal of Physics D: Applied Physics 10, 1801 (1977).  
+
+<|ref|>text<|/ref|><|det|>[[77, 111, 944, 169]]<|/det|>
+25 Fu, J. et al. Achieving ultrahigh output energy density of triboelectric nanogenerators in high-pressure gas environment. Advanced Science 7, 2001757 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[77, 181, 944, 276]]<|/det|>
+26 Moon, K. S., Lee, J. & Whang, K.- W. Electron ejection from MgO thin films by low energy noble gas ions: Energy dependence and initial instability of the secondary electron emission coefficient. Journal of applied physics 86, 4049- 4051 (1999).  
+
+<|ref|>text<|/ref|><|det|>[[77, 286, 944, 345]]<|/det|>
+27 Auday, G., Guillot, P. & Galy, J. Secondary emission of dielectrics used in plasma display panels. Journal of Applied Physics 88, 4871- 4874 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[77, 355, 944, 413]]<|/det|>
+28 Druyvesteyn, M. & Penning, F. M. The mechanism of electrical discharges in gases of low pressure. Reviews of Modern Physics 12, 87 (1940).  
+
+<|ref|>text<|/ref|><|det|>[[77, 424, 675, 446]]<|/det|>
+29 Little, P. Secondary effects. Handbuch der Physik 4, 574- 662 (1956).  
+
+<|ref|>text<|/ref|><|det|>[[77, 457, 944, 515]]<|/det|>
+30 Phelps, A. & Petrovic, Z. L. Cold- cathode discharges and breakdown in argon: surface and gas phase production of secondary electrons. Plasma Sources Science and Technology 8, R21 (1999).  
+
+<|ref|>text<|/ref|><|det|>[[77, 526, 944, 585]]<|/det|>
+31 Chiang, C.- L. et al. Secondary electron emission characteristics of oxide electrodes in flat electron emission lamp. Aip Advances 6, 015317 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[77, 596, 760, 618]]<|/det|>
+32 Matsuyama, T. in AIP Conference Proceedings. 020001 (AIP Publishing LLC).  
+
+<|ref|>text<|/ref|><|det|>[[77, 629, 944, 687]]<|/det|>
+33 Davies, D. K. & Biondi, M. A. Detection of electrode vapor between plane parallel copper electrodes prior to current amplification and breakdown in vacuum. Journal of Applied Physics 41, 88- 93 (1970).  
+
+<|ref|>text<|/ref|><|det|>[[77, 698, 944, 756]]<|/det|>
+34 Davies, D. K. & Biondi, M. A. Mechanism of dc electrical breakdown between extended electrodes in vacuum. Journal of Applied Physics 42, 3089- 3107 (1971).  
+
+<|ref|>text<|/ref|><|det|>[[77, 768, 944, 826]]<|/det|>
+35 Farrall, G. Electrical breakdown in vacuum. IEEE transactions on electrical insulation, 815- 841 (1985).  
+
+<|ref|>text<|/ref|><|det|>[[77, 837, 944, 875]]<|/det|>
+36 McCracken, G. The behaviour of surfaces under ion bombardment. Reports on Progress in Physics 38, 241 (1975).  
+
+<|ref|>text<|/ref|><|det|>[[77, 877, 944, 935]]<|/det|>
+37 Diaz, A. & Felix- Navarro, R. A semi- quantitative tribo- electric series for polymeric materials: the influence of chemical structure and properties. Journal of Electrostatics 62, 277- 290 (2004).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[30, 44, 844, 63]]<|/det|>
+260 38 Zou, H. et al. Quantifying the triboelectric series. Nature communications 10, 1427 (2019).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[90, 42, 926, 320]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[88, 345, 944, 402]]<|/det|>
+<center>Fig. 1 | Phenomenon description and test setup. a, Stages of a typical contact electrification cycle between insulators in a homogeneous gas with constant pressure: 1) Electrically neutral surfaces forced into intimate </center>  
+
+<|ref|>text<|/ref|><|det|>[[87, 412, 945, 644]]<|/det|>
+Fig. 1 | Phenomenon description and test setup. a, Stages of a typical contact electrification cycle between insulators in a homogeneous gas with constant pressure: 1) Electrically neutral surfaces forced into intimate contact. 2) Surface separation immediately after disengaging, where the raw amount of surface charge is maintained while the gap is filled with surrounding gas. 3) The first gas breakdown event as the gap voltage approaches the breakdown threshold. Discharge occurs in the form of self-sustaining cascades of Townsend avalanches, partially reducing the surface charge density and thus the gap voltage. More breakdown events follow as the gap increases. b, Test apparatus in an acrylic vacuum chamber. c, Fabrication of the PDMS surface, where liquid PDMS is cast in a 3D- printed plastic mold pressed against a clear polyester sheet on a flat surface. The PDMS sample has a larger diameter (95 mm) than the acrylic disc (76.2 mm).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[88, 45, 510, 636]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[88, 655, 944, 884]]<|/det|>
+<center>Fig. 2 | Test strategies. a, Pseudo-constant-pressure tests. b, Pseudo-constant-distance tests. The contact electrification (state 1) practically involves multiple contact cycles to deposit an adequate amount of surface charge, and state 1 represents the contact phase in the final cycle. In a, from state 1 to 2 the gap is first raised to overcome Van der Waals adhesion and then reduced, which may also apply to b depending on the target gap size at state 2. Two paths are implemented in each case, starting with contact electrification under low and high gas pressures, respectively, where partial breakdown discharge is inevitably present in contact cycles at high pressure. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[87, 40, 512, 670]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[90, 688, 941, 708]]<|/det|>
+<center>Fig. 3 | Measured breakdown voltage of nitrogen between electrified PDMS and acrylic surfaces. a, </center>  
+
+<|ref|>text<|/ref|><|det|>[[88, 721, 943, 847]]<|/det|>
+Pseudo- constant- pressure test results at multiple target gas pressures as labeled with tolerances. b, Pseudo- constant- distance test results at multiple target gap distances labeled with their nominal values at zero load cell deflection. c, Collected gap voltage readings at breakdown events detected from test runs in both strategies.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[92, 49, 506, 243]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[40, 270, 941, 356]]<|/det|>
+<center>Fig. 4 | Effective secondary-electron-emission coefficient of PDMS surface under nitrogen cation bombardments at room temperature \(20^{\circ}\mathrm{C}\) , estimated from test results of nitrogen breakdown voltage between a PDMS-acrylic contact pair. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[80, 42, 510, 525]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[85, 544, 944, 570]]<|/det|>
+<center>Fig. 5 | Surface charge accumulation in PDMS-acrylic contact electrification cycles. a, Time history of </center>  
+
+<|ref|>text<|/ref|><|det|>[[84, 580, 944, 809]]<|/det|>
+the bottom load cell readings, positive being attraction. b, Time history of the top load cell readings, positive being attraction. c, Surface charge density estimated via Coulomb force readings at time instants of infinitesimal gap distance immediately before the surfaces engage in each cycle. d, Details of the top load cell reading in a labeled time window, with cyclic states demonstrated in e, including 1) surfaces separated to maximum distance, 2) surfaces at an infinitesimal gap immediately before engaging, 3) surfaces compressed to peak contact force and 4) surfaces experiencing peak Van der Waals adhesion immediately before disengaging.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[92, 84, 375, 103]]<|/det|>
+## Sample fabrication and test setup  
+
+<|ref|>text<|/ref|><|det|>[[88, 116, 945, 456]]<|/det|>
+The bottom sample (acrylic disc) is \(76.2\mathrm{mm}\) (3 inches) in diameter and \(6.4\mathrm{mm}\) (1/4 inches) thick (commercially available, McMaster- Carr). The PDMS sample is fabricated following steps shown in Fig. 1c with details displayed in Extended Data Fig. 1a. The 3D- printed mold (with ASA filament, on Prusa MK3S) is pressed against a clear polyester sheet (Grafix, 0.18 mm thickness, cleaned with isopropyl alcohol) on a flat surface, and then degassed liquid PDMS (Sylgard 184) is poured inside via an array of channels on its bottom. The PDMS is then cured in atmosphere under room temperature for 48 hours, while the channels ventilate extra bubbles generated during the casting and curing processes. Overhang structures printed on the floor of the mold serve as buried mechanical locks that seize the cured PDMS to prevent it from peeling off the mold floor or walls under strong Van der Waals adhesion (Fig. 5) during surface separation in contact electrification cycles. The overhangs are \(3.4\mathrm{mm}\) in height and the bulk PDMS above is \(6.6\mathrm{mm}\) thick.  
+
+<|ref|>text<|/ref|><|det|>[[88, 468, 945, 944]]<|/det|>
+The test setup is illustrated in Fig. 1b along with a picture shown in Extended Data Fig. 1b. The acrylic vacuum chamber is customized (Sanatron) to be \(305\mathrm{mm}\) (12 inches) cubic with inlet and outlet connected to the gas cylinder (Indiana Oxygen, nitrogen, \(>99.998\%\) purity) and a 2- stage mechanical pump (Across International SuperVac- 5C, 5.6 cfm, mounted on a separate table for vibration isolation), respectively. Gas pressure in the chamber is measured with a Pirani gauge (Instructech Stinger CVM 211) with a log- linear output calibrated for nitrogen. The PDMS sample is mounted on the top load cell (Mark- 10, MR03- 5, 25 N capacity), whose vertical motion is driven by 2 synchronized stepper motors with a displacement resolution of \(0.02\mathrm{mm}\) , where motor heat is dissipated by conduction through the chamber body. The acrylic sample is fixed on the bottom load cell (Futek, LRF 400, 1.2 N capacity) which is mounted on the base by 4 leveling screws. The overload of the bottom load cell during the contact cycles is kept within its safety level of 200 N. The surface alignment error is calibrated to under \(0.02\mathrm{mm}\) using the leveling screws by sliding a thin strip of paper from several directions into the gap and then comparing the friction when the paper is being pulled out after the gap is closed (Extended Data Fig. 1c). Drift and hysteresis of the bottom load cell readings under gas pressure variations is shown in Extended Data Fig. 1e where the load cell is zeroed when the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[88, 43, 944, 70]]<|/det|>
+surfaces are separated to a gap of \(30\mathrm{mm}\) and the gas pressure is swept from \(20\mathrm{Pa}\) to \(100\mathrm{kPa}\) for 2 cycles.  
+
+<|ref|>text<|/ref|><|det|>[[88, 80, 945, 560]]<|/det|>
+This error is not compensated for in the results reported in Fig. 3 since in pseudo- constant- distance tests the deviation is trivial at high voltages while at low voltages the gas pressure is in general lower than \(10\mathrm{kPa}\) where the load cell error is within \(0.1\mathrm{mN}\) , and in pseudo- constant- pressure tests the breakdown events under pressures higher than \(10\mathrm{kPa}\) generally occur at small gap distances where load cell error \((< 1\mathrm{mN})\) is trivial compared to the magnitude of the Coulomb forces corresponding to the breakdown voltages. Readings of 3 typical contact cycles in the charging phase of a test run are shown in Extended Data Fig. 2 for the bottom load cell, from which a linear stiffness of \(4.57\mathrm{N / mm}\) (Extended Data Fig. 2c) is estimated for deflections of the load cells to be compensated in the recorded gap distance and the calculation of gap voltage thereafter. All structural parts in the test apparatus are 3D- printed in ASA and PLA, CAD files available upon request, and the samples are kept away from any grounded conductor to prevent any induced charge from disturbing the electric field. The load cell output and stepper motor input are transmitted via a wire feedthrough on the back of the vacuum chamber, for which disturbance and noise are minimized by disabling the stepper motors 0.4 seconds before whenever a Coulomb force reading is taken. Data acquisition, visualization and test programming are integrated in a user interface on the Qt framework with its serial communication module.  
+
+<|ref|>sub_title<|/ref|><|det|>[[90, 568, 424, 587]]<|/det|>
+## Supplementary tests in atmospheric air  
+
+<|ref|>text<|/ref|><|det|>[[88, 599, 945, 937]]<|/det|>
+Preliminary results of a supplementary test conducted in room air provide evidence for the accuracy of surface charge density estimation via Coulomb force measurement while determining the polarity of contact electrification, which is hardly feasible in the vacuum chamber. Two sample surfaces are first brought into controlled contact cycles and then transferred onto a low- capacity load cell where the Coulomb force is measured at a fixed destination gap distance, as shown in Extended Data Fig. 3. A sharp- tip electrode (brush) grounded through an electrometer (Keithley 6514) is then swept over each surface successively so that the majority of the surface charge is collected via tip breakdown by the brush and measured by the electrometer. Extended Data Fig. 3c depicts the time histories of load cell and electrometer readings for a typical test run using a PDMS- PTFE contact pair with a \(35\mathrm{- mm}\) - diameter circular effective contact area. The samples undergo 5 contact cycles with a peak force of \(4\mathrm{N}\) (4.16 kPa) and then separate to a significant gap around
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[88, 43, 944, 239]]<|/det|>
+50 mm to induce adequate air breakdown discharge so that the surface charge remains quasi- constant in later steps when the sample is transferred to the low- capacity load cell and the gap is closed. The surfaces are brought to a destination gap distance of \(1.5 \mathrm{mm}\) where a \(6 \mathrm{mN}\) Coulomb force is measured to give an estimated surface charge density of \(11.70 \mu \mathrm{C} / \mathrm{m}^2\) , while direct surface charge collection by the brush electrode reads \(10.7 \mathrm{nC}\) for the PDMS surface and \(- 11 \mathrm{nC}\) for the PTFE surface, yielding surface charge densities of \(11.12 \mu \mathrm{C} / \mathrm{m}^2\) and \(- 11.43 \mu \mathrm{C} / \mathrm{m}^2\) , respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[90, 253, 563, 273]]<|/det|>
+## Townsend process of gas breakdown between electrodes  
+
+<|ref|>text<|/ref|><|det|>[[88, 287, 944, 412]]<|/det|>
+While the methodology in the present work is ideally applicable to characterizing gas breakdown between either conductive or insulative surfaces, Paschen's law is typically examined between electrodes (conductive) connected to a voltage source. Extended Data Fig. 4 displays the classical Townsend's theory for this process to clarify corresponding discussions in the main text.  
+
+<|ref|>sub_title<|/ref|><|det|>[[90, 426, 464, 447]]<|/det|>
+## Evaluation of gap voltage by Coulomb force  
+
+<|ref|>text<|/ref|><|det|>[[88, 460, 944, 553]]<|/det|>
+In the demonstrated tests of nitrogen breakdown between PDMS and acrylic surfaces, assuming a uniform surface charge density \(\pm \sigma\) on both samples (circular), the magnitude of the attractive Coulomb force at gap distance \(d\) is given by  
+
+<|ref|>equation<|/ref|><|det|>[[255, 565, 778, 612]]<|/det|>
+\[F_{\mathrm{c}} = \frac{\sigma^{2}d}{2\epsilon_{\mathrm{gas}}}\int_{0}^{2\pi}\int_{0}^{R}\int_{0}^{R}\frac{r_{1}r_{2}}{(r_{1}^{2} + r_{2}^{2} - 2r_{1}r_{2}\cos\theta + d^{2})^{3 / 2}}\mathrm{d}r_{1}\mathrm{d}r_{2}\mathrm{d}\theta\]  
+
+<|ref|>text<|/ref|><|det|>[[88, 625, 944, 793]]<|/det|>
+where \(\epsilon_{\mathrm{gas}} \approx \epsilon_{\mathrm{vacuum}} \approx 8.85 \times 10^{- 12} \mathrm{F} \cdot \mathrm{m}^{- 1}\) is the permittivity of the operating gas and \(R\) the sample radius. Once the surface charge density is derived from the Coulomb force reading, the corresponding gap voltage is evaluated using the infinite- parallel- plate assumption so that \(V = \sigma d / \epsilon_{\mathrm{gas}}\) . This is based on the assumption that gas breakdown events are generally triggered at locations with the highest voltage, in this case the center of the discs, where the voltage across the gap along a linear path connecting the center of the two surfaces is  
+
+<|ref|>equation<|/ref|><|det|>[[333, 804, 700, 850]]<|/det|>
+\[V = \frac{\sigma}{4\pi\epsilon_{\mathrm{gas}}}\int_{0}^{d}\int_{0}^{2\pi}\int_{0}^{R}\frac{hr}{(r^{2} + h^{2})^{3 / 2}}\mathrm{d}r\mathrm{d}\theta \mathrm{d}h\]  
+
+<|ref|>text<|/ref|><|det|>[[88, 865, 943, 920]]<|/det|>
+so that, for example, in the presented test runs the error of voltage estimation is kept below \(5\%\) when the gap distance is lower than \(4 \mathrm{mm}\) .
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[87, 44, 460, 63]]<|/det|>
+## Mechanisms of secondary electron emission  
+
+<|ref|>text<|/ref|><|det|>[[88, 78, 943, 168]]<|/det|>
+Extended Data Fig. 5 describes theoretical mechanisms of secondary electron emission from the negatively charged surface to clarify discussions in the main text regarding variations in the estimated secondary- electron- emission coefficients shown in Fig. 4.  
+
+<|ref|>sub_title<|/ref|><|det|>[[89, 183, 238, 202]]<|/det|>
+## Data Availability  
+
+<|ref|>text<|/ref|><|det|>[[88, 217, 927, 239]]<|/det|>
+The datasets analyzed in Figs. 3 and 4 are available from the figshare repository linked to this manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[89, 253, 241, 272]]<|/det|>
+## Code Availability  
+
+<|ref|>text<|/ref|><|det|>[[88, 287, 943, 343]]<|/det|>
+Source code for control, data acquisition and interface of the customized test apparatus is available at https://github.com/adamsPurdue/customized- contact- electrification- ender- futek- mark10  
+
+<|ref|>sub_title<|/ref|><|det|>[[89, 357, 257, 376]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[88, 391, 944, 516]]<|/det|>
+The authors thank Professors Chelsea Davis, Arvind Raman and Jefferey Rhoads from Purdue University for discussions and advice. The authors thank the faculty and staff members of the Ray W. Herrick Laboratories for facilitating the experimental setup. This work is supported by the National Science Foundation under grant CMMI 1662925 and CAREER Award CMMI 2145803.  
+
+<|ref|>sub_title<|/ref|><|det|>[[89, 531, 279, 550]]<|/det|>
+## Author Contributions  
+
+<|ref|>text<|/ref|><|det|>[[88, 565, 943, 621]]<|/det|>
+JG led the proposal of the methodology and analysis. HT implemented the experimental setup. HT and JG processed the data and prepared the manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[89, 636, 360, 655]]<|/det|>
+## Competing Interest Declaration  
+
+<|ref|>text<|/ref|><|det|>[[88, 670, 550, 690]]<|/det|>
+The authors declare that they have no conflicts of interest.  
+
+<|ref|>sub_title<|/ref|><|det|>[[89, 705, 290, 724]]<|/det|>
+## Additional Information  
+
+<|ref|>text<|/ref|><|det|>[[88, 739, 872, 760]]<|/det|>
+Correspondence and requests for materials should be addressed to JG (email: jgibert@purdue.edu).  
+
+<|ref|>text<|/ref|><|det|>[[88, 774, 708, 794]]<|/det|>
+Reprints and permissions information is available at www.nature.com/reprints
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[100, 45, 508, 388]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[37, 400, 944, 567]]<|/det|>
+Extended Data Fig. 1 | Test setup and configuration. a, Structure of the ASA plastic mold for the PDMS sample showing overhang locks to be buried in cured PDMS. b, Picture of the test apparatus excluding the gas cylinder, vacuum pump, load cell amplifiers and data acquisition terminal. c, Calibration of surface alignment using a paper strip. d, Leveling screws connecting the bottom load cell and the base. e, Drift of bottom load cell reading with respect to gas pressure.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[90, 42, 508, 260]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[35, 285, 943, 375]]<|/det|>
+Extended Data Fig. 2 | Load cell deflection. a, Displacement time history of the top sample- load cell assembly. b, Time history of the bottom load cell readings. c, Estimation of load cell deflection stiffness from the linear operational region of the bottom load cell.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[90, 49, 508, 435]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[89, 460, 943, 520]]<|/det|>
+Extended Data Fig. 3 | Preliminary tests in room air. a, Test apparatus. b, Test steps including 1) contact electrification cycles, 2) Coulomb force measurement and 3) surface charge collection using a copper brush grounded through an electrometer. c, A typical test with time histories of load cell and electrometer readings.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[90, 42, 508, 375]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[88, 395, 944, 627]]<|/det|>
+Extended Data Fig. 4 | Gas breakdown between electrodes. a, Test setup with electrodes submerged in the examined gas, connected to a voltage source with continuous charge supply. b, Typical relation between the current across the gap and the voltage applied, for a fixed combination of gap distance and gas pressure, with definition of the corresponding breakdown voltage. c, General shape of Paschen's law (the Paschen curve) for a fixed combination of gas type and electrode materials, showing the dependence of breakdown voltage on the product of gap distance and gas pressure. d, Mechanism of gas breakdown explained as self- sustaining cascades of Townsend avalanches.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[95, 52, 510, 380]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[88, 405, 944, 671]]<|/det|>
+Extended Data Fig. 5 | Theoretical mechanisms of secondary electron emission from the negatively charged surface that contribute to sustaining Townsend avalanches in the gas breakdown during contact electrification. These include but are not limited to that by gas cations, excited or metastable gas molecules as well as photons emitted from them, fast gas molecules created by neutralization of accelerated cations near the negatively charged surface, and photons released from the positively charged surface by incident electrons, along with potential secondary electron emission from ionizing collisions between gas cations and molecules. Primary electron emission mechanisms include background radiation and potential field emission.
+
+<--- Page Split --->

preprint/preprint__7eb637ef2168937a7f6622ee4f96ee67b231f78473941085abc2273f36690319/images_list.json

@@ -0,0 +1,47 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. Dissolution mechanism of \\(\\mathrm{Ca}_{\\alpha}\\) from the \\(\\mathrm{Ca_3SiO_5}\\) surface. a, c The one-dimensional free energy profiles with respect to \\(\\mathrm{CN(Ca - O_w)}\\) and \\(\\mathrm{CN(Ca - O_w)}\\) , respectively. b The twodimensional free energy surface with variables of \\(\\mathrm{CN(Ca - O_s)}\\) and \\(\\mathrm{CN(Ca - O_w)}\\) . d The configurations of the free energy minimum states on the FES and the corresponding reaction pathways. The state number, coordinates on the FES and the Helmholtz free energy values (kJ/mol) relative to state A are at the upper right. The values (kJ/mol) in red are free energy barriers and the values under the arrows in black are overall changes in free energies between two states. The yellow, blue, cyan, red and white spheres are indicted to the silicon, calcium (no bias potential), calcium (with bias potential), oxygen and hydrogen ions, respectively. For simplicity, the solute is shown in the transparent stick type.",
+    "footnote": [],
+    "bbox": [
+      [
+        130,
+        88,
+        890,
+        523
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Dissolution mechanism of \\(\\mathrm{Ca}_{\\beta}\\) from the \\(\\mathrm{Ca}_3\\mathrm{SiO}_5\\) surface with \\(\\mathrm{CN}(\\mathrm{Ca - O}_5)\\) from 5 to",
+    "footnote": [],
+    "bbox": [
+      [
+        130,
+        254,
+        888,
+        860
+      ]
+    ],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. The further dissolution mechanism of \\(\\mathrm{Ca}_{\\beta}\\) from the \\(\\mathrm{Ca}_3\\mathrm{SiO}_5\\) surface with \\(\\mathrm{CN}(\\mathrm{Ca - O}_s)\\) from 2 to 0. a The two-dimensional free energy surface with variables of \\(\\mathrm{CN}(\\mathrm{Ca - O}_s)\\) and \\(\\mathrm{CN}(\\mathrm{Ca - O_w})\\) . b The configurations of the free energy minimum states on the FES and the corresponding reaction pathways. The state number, coordinates on the FES and the Helmholtz free energy values (kJ/mol) relative to state H are at the upper right. The values (kJ/mol) in red",
+    "footnote": [],
+    "bbox": [
+      [
+        137,
+        191,
+        880,
+        710
+      ]
+    ],
+    "page_idx": 10
+  }
+]

preprint/preprint__7eb637ef2168937a7f6622ee4f96ee67b231f78473941085abc2273f36690319/preprint__7eb637ef2168937a7f6622ee4f96ee67b231f78473941085abc2273f36690319.mmd

@@ -0,0 +1,290 @@
+
+# Ab initio mechanism revealing for tricalcium silicate dissolution  
+
+Yunjian Li  University of Macau  Hui Pan  University of Macau  https://orcid.org/0000- 0002- 6515- 4970  Xing Ming  University of Macau  Zongjin Li ( zongjinli@um.edu.mo )  University of Macau  
+
+## Article  
+
+Keywords: ab initio molecular dynamics, reaction pathways, tricalcium silicate  
+
+Posted Date: November 15th, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 1066982/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 March 10th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 28932- 2.
+
+<--- Page Split --->
+
+# Ab initio mechanism revealing for tricalcium silicate dissolution  
+
+Yunjian Li \(^{a}\) , Hui Pan \(^{a,b}\) , Xing Ming \(^{a}\) , Zongjin Li \(^{a*}\)  
+
+\(^{a}\) Institute of Applied Physics and Materials Engineering, University of Macau, Macao SAR, 999078, P. R. China  
+
+\(^{b}\) Department of Physics and Chemistry, Faculty of Science and Technology, University of Macau, Macao SAR, 999078, P. R. China  
+
+Zongjin Li: E- mail: zongjini@um.edu.mo  
+
+## Abstract  
+
+Dissolution of mineral in water is ubiquitous in nature and industry, especially for the calcium silicate species. However, the behavior of such a complex chemical reaction is still unclear at atomic level. Here, we show that the ab initio molecular dynamics and metadynamics simulations enable quantitative analyses of reaction pathways, and the thermodynamics and kinetics of calcium ion dissolution from the tricalcium silicate \((\mathrm{Ca}_3\mathrm{SiO}_5)\) surface. The calcium sites with different coordination environment leads to different reaction pathways and free energy barriers. The low free energy barriers lead to that the detachment of calcium ions is a ligand exchange and auto- catalytic process. Moreover, the water adsorption, proton exchange and diffusion of water into the surface layer accelerate the leaching of calcium ions from the surface step by step. The discovery in this work thus would be a landmark for revealing the mechanism of cement hydration.
+
+<--- Page Split --->
+
+## Introduction  
+
+Exploring the kinetics of dissolution and dynamic properties at the water/solid interface on the atomic scale is of great significance to understand the natural process and instruct the industrial production at macroscopic scale. This has been at the heart of numerous research fields, such as geochemistry \(^{1,2}\) , drug release \(^{3}\) , water treatment \(^{4}\) and degradation of catalysis \(^{5}\) . Calcium silicate is an essential constituent in many natural minerals and has been used in a variety of fields from building materials \(^{6,7,8,9}\) to pharmaceutical products \(^{3,10}\) . Because of its bioactivity, biocompatibility and hydraulic nature, it is also a candidate for drug delivery \(^{11,12}\) , filling and regeneration material in dentistry \(^{13,14}\) and bone tissue \(^{15}\) . Above all, its application in cement is of great interest due to huge amount of usage in world widely. Tricalcium silicate (Ca \(_3\) SiO \(_5\) ) is the main and most reactive calcium silicate species in ordinary Portland cement (OPC) \(^{6}\) . It is well known that the cement hydration is stimulated by the dissolution of calcium ions from the Ca \(_3\) SiO \(_5\) surfaces accompanied by the precipitation of lamellar calcium-silicate-hydrate (C- S- H), which is responsible for the cohesivity, durability and mechanical properties of concrete \(^{16}\) .  
+
+The Ca \(_3\) SiO \(_5\) hydration exhibits clear stages (initial, induction, acceleration and deceleration) and is governed by multiple coupled parameters diverging in different time scales (from fs to years) and space scales (from nanoscale to macroscale), which is extremely complex to depict precisely. The experimental studies found that during the dissolution process the surface topography undergoes a complicated transformation with the formation of etch pits, point defects and screw dislocation \(^{17}\) . Besides, the hydrated silicate species above the surfaces reconstruct with the remaining Ca ions after the detachment of Ca ions \(^{18,19}\) . In general, the dissolution rate is well accepted to be affected by the grain particles size, overall reactive surface area, temperature, components of solution and dislocations on the solid surface \(^{20}\) on the macroscopic scale. Alongside these, the global dissolution rate is also controlled by the slowest step, which depends on the individual stage during reaction. However, the case would
+
+<--- Page Split --->
+
+be more intricate for \(\mathrm{Ca}_3\mathrm{SiO}_5\) due to the coupling effect with the precipitation of hydrate product. It has been observed the dissolution rate of \(\mathrm{Ca}_3\mathrm{SiO}_5\) is extremely fast initially and then decreases dramatically from the highest to the lowest<sup>21</sup>. The reasons for this phenomenon are still on debate. Firstly, the hydroxylation prior to dissolution may stabilize the surface and therefore lower the solubility of \(\mathrm{Ca}_3\mathrm{SiO}_5\) , as is the case for other minerals<sup>20</sup>. Furthermore, the dissolution theory<sup>17</sup> implies the driving force for the initial swift dissolution rate is the high degrees of undersaturation as it is energetically favorable for etch pits to form. When the composition of the solution is very close to the solubility equilibrium of \(\mathrm{Ca}_3\mathrm{SiO}_5\) , the etch pits no longer form and even step retreat, thus limiting the dissolution rate rather severely<sup>22</sup>, like the natural weathering<sup>23</sup> and other mineral hydration<sup>23</sup>. Moreover, there may also be an electrical double layer<sup>24</sup> and a metastable hydrate phase barrier<sup>21</sup> formed on the \(\mathrm{Ca}_3\mathrm{SiO}_5\) surface. All these hypotheses and the fit of the calorimetric curve of \(\mathrm{Ca}_3\mathrm{SiO}_5\) hydration using thermodynamic calculations<sup>25</sup> may be derived from the ignorance on the interfacial reactions during the \(\mathrm{Ca}_3\mathrm{SiO}_5\) dissolution, especially the dissolution behavior of calcium ions at atomic level.  
+
+Fortunately, atomistic simulations can tackle these problems. The previous density functional theory (DFT) - based geometry optimization calculations<sup>26</sup> indicated the adsorption of water on the Ca ion impairs the bonds strength between the calcium and oxygen ions on the surface. Claverie et al.<sup>27</sup> investigated the proton transfer at the water/Ca<sub>3</sub>SiO<sub>5</sub> interface using ab initio molecular dynamics (AIMD) simulations and found that the hydroxides formed on the surface are highly stable. However, they did not observe an obvious vertical displacement of Ca ions relative to the initial position. In fact, it is very hard to probe a complete calcium dissolution process at the atomic level even using the traditional molecular dynamics (MD) simulations<sup>28, 29</sup> with large timescale (i.e. nanoseconds). Moreover, the classical MD is not appropriate to simulate the chemical reaction involving the breakage and formation of bonds.
+
+<--- Page Split --->
+
+Even using the ReaxFF force field is still not preciously enough to present the reaction pathways for a chemical reaction. Hence, it is indispensable to give an ab initio description of such a fundamental reaction.  
+
+Here, we uncover, for the first time, the dissolution mechanism of \(\mathrm{Ca_3SiO_5}\) at early stage with ab initio method. We calculated the reaction pathways, free energy changes and free energy barriers of \(\mathrm{Ca_3SiO_5}\) dissolution using the ab initio metadynamics simulations. We show that the calcium dissolution at different sites have different reaction pathways and the less coordinated Ca is easier to escape from the surface. Besides, we found that water molecules can reduce the dissolution free energy barriers not only by attractive effect through adsorbing on Ca, but also by repulsive effect through proton penetrating into the surface and water diffusion to the original Ca site. Our findings pave the way to the atomistic understanding of surface reaction for the Ca dissolution from \(\mathrm{Ca_3SiO_5}\) .  
+
+## Results  
+
+Dissolution pathways for \(\mathrm{Ca_a}\) . The chemical reaction in initial \(\mathrm{Ca_3SiO_5}\) hydration, especially the dissolution of Ca ions, is a process of breaking the old \(\mathrm{Ca - O_s}\) bonds and forming new \(\mathrm{Ca - O_w}\) bonds. Therefore, we probe into the coordination environment of Ca to calculate the full dissolution pathways. There are four Ca sites in different chemical environment on the \(\mathrm{Ca_3SiO_5}\) (111) surface and generally they can be classified into three- and five- coordinated Ca species, which are indicated by \(\mathrm{Ca_a}\) and \(\mathrm{Ca_B}\) in this work, respectively. The \(\mathrm{Ca_3SiO_5}\) dissolution rate at different surface site (i.e. flat, step and kink site) is typically different due to different coordination environments around Ca, which changes the dissolution pathways as well as the thermodynamic and kinetic properties. Thus, we would investigate the dissolution mechanism for both \(\mathrm{Ca_a}\) and \(\mathrm{Ca_B}\) .  
+
+For the dissolution of \(\mathrm{Ca_a}\) , we can clearly identify six free energy minima on the two-
+
+<--- Page Split --->
+
+dimensional (2D) free energy surface (FES) (Figure 1b). In addition, the free energy basins and free energy barriers along any one of collective variables (CVs) can also be found through the one- dimensional (1D) FES projected from the 2D FES (Figures 1a and c). When the water encounters with the \(\mathrm{Ca_3SiO_5}\) surface, the stable state changes from (3, 0) (the state before water contacting the substrate) to A(3, 2), indicating two water molecules adsorb on \(\mathrm{Ca}_{\alpha}\) and make the system more stable. After crossing two energy barriers \((\Delta \mathrm{A}^* (\mathrm{A - B}) = 3.57 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}^* (\mathrm{B - C})) = 11.76 \mathrm{kJ / mol}\) ), the system comes to the most stable state C(3, 4) with up to four adsorbed water molecules. This high- coordination (seven- coordinated) state compromises the breakage of the original \(\mathrm{Ca - O_s}\) bond but associated with huge free energy barriers \((\Delta \mathrm{A}^* (\mathrm{C - D})) = 15.71 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}^* (\mathrm{D - E})) = 14.28 \mathrm{kJ / mol}\) and a little increase in free energy changes \((\Delta \mathrm{A}(\mathrm{C - D} = 4.69 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}(\mathrm{D - E}) = 3.48 \mathrm{kJ / mol})\) . The breakage of the \(\mathrm{Ca - O}\) bond with the oxygen ion from silicate is earlier, but more difficult than that with the interstitial oxygen ion (from the state C to D to E). These two sequential steps of breaking \(\mathrm{Ca - O_s}\) bonds decrease the total coordination number from seven to five, making this detached and free \(\mathrm{Ca}\) ion have more chances to accommodate one more water ligand and reform an octahedral structure, albeit at this stage is severely distorted with a trigonal bipyramid \((\mathrm{D_{3h}})\) structure. The free energy barrier and the free energy change for these steps \((\Delta \mathrm{A}^* (\mathrm{E - F}) = 8.70 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}(\mathrm{E - F}) = - 3.87 \mathrm{kJ / mol})\) are relatively high compared to the same fivefold to sixfold coordination transition step (A- B).
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1. Dissolution mechanism of \(\mathrm{Ca}_{\alpha}\) from the \(\mathrm{Ca_3SiO_5}\) surface. a, c The one-dimensional free energy profiles with respect to \(\mathrm{CN(Ca - O_w)}\) and \(\mathrm{CN(Ca - O_w)}\) , respectively. b The twodimensional free energy surface with variables of \(\mathrm{CN(Ca - O_s)}\) and \(\mathrm{CN(Ca - O_w)}\) . d The configurations of the free energy minimum states on the FES and the corresponding reaction pathways. The state number, coordinates on the FES and the Helmholtz free energy values (kJ/mol) relative to state A are at the upper right. The values (kJ/mol) in red are free energy barriers and the values under the arrows in black are overall changes in free energies between two states. The yellow, blue, cyan, red and white spheres are indicted to the silicon, calcium (no bias potential), calcium (with bias potential), oxygen and hydrogen ions, respectively. For simplicity, the solute is shown in the transparent stick type. </center>
+
+<--- Page Split --->
+
+Dissolution pathways for \(\mathbf{Ca}_{\beta}\) with CN(Ca- O) from 5 to 2. The surface reactivity varies with the different sites on the \(\mathrm{Ca_3SiO_5}\) surface and the hydration process is closely correlated with the coordination number of surface Ca ions<sup>29</sup>. Thus, we carried out the comparative WT- MetaD simulations for \(\mathrm{Ca}_{\beta}\) to investigate whether the dissolution pathway alters with initial coordination environment. Obviously, the FES for the detachment of \(\mathrm{Ca}_{\beta}\) is different from that for \(\mathrm{Ca}_{\alpha}\) (Figures 2a, b and c) and becomes more complicated with more possible reaction pathways. When water molecules come to the \(\mathrm{Ca_3SiO_5}\) surface, the first stable state is A(5, 1). Albeit the number of water molecule is less than that for \(\mathrm{Ca}_{\alpha}\) , the total coordination number is same with six. However, the reaction pathway for the next step is more complex. The state A has two potential reaction paths to adsorb more water molecules. The first one adsorbs one more water molecule directly without breaking the \(\mathrm{Ca - O_5}\) bond (A- B). While the other one does it by breaking two \(\mathrm{Ca - O_5}\) bonds at the same time (A- D). From a thermodynamic point of view, it is more energetically favourable to pass through state B first due to the larger free energy change between the state A and the second free energy minimum. Nevertheless, from a kinetic point of view, it is rapider to react along the pathway of A- D because of its lower free energy barriers \((\Delta \mathrm{A}^+ (\mathrm{A} - \mathrm{D}) = 6.49 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}^+ (\mathrm{A} - \mathrm{B}) = 7.01 \mathrm{kJ / mol}\) ). The reaction pathway of A- B- C- D is similar to the B- C- D- E for the dissolution of \(\mathrm{Ca}_{\alpha}\) , in which the start of \(\mathrm{Ca - O_5}\) bond cleavage is from the sevenfold coordination state. Noticeably, the seven- coordinated species is an essential intermediate in the dissolution of \(\mathrm{Ca}\) , which is similar to the decomposition of \(\gamma\) - \(\mathrm{Al}_2\mathrm{O}_3^{30}\) . But the free energy barriers along the B- C- D \((\Delta \mathrm{A}^+ (\mathrm{B} - \mathrm{C}) = 7.38 \mathrm{kJ / mol}\) , \(\Delta \mathrm{A}^+ (\mathrm{C} - \mathrm{D}) = 4.24 \mathrm{kJ / mol}\) ) is much smaller than that along C- D- E for \(\mathrm{Ca}_{\alpha}\) . The state D(3, 2) is the most stable state with the same coordinate of the start point, A, on the FES of \(\mathrm{Ca}_{\alpha}\) and is also a new outset for other two different reaction pathways towards breaking one more \(\mathrm{Ca - O_5}\) bond. The reaction would be more possible to proceed in the direction from D to E(2, 2) due to the lower free energy barrier \((\Delta \mathrm{A}^+ (\mathrm{D} - \mathrm{E}) = 19.60 \mathrm{kJ / mol}\) ) compared to the route from D to G(2, 3) \((\Delta \mathrm{A}^+ (\mathrm{D} - \mathrm{E})\)
+
+<--- Page Split --->
+
+\(= 29.02 \mathrm{kJ / mol}\) , \(\Delta \mathrm{A}^{*}(\mathrm{E - F}) = 26.11 \mathrm{kJ / mol})\) ) and it is the rate- controlling step among all the reactions. It should be noted that the further \(\mathrm{Ca - O}_5\) bond cleavage from 2 to 0 is not accessible during this simulation due to the large free energy barriers required. Therefore, to uncover the subsequent dissolution mechanism of \(\mathrm{Ca}\) with less than two- and even zero- coordinated \(\mathrm{O}_5\) , it is necessary to add a 'wall' to constrain the CVs in the region of interest.  
+
+![](images/Figure_2.jpg)
+
+<center>Figure 2. Dissolution mechanism of \(\mathrm{Ca}_{\beta}\) from the \(\mathrm{Ca}_3\mathrm{SiO}_5\) surface with \(\mathrm{CN}(\mathrm{Ca - O}_5)\) from 5 to </center>
+
+<--- Page Split --->
+
+2. a, c The one-dimensional free energy profiles with respect to \(\mathrm{CN(Ca - O_s)}\) and \(\mathrm{CN(Ca - O_w)}\) , respectively. b The two-dimensional free energy surface with variables of \(\mathrm{CN(Ca - O_s)}\) and \(\mathrm{CN(Ca - O_w)}\) . d The overhead sketches for the first layer of the \(\mathrm{Ca_3SiO_5}\) surface of the free energy minimum states on the FES (the all-atom configurations is presented in Supplementary Figure 1) and the corresponding reaction pathways. The state number, coordinates on the FES and the Helmholtz free energy values (kJ/mol) relative to state A are presented around the corresponding structure. The values (kJ/mol) in red are free energy barriers and the values under the arrows in black are overall changes in free energies between two states. The biased \(\mathrm{Ca_B}\) is bolded, and the \(\mathrm{O_s}\) and \(\mathrm{O_w}\) bonded with \(\mathrm{Ca_B}\) were highlighted in red and blue, respectively.  
+
+Complete dissolution of \(\mathrm{Ca_B}\) with \(\mathrm{CN(Ca - O_s)}\) from 2 to 0. For the further detachment of \(\mathrm{Ca_B}\) with \(\mathrm{CN(Ca - O_s)}\) from \(2\) to \(0\) , we assume that the start point is the most stable state \(\mathrm{H}(1,4)\) on the FES (Figure 3a) as it is most possible to exist in reality. To further dissolve, \(\mathrm{Ca_B}\) needs to guest a water molecule first to achieve an octahedral structure crossing over a \(26.99\mathrm{kJ / mol}\) free energy barrier and coming to the state \(\mathrm{I}(1,5)\) (Figure 3b). The next step of welcoming one more water molecule from the state \(\mathrm{I}\) to \(\mathrm{K}(1,6)\) is the rate- controlling step due to the highest free energy barrier of \(28.08\mathrm{kJ / mol}\) . After that, \(\mathrm{Ca_B}\) detaches from the original position progressively and finally gets rid of the confinement of \(\mathrm{O_s}\) network totally, which is hydrated by the surrounding water molecules to the six- or seven- coordinated solute ion. The five- , six- and seven- coordinated \(\mathrm{Ca_B}\) ion can be transformed to each other. However, the sixfold coordination state processes the greatest possibility, not only because the reactions from the fivefold state \(\mathrm{M}(0,5)\) and sevenfold state \(\mathrm{L}(0,7)\) to the sixfold state \(\mathrm{K}(0,6)\) are spontaneous \((\Delta \mathrm{A}(\mathrm{M - K}) = - 2.09\mathrm{kJ / mol}\) and \(\Delta \mathrm{A}(\mathrm{L - K}) = - 10.30\mathrm{kJ / mol})\) , but also the free energy barriers \((\Delta \mathrm{A}^*(\mathrm{M - K}) = 10.34\mathrm{kJ / mol}\) , \(\Delta \mathrm{A}^*(\mathrm{L - K}) = 6.96\mathrm{kJ / mol})\) are lower than those of the reverse reactions, which is confirmed by a further \(30\mathrm{ps}\) equilibrium AIMD simulations. The AIMD also show that the
+
+<--- Page Split --->
+
+coordinated \(\mathrm{O_w}\) with \(\mathrm{Ca}_{\beta}\) increased slightly and the dissolved Ca forms a more regular octahedral structure with water and hydroxyl with the evolution of time.  
+
+![](images/Figure_3.jpg)
+
+<center>Figure 3. The further dissolution mechanism of \(\mathrm{Ca}_{\beta}\) from the \(\mathrm{Ca}_3\mathrm{SiO}_5\) surface with \(\mathrm{CN}(\mathrm{Ca - O}_s)\) from 2 to 0. a The two-dimensional free energy surface with variables of \(\mathrm{CN}(\mathrm{Ca - O}_s)\) and \(\mathrm{CN}(\mathrm{Ca - O_w})\) . b The configurations of the free energy minimum states on the FES and the corresponding reaction pathways. The state number, coordinates on the FES and the Helmholtz free energy values (kJ/mol) relative to state H are at the upper right. The values (kJ/mol) in red </center>
+
+<--- Page Split --->
+
+are free energy barriers and the values under the arrows in black are overall changes in free energies between two states.  
+
+Structural and dynamic analysis of the water/Ca \(_3\) SiO \(_5\) interface after the dissolution of the Ca ion. According to the atomic density and atomic excess profiles (Figure 4a), we define six regions for the water structure along the z direction on the Ca \(_3\) SiO \(_5\) surface. It is obvious that the H ion penetrates into the second layer of the surface ( \(z = 6 \mathring{\mathrm{A}}\) ) due to the escape of the Ca ion and the proton exchange, which is the first time for such a phenomenon in the hydration of Ca \(_3\) SiO \(_5\) to be observed by the AIMD simulation. The region II is the chemisorbed water molecule and III is a mixture of the physisorbed water molecule for the surface and the first hydration shell of the dissolved Ca, which expands the destruction area of the layered water structure compared to the surface before dissolution. The region IV, V and VI are the transition layer, bulk layer and water/vacuum layer, respectively, which are similar to those on the perfect Ca \(_3\) SiO \(_5\) surface. The radial distribution function (RDF) (Figure 3b) shows that the first and second hydration shells of the dissolved Ca are more than that of the surface Ca. It also presents that the Ca- O \(_w\) bond length for the dissolved Ca is \(2.39 \mathring{\mathrm{A}}\) , which is shorter than that for surface Ca with \(2.50 \mathring{\mathrm{A}}\) , indicating a stronger interaction between Ca and the water molecule after dissolving. The mean square displacement (MSD) (Figure 3c) presents a more dynamic property and stronger diffusion ability for the dissolved Ca compared to the previous surface state. We also calculated the infrared (IR) spectra for the system and extract the parts for Si- OH and Ca- OH. It clearly shows one band at \(916 \mathrm{cm}^{- 1}\) (Figure 4d), which is characteristic for the Si- OH bond raised in experimental IR spectra upon Ca \(_3\) SiO \(_5\) hydration \(^{19}\) . In addition, two bands arising at \(700 - 1000 \mathrm{cm}^{- 1}\) and \(3640 \mathrm{cm}^{- 1}\) (Figure 4e) shows the formation of Ca- OH during the dissolution of Ca ion as indicated by the experimental results \(^{19,31}\) . From the transmission electron microscopy (TEM) photograph (figure 4f), we can clear see when Ca \(_3\) SiO \(_5\) encounters
+
+<--- Page Split --->
+
+with water, the Ca cations are released from the \(\mathrm{Ca_3SiO_5}\) surface and form the \(\mathrm{Ca(OH)_2}\) in the aqueous solution. An atomistic interpretation of this natural phenomenon can be obtained from our AIMD simulation, which shows the Ca ion coordinated with five water molecules and one hydroxyl group releases from the \(\mathrm{Ca_3SiO_5}\) surface following the proton transfer from the water molecule to the second layer of the interstitial oxygen ion and the occupation of the initial Ca site by one water molecule. This water molecule is parallel to the surface and form hydrogen bond with the new formed O- H and \(\mathrm{O_s}\) .
+
+<--- Page Split --->
+![PLACEHOLDER_13_0]
+
+
+<--- Page Split --->
+
+Figure 4. a Atomic density for \(\mathrm{O_w}\) and H, and atomic excess profiles as a function of the height beginning at \(6\mathrm{\AA}\) from the bottom of the \(\mathrm{Ca_3SiO_5}\) surface. b Radial distribution function (RDF) between the dissolved Ca ion ( \(\mathrm{Ca_d}\) ) and \(\mathrm{O_w}\) as well as between the surface Ca ion in the same site ( \(\mathrm{Ca_s}\) ) and \(\mathrm{O_w}\) . c The mean square displacement (MSD) evolution of \(\mathrm{Ca_d}\) and \(\mathrm{Ca_s}\) . d, e IR spectrums for Si- OH group and Ca- OH group, respectively. f TEM image of the dissolution of Ca from the \(\mathrm{Ca_3SiO_5}\) surface and the snapshot of the equilibrium AIMD at \(30\mathrm{ps}\) . The composition and the electron diffraction patterns of \(\mathrm{Ca(OH)_2}\) refers to \(^{32}\)  
+
+## Discussion  
+
+The calcium sites with different coordination environments lead to different reaction pathways and free energy barriers. The dissolution of three- coordinated \(\mathrm{Ca_d}\) is easier than the five- coordinated \(\mathrm{Ca_{\beta}}\) not only because of its initial less restraint from the \(\mathrm{Ca_3SiO_5}\) surface, but also the smaller free energy barriers along the reaction path. In addition, The free energy barriers between the two stable states on either FES of \(\mathrm{Ca_d}\) and \(\mathrm{Ca_{\beta}}\) tends to be larger as the \(\mathrm{Ca - O_s}\) bonds become less, which means the water adsorption on the \(\mathrm{Ca_3SiO_5}\) surface is easier than the detachment of Ca and the kinetic rate decreases gradually as this process proceeds. Nonetheless even the highest free energy barrier is only \(29.02\mathrm{kJ / mol}\) , which is easy to be crossed, suggesting the dissolution is an auto- catalytic process. It should be noted that there is no human intervention during the simulation of crossing over the free energy barriers, which ensures the reliability of our results. Thus, the results obtained in this work are general and applicable to other kinds of calcium silicate species.  
+
+Our ab initio WT- MetaD simulations with explicit water solvation highlight the importance of the water molecule on the detachment of Ca during the \(\mathrm{Ca_3SiO_5}\) hydration. In short, the dissolution of Ca can be explained in terms of a ligand exchange process. It initially stimulated by the water adsorption, raising the total coordination number with oxygen ions to
+
+<--- Page Split --->
+
+a high level (five to seven). The adsorbed water molecules can reduce the free energy barriers for breaking the Ca- O<sub>s</sub> bond, and thus provide an opportunity for Ca to break the Ca- O<sub>s</sub> bonds with the optimal six coordination number unchanged. In fact, the breakage of Ca- O<sub>s</sub> bonds is a multi- step and multi- orientation chemical reaction, and every step needs a relatively high free energy barrier, which is straight to cross for traditional AIMD simulations within 100 ps but easy in reality. In addition, the dissolution of Ca is further promoted by the proton exchange and the diffusion of water molecule from the chemisorbed layer into the second surface layer. On the one hand, the H ion penetrate into the second layer of the surface and bonds to the interstitial oxygen ion previously bonded to the dissolved Ca ion, leading to a repulsive force pushing Ca out of the surface. At the same time, this diffused water molecule resides in the position of the Ca ion before dissolution and forms the hydron bond network with O<sub>s</sub>, O<sub>w</sub> and H, which undermines the attractive force to this Ca.  
+
+In summary, an atomistic and mechanistic picture of Ca dissolution from the Ca<sub>3</sub>SiO<sub>5</sub> surface in water solution at the initial stage of Ca<sub>3</sub>SiO<sub>5</sub> hydration is investigated using the ab initio molecular dynamics and metadynamics simulations. We find that the Ca dissolution from the Ca<sub>3</sub>SiO<sub>5</sub> are multi- step and multi- orientation chemical reactions accompanied by the water adsorption, proton exchange, breakage of Ca- O<sub>s</sub> bonds and water diffusion. The thermodynamic and kinetic analyses show that the detachment of Ca is spontaneous when the Ca is not fully dissolved and unspontaneous when Ca is no more coordinated with O<sub>s</sub>. Additionally, the Ca dissolution is an auto- catalytic process with the highest free energy barriers of only 29 kJ/mol. The reaction pathways for Ca in different coordination environments are different and the less coordinated Ca is easier to leach from the surface. Besides, we find the water molecules provide not only the attractive forces pulling Ca out of the surface, but also repulsive forces in filling the precious Ca site and pushing it away. The present achievement thus provides an insight into the cement hydration and also predict the evolution
+
+<--- Page Split --->
+
+of other complex geochemical and catalytic systems.  
+
+## Methods  
+
+## Atomistic model  
+
+The \(\mathrm{Ca_3SiO_5}\) (111) surface was cleaved from the M3 type of \(\mathrm{Ca_3SiO_5}\) (obtained from CCSD \(^{33}\) ), which is most frequently observed in industrial clinkers \(^{34}\) . The details of the DFT- based geometry optimization of the bulk crystal and surface slab were listed in Supplementary Methods. Considering the adsorption of water molecule on the \(\mathrm{Ca_3SiO_5}\) surface occurs even before contacting bulk water \(^{35}\) , we firstly adsorbed isolated water molecules to saturate the dangling bond on the \(\mathrm{Ca_3SiO_5}\) surface. Then, we put a \(20\mathrm{\AA}\) thick layer of water with density of \(1\mathrm{g / cm^3}\) on the \(\mathrm{Ca_3SiO_5}\) (111) surface (totally 333 atoms) for the dissolution simulations. The lattice parameters were \(14.21\mathrm{\AA}\times 11.72\mathrm{\AA}\times 36\mathrm{\AA}\) after setting a vacuum of \(15\mathrm{\AA}\) along z direction. After the detachment of the \(\mathrm{Ca_{\beta}}\) ion from its initial position to the state of D, we further constructed a new model ( \(14.21\mathrm{\AA}\times 11.72\mathrm{\AA}\times 48\mathrm{\AA}\) ) by adding another \(10\mathrm{\AA}\) thick layer of water on the previous system (totally 498 atoms) to calculate the structural and dynamic properties of the equilibrium state.  
+
+## AIMD simulations  
+
+All the AIMD simulations reported in this work were performed within the framework of DFT with the generalized gradient approximation (GGA) using the Perdew- Burke- Ernzerhof (PBE) \(^{36}\) functional and Grimme D3 correction \(^{37}\) , which was implemented in the CP2K/Quickstep \(^{38}\) . The Core electrons were described by Goedecker- Teter- Hutter (GTH) pseudopotentials \(^{39,40}\) and the valence electrons were described by a mixed Gaussian and plane waves basis (GPW) \(^{41}\) . The wave functions were expanded on a double- \(\zeta\) valence polarized
+
+<--- Page Split --->
+
+(DZVP) basis set along with an auxiliary plane wave basis set at a cutoff energy of 500 Ry. The Brillouin zone was sampled by the gamma approximation. During AIMD, the nuclei were treated within the Born–Oppenheimer approximation with a timestep of 0.5 fs for equilibrium simulation, while 1 fs for metadynamics simulations with the replacement of hydrogen by deuterium to accelerate the structural evolution without energy drifts<sup>27,42</sup>. The temperature was maintained at 300 K using a Nosé-Hoover thermostat<sup>43, 44</sup> coupled to the system with a time constant of 1000 fs in the Canonical ensemble (NVT). The convergence criterion for energy was set to \(10^{- 12}\) Hartree and for self- consistent field was \(10^{- 6}\) Hartree. All the system were first optimized to a stable state and then thermalized for at least 2.5 ps before the production run for statistical analysis. The duration of the AIMD simulations for dissolution of \(\mathrm{Ca}_{\alpha}\) , dissolution of \(\mathrm{Ca}_{\beta}\) with \(\mathrm{CN}(\mathrm{Ca - O}_s)\) from 5 to 2, further dissolution of \(\mathrm{Ca}_{\beta}\) with \(\mathrm{CN}(\mathrm{Ca - Os})\) from 2 to 0, equilibrium of final state of \(\mathrm{Ca}_{\beta}\) were 100 ps, 63 ps, 46ps, and 30ps, respectively.  
+
+## Metadynamics simulations  
+
+In the well- tempered ab initio metadynamics<sup>45</sup> (WT- MetaD) simulations, we utilized a two- dimensional collective variables (CVs) characterized by the coordination number (CN) to monitor the dissolution process. The \(\mathrm{CN}(\mathrm{Ca - Os})\) is the coordination number of the Ca ion with all oxygen ions from the surface slab, while \(\mathrm{CN}(\mathrm{Ca - Os})\) is the coordination number of the Ca ion. with all oxygen ions from water molecules. As defined in the PLUMED code<sup>46</sup>, the CN have the expression as follows:  
+
+\[CN(Ca,O_{w / s}) = \sum_{j\in O_{w / s}}s_{ij}(r_{ij}) = \sum_{j\in O_{w / s}}\frac{1 - \left(\frac{r_{ij} - d_0}{r_0}\right)^n}{1 - \left(\frac{r_{ij} - d_0}{r_0}\frac{m}{n}\right)^m} \quad (1)\]  
+
+where \(\mathrm{r_{ij}}\) is the distance between atom i and atom j. \(\mathrm{s_{ij}(r_{ij})}\) is a rational type of switching function describing the coordination between atom i and j. \(\mathrm{d_0}\) is the central value of the function. \(\mathrm{r_0}\) is the acceptance distance of the switching function, where the function well be \(\mathrm{n / m}\) at \(\mathrm{d_0 + r_0}\) .
+
+<--- Page Split --->
+
+Here, we define do is 2.42 Å, which is the equilibrium bond length between the Ca and O ions \(^{47}\) ; ro is 0.4 Å, which is around half of the full width at half maximum of the radial distribution function of Ca- O \(^{48}\) and n and m are 6 and 12, respectively.  
+
+The Gaussian hills was deposited every 30 timesteps with the initial height of \(3.5 \mathrm{kJ / mol}\) and width of 0.15 for both CVs. The biasfactor were 15 for simulations of \(\mathrm{Ca}_{\alpha}\) and 24 for simulation of \(\mathrm{Ca}_{\beta}\) . In addition, to further investigation of the dissolution process to a larger extent, we added a quadratic wall with the force constant of \(500 \mathrm{kJ / mol}\) at the position of \(\mathrm{CN}(\mathrm{Ca - O}_s)\) equals to 1.5 to restrict the simulation of further dissolution of \(\mathrm{Ca}_{\beta}\) on the regions of free energy surface with \(\mathrm{CN}(\mathrm{Ca - O}_s)\) less than 1.5. The time evolutions of the CV1 and CV2, the convergence tests for the free energy surfaces and the errors between the free energy minima were shown in Supplementary Figures 2- 7.  
+
+## Structural and spectroscopy calculations  
+
+The atomic excess (ae) is defined as:  
+
+\[a e = \frac{2[O w] - 2[H]}{2[O w] + 2[H]} \quad (2)\]  
+
+where \([O w]\) and \([H]\) are atomic density for Ow and H, respectively. The negative value for ae indicates an excess of Ow, while the positive one indicates an excess of H.  
+
+For the vibrational spectra, we use the last \(30 \mathrm{ps}\) equilibrium AIMD trajectory to calculate the infrared (IR) spectrum with the TRAVIS program \(^{46}\) . The molecular electric properties were calculated every 4 fs (8 timesteps) using the Voronoi integration approach \(^{49}\) . The IR spectrum of particular components of a system were computed though the Fourier transform of the molecular dipole autocorrelation function as follows:  
+
+\[A(\omega) \propto \int \langle \dot{\mu} (\tau) \dot{\mu} (t + \tau) \rangle_{\tau} e^{-i\omega t} dt \quad (3)\]  
+
+where A is the absorption cross section, \(\omega\) is frequency, and \(\dot{\mu}\) is the time derivative of the
+
+<--- Page Split --->
+
+dipole moment leading to the dipole- velocity autocorrelation function.  
+
+## Synthesis of \(\mathrm{Ca_3SiO_5}\) and characterization of dissolution of \(\mathrm{Ca_3SiO_5}\)  
+
+The synthesis of \(\mathrm{Ca_3SiO_5}\) was conducted through a traditional method \(^{50}\) . Calcium carbonate and silica with molar ratio of 3:1 was first ground to pass through a 63- \(\mu \mathrm{m}\) sieve and then evenly mixed for \(2\mathrm{h}\) . The obtained fine powder was compressed into pancakes in a lab press and put into a Pt crucible for calcination at \(1500^{\circ}\mathrm{C}\) for \(5\mathrm{h}\) , and quickly cooled down to ambient temperature within \(10\mathrm{min}\) . The obtained product was ground into fine powder, compressed and reclaimed for four repeating cycles before the production of pure \(\mathrm{Ca_3SiO_5}\) . The synthetic \(\mathrm{Ca_3SiO_5}\) was then dispersed in the water solution for hydrolysis.  
+
+The morphology of dissolution of \(\mathrm{Ca_3SiO_5}\) was characterized by transmission electron microscopy (JEOL, JEM- 2100, \(200\mathrm{kV}\) ), equipped with an energy dispersive spectroscopy system.
+
+<--- Page Split --->
+
+## References  
+
+1. Schliemann R, Churakov SV. Atomic scale mechanism of clay minerals dissolution revealed by ab initio simulations. Geochimica et Cosmochimica Acta 2021, 293: 438-460.  
+2. Ye S, Feng P, Liu Y, Liu J, Bullard JW. In situ nano-scale observation of C3A dissolution in water. Cement and Concrete Research 2020, 132: 106044.  
+3. Sato H, Miyagawa Y, Okabe T, Miyajima M, Sunada H. Dissolution mechanism of diclofenac sodium from wax matrix granules. Journal of pharmaceutical sciences 1997, 86(8): 929-934.  
+4. Pirkanniemi K, Sillanpää M. Heterogeneous water phase catalysis as an environmental application: a review. Chemosphere 2002, 48(10): 1047-1060.  
+5. Ravenelle RM, Copeland JR, Kim W-G, Crittenden JC, Sievers C. Structural changes of γ-Al2O3-supported catalysts in hot liquid water. Acs Catalysis 2011, 1(5): 552-561.  
+6. Taylor HF. Cement chemistry, vol. 2. Thomas Telford London, 1997.  
+7. Vogel W. Glass chemistry. Springer Science & Business Media, 2012.  
+8. Rodriguez J, Rodriguez M, De Aza S, Pena P. Reaction sintering of zircon-dolomite mixtures. Journal of the European Ceramic Society 2001, 21(3): 343-354.  
+9. Zheng Q, Wang W. Calcium silicate based high efficiency thermal insulation. British ceramic transactions 2000, 99(4): 187-190.  
+10. Kinoshita M, Baba K, Nagayasu A, Yamabe K, Shimooka T, Takeichi Yi, et al. Improvement of solubility and oral bioavailability of a poorly water-soluble drug, TAS-301, by its melt-adsorption on a porous calcium silicate. Journal of pharmaceutical sciences 2002, 91(2): 362-370.  
+11. Zhu Y-J, Guo X-X, Sham T-K. Calcium silicate-based drug delivery systems. Expert opinion on drug delivery 2017, 14(2): 215-228.  
+12. Jain SK, Awasthi A, Jain N, Agrawal G. Calcium silicate based microspheres of repaglinide for gastroenterite floating drug delivery: Preparation and in vitro characterization. Journal of controlled release 2005, 107(2): 300-309.  
+13. Prati C, Gandolfi MG. Calcium silicate bioactive cements: Biological perspectives and clinical applications. Dental materials 2015, 31(4): 351-370.  
+14. Xu S, Lin K, Wang Z, Chang J, Wang L, Lu J, et al. Reconstruction of calvarial defect of rabbits using porous calcium silicate bioactive ceramics. Biomaterials 2008, 29(17): 2588-2596.
+
+<--- Page Split --->
+
+15. Ding S-J, Shie M-Y, Wang C-Y. Novel fast-setting calcium silicate bone cements with high bioactivity and enhanced osteogenesis in vitro. Journal of Materials Chemistry 2009, 19(8): 1183-1190.  
+
+16. Zongjin L. Advanced concrete technology. Hoboken, New Jersey: John Wiley & Sons, Inc 2011.  
+
+17. Juilland P, Gallucci E, Flatt R, Scrivener K. Dissolution theory applied to the induction period in alite hydration. Cement and Concrete Research 2010, 40(6): 831-844.  
+
+18. Pustovgar E, Sangodkar RP, Andreev AS, Palacios M, Chmelka BF, Flatt RJ, et al. Understanding silicate hydration from quantitative analyses of hydrating tricalcium silicates. Nature communications 2016, 7(1): 1-9.  
+
+19. Higl J, Hinder D, Rathgeber C, Ramming B, Lindén M. Detailed in situ ATR-FTIR spectroscopy study of the early stages of CSH formation during hydration of monoclinic C3S. Cement and Concrete Research 2021, 142: 106367.  
+
+20. Bellmann F, Sowoidnich T, Ludwig H-M, Damidot D. Dissolution rates during the early hydration of tricalcium silicate. Cement and Concrete Research 2015, 72: 108-116.  
+
+21. Naber C, Bellmann F, Neubauer J. Influence of w/s ratio on alite dissolution and CSH precipitation rates during hydration. Cement and Concrete Research 2020, 134: 106087.  
+
+22. Nicoleau L, Bertolim MA. Analytical model for the alite (C3S) dissolution topography. Journal of the American Ceramic Society 2016, 99(3): 773-786.  
+
+23. Lasaga AC, Luttge A. Variation of crystal dissolution rate based on a dissolution stepwave model. Science 2001, 291(5512): 2400-2404.  
+
+24. Tadros M, Skalny J, Kalyoncu R. Early hydration of tricalcium silicate. Journal of the American Ceramic Society 1976, 59(7-8): 344-347.  
+
+25. Zhang Z, Han F, Yan P. Modelling the dissolution and precipitation process of the early hydration of C3S. Cement and Concrete Research 2020, 136: 106174.  
+
+26. Qi C, Spagnoli D, Fourie A. DFT-D study of single water adsorption on low-index surfaces of calcium silicate phases in cement. Applied Surface Science 2020, 518: 146255.  
+
+27. Claverie J, Bernard F, Cordeiro JMM, Kamali-Bernard S. Ab initio molecular dynamics description of proton transfer at water-tricalcium silicate interface. Cement and Concrete Research 2020, 136: 106162.  
+
+28. Claverie J, Bernard F, Cordeiro JMM, Kamali-Bernard S. Water's behaviour on Ca-rich tricalcium
+
+<--- Page Split --->
+
+silicate surfaces for various degrees of hydration: A molecular dynamics investigation. Journal of Physics and Chemistry of Solids 2019, 132: 48- 55.  
+
+29. Qi C, Manzano H, Spagnoli D, Chen Q, Fourie A. Initial hydration process of calcium silicates in Portland cement: A comprehensive comparison from molecular dynamics simulations. Cement and Concrete Research 2021, 149: 106576.  
+
+30. Reocreux R, Girel É, Clabaut P, Tuel A, Besson M, Chaumonnot A, et al. Reactivity of shape-controlled crystals and metadynamics simulations locate the weak spots of alumina in water. Nature communications 2019, 10(1): 1-8.  
+
+31. Khachani M, El Hamidi A, Halim M, Arsalan S. Non-isothermal kinetic and thermodynamic studies of the dehydroxylation process of synthetic calcium hydroxide Ca (OH) 2. J Mater Environ Sci 2014, 5(2): 615-624.  
+
+32. Sun G, Li Z, Liang R, Weng L-T, Zhang L. Super stretchable hydrogel achieved by non-aggregated spherulites with diameters < 5 nm. Nature communications 2016, 7(1): 1-8.  
+
+33. Mumme WG. CRYSTAL-STRUCTURE OF TRICALCIUM SILICATE FROM A PORTLAND-CEMENT CLINKER AND ITS APPLICATION TO QUANTITATIVE XRD ANALYSIS. Neues Jahrbuch Fur Mineralogie-Monatshefte 1995(4): 145-160.  
+
+34. De Noirfontaine M-N, Dunstetter F, Courtial M, Gasecki G, Signes-Frehel M. Polymorphism of tricalcium silicate, the major compound of Portland cement clinker: 2. Modelling alite for Rietveld analysis, an industrial challenge. Cement and concrete research 2006, 36(1): 54-64.  
+
+35. Thissen P, Natzeck C, Giraudo N, Weidler P, Woll C. Hydration of Concrete: The First Steps. Chemistry-A European Journal 2018, 24(34): 8603-8608.  
+
+36. Perdew JP, Burke K, Ernzerhof M. Generalized gradient approximation made simple. Physical review letters 1996, 77(18): 3865.  
+
+37. 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. The Journal of chemical physics 2010, 132(15): 154104.  
+
+38. VandeVondele J, Krack M, Mohamed F, Parrinello M, Chassaing T, Hutter J. Quickstep: Fast and accurate density functional calculations using a mixed Gaussian and plane waves approach. Computer Physics Communications 2005, 167(2): 103-128.  
+
+39. Goedecker S, Teter M, Hutter J. Separable dual-space Gaussian pseudopotentials. Physical Review B 1996, 54(3): 1703.  
+
+40. Hartwigsen C, Goedecker S, Hutter J. Relativistic separable dual-space Gaussian pseudopotentials
+
+<--- Page Split --->
+
+from H to Rn. Physical Review B 1998, 58(7): 3641.  
+
+41. VandeVondele J, Hutter J. Gaussian basis sets for accurate calculations on molecular systems in gas and condensed phases. The Journal of chemical physics 2007, 127(11): 114105.  
+
+42. Leung K, Rempe SB. Ab initio rigid water: Effect on water structure, ion hydration, and thermodynamics. Physical Chemistry Chemical Physics 2006, 8(18): 2153-2162.  
+
+43. Martyna GJ, Klein ML, Tuckerman M. Nose-Hoover chains: The canonical ensemble via continuous dynamics. The Journal of chemical physics 1992, 97(4): 2635-2643.  
+
+44. Nose S. A molecular dynamics method for simulations in the canonical ensemble. Molecular physics 1984, 52(2): 255-268.  
+
+45. Bonomi M, Parrinello M. Enhanced sampling in the well-tempered ensemble. Physical review letters 2010, 104(19): 190601.  
+
+46. Brehm M, Thomas M, Gehrke S, Kirchner B. TRAVIS—A free analyzer for trajectories from molecular simulation. The Journal of chemical physics 2020, 152(16): 164105.  
+
+47. Jalilehvand F, Spangberg D, Lindqvist-Reis P, Hermansson K, Persson I, Sandström M. Hydration of the calcium ion. An EXAFS, large-angle X-ray scattering, and molecular dynamics simulation study. Journal of the American Chemical Society 2001, 123(3): 431-441.  
+
+48. Zhang W, Li S, Hou D, Geng Y, Zhang S, Yin B, et al. Study on unsaturated transport of cement-based silane sol coating materials. Coatings 2019, 9(7): 427.  
+
+49. Thomas M, Brehm M, Kirchner B. Voronoi dipole moments for the simulation of bulk phase vibrational spectra. Physical Chemistry Chemical Physics 2015, 17(5): 3207-3213.  
+
+50. Wesselsky A, Jensen OM. Synthesis of pure Portland cement phases. Cement and concrete research 2009, 39(11): 973-980.
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+Supplementaryinformation.docx VideoS1.mp4
+
+<--- Page Split --->

preprint/preprint__7eb637ef2168937a7f6622ee4f96ee67b231f78473941085abc2273f36690319/preprint__7eb637ef2168937a7f6622ee4f96ee67b231f78473941085abc2273f36690319_det.mmd

@@ -0,0 +1,373 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 953, 174]]<|/det|>
+# Ab initio mechanism revealing for tricalcium silicate dissolution  
+
+<|ref|>text<|/ref|><|det|>[[44, 196, 590, 379]]<|/det|>
+Yunjian Li  University of Macau  Hui Pan  University of Macau  https://orcid.org/0000- 0002- 6515- 4970  Xing Ming  University of Macau  Zongjin Li ( zongjinli@um.edu.mo )  University of Macau  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 418, 102, 436]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 455, 720, 475]]<|/det|>
+Keywords: ab initio molecular dynamics, reaction pathways, tricalcium silicate  
+
+<|ref|>text<|/ref|><|det|>[[44, 494, 345, 512]]<|/det|>
+Posted Date: November 15th, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 531, 474, 551]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 1066982/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 569, 910, 611]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License.  Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 648, 925, 691]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on March 10th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 28932- 2.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[118, 85, 815, 108]]<|/det|>
+# Ab initio mechanism revealing for tricalcium silicate dissolution  
+
+<|ref|>text<|/ref|><|det|>[[118, 127, 525, 145]]<|/det|>
+Yunjian Li \(^{a}\) , Hui Pan \(^{a,b}\) , Xing Ming \(^{a}\) , Zongjin Li \(^{a*}\)  
+
+<|ref|>text<|/ref|><|det|>[[118, 159, 870, 203]]<|/det|>
+\(^{a}\) Institute of Applied Physics and Materials Engineering, University of Macau, Macao SAR, 999078, P. R. China  
+
+<|ref|>text<|/ref|><|det|>[[118, 217, 870, 251]]<|/det|>
+\(^{b}\) Department of Physics and Chemistry, Faculty of Science and Technology, University of Macau, Macao SAR, 999078, P. R. China  
+
+<|ref|>text<|/ref|><|det|>[[118, 260, 422, 276]]<|/det|>
+Zongjin Li: E- mail: zongjini@um.edu.mo  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 337, 197, 353]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[115, 384, 883, 732]]<|/det|>
+Dissolution of mineral in water is ubiquitous in nature and industry, especially for the calcium silicate species. However, the behavior of such a complex chemical reaction is still unclear at atomic level. Here, we show that the ab initio molecular dynamics and metadynamics simulations enable quantitative analyses of reaction pathways, and the thermodynamics and kinetics of calcium ion dissolution from the tricalcium silicate \((\mathrm{Ca}_3\mathrm{SiO}_5)\) surface. The calcium sites with different coordination environment leads to different reaction pathways and free energy barriers. The low free energy barriers lead to that the detachment of calcium ions is a ligand exchange and auto- catalytic process. Moreover, the water adsorption, proton exchange and diffusion of water into the surface layer accelerate the leaching of calcium ions from the surface step by step. The discovery in this work thus would be a landmark for revealing the mechanism of cement hydration.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 230, 101]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[115, 130, 884, 550]]<|/det|>
+Exploring the kinetics of dissolution and dynamic properties at the water/solid interface on the atomic scale is of great significance to understand the natural process and instruct the industrial production at macroscopic scale. This has been at the heart of numerous research fields, such as geochemistry \(^{1,2}\) , drug release \(^{3}\) , water treatment \(^{4}\) and degradation of catalysis \(^{5}\) . Calcium silicate is an essential constituent in many natural minerals and has been used in a variety of fields from building materials \(^{6,7,8,9}\) to pharmaceutical products \(^{3,10}\) . Because of its bioactivity, biocompatibility and hydraulic nature, it is also a candidate for drug delivery \(^{11,12}\) , filling and regeneration material in dentistry \(^{13,14}\) and bone tissue \(^{15}\) . Above all, its application in cement is of great interest due to huge amount of usage in world widely. Tricalcium silicate (Ca \(_3\) SiO \(_5\) ) is the main and most reactive calcium silicate species in ordinary Portland cement (OPC) \(^{6}\) . It is well known that the cement hydration is stimulated by the dissolution of calcium ions from the Ca \(_3\) SiO \(_5\) surfaces accompanied by the precipitation of lamellar calcium-silicate-hydrate (C- S- H), which is responsible for the cohesivity, durability and mechanical properties of concrete \(^{16}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 559, 884, 906]]<|/det|>
+The Ca \(_3\) SiO \(_5\) hydration exhibits clear stages (initial, induction, acceleration and deceleration) and is governed by multiple coupled parameters diverging in different time scales (from fs to years) and space scales (from nanoscale to macroscale), which is extremely complex to depict precisely. The experimental studies found that during the dissolution process the surface topography undergoes a complicated transformation with the formation of etch pits, point defects and screw dislocation \(^{17}\) . Besides, the hydrated silicate species above the surfaces reconstruct with the remaining Ca ions after the detachment of Ca ions \(^{18,19}\) . In general, the dissolution rate is well accepted to be affected by the grain particles size, overall reactive surface area, temperature, components of solution and dislocations on the solid surface \(^{20}\) on the macroscopic scale. Alongside these, the global dissolution rate is also controlled by the slowest step, which depends on the individual stage during reaction. However, the case would
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 81, 884, 563]]<|/det|>
+be more intricate for \(\mathrm{Ca}_3\mathrm{SiO}_5\) due to the coupling effect with the precipitation of hydrate product. It has been observed the dissolution rate of \(\mathrm{Ca}_3\mathrm{SiO}_5\) is extremely fast initially and then decreases dramatically from the highest to the lowest<sup>21</sup>. The reasons for this phenomenon are still on debate. Firstly, the hydroxylation prior to dissolution may stabilize the surface and therefore lower the solubility of \(\mathrm{Ca}_3\mathrm{SiO}_5\) , as is the case for other minerals<sup>20</sup>. Furthermore, the dissolution theory<sup>17</sup> implies the driving force for the initial swift dissolution rate is the high degrees of undersaturation as it is energetically favorable for etch pits to form. When the composition of the solution is very close to the solubility equilibrium of \(\mathrm{Ca}_3\mathrm{SiO}_5\) , the etch pits no longer form and even step retreat, thus limiting the dissolution rate rather severely<sup>22</sup>, like the natural weathering<sup>23</sup> and other mineral hydration<sup>23</sup>. Moreover, there may also be an electrical double layer<sup>24</sup> and a metastable hydrate phase barrier<sup>21</sup> formed on the \(\mathrm{Ca}_3\mathrm{SiO}_5\) surface. All these hypotheses and the fit of the calorimetric curve of \(\mathrm{Ca}_3\mathrm{SiO}_5\) hydration using thermodynamic calculations<sup>25</sup> may be derived from the ignorance on the interfacial reactions during the \(\mathrm{Ca}_3\mathrm{SiO}_5\) dissolution, especially the dissolution behavior of calcium ions at atomic level.  
+
+<|ref|>text<|/ref|><|det|>[[115, 575, 884, 889]]<|/det|>
+Fortunately, atomistic simulations can tackle these problems. The previous density functional theory (DFT) - based geometry optimization calculations<sup>26</sup> indicated the adsorption of water on the Ca ion impairs the bonds strength between the calcium and oxygen ions on the surface. Claverie et al.<sup>27</sup> investigated the proton transfer at the water/Ca<sub>3</sub>SiO<sub>5</sub> interface using ab initio molecular dynamics (AIMD) simulations and found that the hydroxides formed on the surface are highly stable. However, they did not observe an obvious vertical displacement of Ca ions relative to the initial position. In fact, it is very hard to probe a complete calcium dissolution process at the atomic level even using the traditional molecular dynamics (MD) simulations<sup>28, 29</sup> with large timescale (i.e. nanoseconds). Moreover, the classical MD is not appropriate to simulate the chemical reaction involving the breakage and formation of bonds.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 83, 882, 167]]<|/det|>
+Even using the ReaxFF force field is still not preciously enough to present the reaction pathways for a chemical reaction. Hence, it is indispensable to give an ab initio description of such a fundamental reaction.  
+
+<|ref|>text<|/ref|><|det|>[[116, 182, 883, 465]]<|/det|>
+Here, we uncover, for the first time, the dissolution mechanism of \(\mathrm{Ca_3SiO_5}\) at early stage with ab initio method. We calculated the reaction pathways, free energy changes and free energy barriers of \(\mathrm{Ca_3SiO_5}\) dissolution using the ab initio metadynamics simulations. We show that the calcium dissolution at different sites have different reaction pathways and the less coordinated Ca is easier to escape from the surface. Besides, we found that water molecules can reduce the dissolution free energy barriers not only by attractive effect through adsorbing on Ca, but also by repulsive effect through proton penetrating into the surface and water diffusion to the original Ca site. Our findings pave the way to the atomistic understanding of surface reaction for the Ca dissolution from \(\mathrm{Ca_3SiO_5}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 495, 184, 511]]<|/det|>
+## Results  
+
+<|ref|>text<|/ref|><|det|>[[116, 541, 883, 857]]<|/det|>
+Dissolution pathways for \(\mathrm{Ca_a}\) . The chemical reaction in initial \(\mathrm{Ca_3SiO_5}\) hydration, especially the dissolution of Ca ions, is a process of breaking the old \(\mathrm{Ca - O_s}\) bonds and forming new \(\mathrm{Ca - O_w}\) bonds. Therefore, we probe into the coordination environment of Ca to calculate the full dissolution pathways. There are four Ca sites in different chemical environment on the \(\mathrm{Ca_3SiO_5}\) (111) surface and generally they can be classified into three- and five- coordinated Ca species, which are indicated by \(\mathrm{Ca_a}\) and \(\mathrm{Ca_B}\) in this work, respectively. The \(\mathrm{Ca_3SiO_5}\) dissolution rate at different surface site (i.e. flat, step and kink site) is typically different due to different coordination environments around Ca, which changes the dissolution pathways as well as the thermodynamic and kinetic properties. Thus, we would investigate the dissolution mechanism for both \(\mathrm{Ca_a}\) and \(\mathrm{Ca_B}\) .  
+
+<|ref|>text<|/ref|><|det|>[[154, 870, 878, 890]]<|/det|>
+For the dissolution of \(\mathrm{Ca_a}\) , we can clearly identify six free energy minima on the two-
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 75, 884, 700]]<|/det|>
+dimensional (2D) free energy surface (FES) (Figure 1b). In addition, the free energy basins and free energy barriers along any one of collective variables (CVs) can also be found through the one- dimensional (1D) FES projected from the 2D FES (Figures 1a and c). When the water encounters with the \(\mathrm{Ca_3SiO_5}\) surface, the stable state changes from (3, 0) (the state before water contacting the substrate) to A(3, 2), indicating two water molecules adsorb on \(\mathrm{Ca}_{\alpha}\) and make the system more stable. After crossing two energy barriers \((\Delta \mathrm{A}^* (\mathrm{A - B}) = 3.57 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}^* (\mathrm{B - C})) = 11.76 \mathrm{kJ / mol}\) ), the system comes to the most stable state C(3, 4) with up to four adsorbed water molecules. This high- coordination (seven- coordinated) state compromises the breakage of the original \(\mathrm{Ca - O_s}\) bond but associated with huge free energy barriers \((\Delta \mathrm{A}^* (\mathrm{C - D})) = 15.71 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}^* (\mathrm{D - E})) = 14.28 \mathrm{kJ / mol}\) and a little increase in free energy changes \((\Delta \mathrm{A}(\mathrm{C - D} = 4.69 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}(\mathrm{D - E}) = 3.48 \mathrm{kJ / mol})\) . The breakage of the \(\mathrm{Ca - O}\) bond with the oxygen ion from silicate is earlier, but more difficult than that with the interstitial oxygen ion (from the state C to D to E). These two sequential steps of breaking \(\mathrm{Ca - O_s}\) bonds decrease the total coordination number from seven to five, making this detached and free \(\mathrm{Ca}\) ion have more chances to accommodate one more water ligand and reform an octahedral structure, albeit at this stage is severely distorted with a trigonal bipyramid \((\mathrm{D_{3h}})\) structure. The free energy barrier and the free energy change for these steps \((\Delta \mathrm{A}^* (\mathrm{E - F}) = 8.70 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}(\mathrm{E - F}) = - 3.87 \mathrm{kJ / mol})\) are relatively high compared to the same fivefold to sixfold coordination transition step (A- B).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[130, 88, 890, 523]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 536, 895, 852]]<|/det|>
+<center>Figure 1. Dissolution mechanism of \(\mathrm{Ca}_{\alpha}\) from the \(\mathrm{Ca_3SiO_5}\) surface. a, c The one-dimensional free energy profiles with respect to \(\mathrm{CN(Ca - O_w)}\) and \(\mathrm{CN(Ca - O_w)}\) , respectively. b The twodimensional free energy surface with variables of \(\mathrm{CN(Ca - O_s)}\) and \(\mathrm{CN(Ca - O_w)}\) . d The configurations of the free energy minimum states on the FES and the corresponding reaction pathways. The state number, coordinates on the FES and the Helmholtz free energy values (kJ/mol) relative to state A are at the upper right. The values (kJ/mol) in red are free energy barriers and the values under the arrows in black are overall changes in free energies between two states. The yellow, blue, cyan, red and white spheres are indicted to the silicon, calcium (no bias potential), calcium (with bias potential), oxygen and hydrogen ions, respectively. For simplicity, the solute is shown in the transparent stick type. </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 80, 884, 895]]<|/det|>
+Dissolution pathways for \(\mathbf{Ca}_{\beta}\) with CN(Ca- O) from 5 to 2. The surface reactivity varies with the different sites on the \(\mathrm{Ca_3SiO_5}\) surface and the hydration process is closely correlated with the coordination number of surface Ca ions<sup>29</sup>. Thus, we carried out the comparative WT- MetaD simulations for \(\mathrm{Ca}_{\beta}\) to investigate whether the dissolution pathway alters with initial coordination environment. Obviously, the FES for the detachment of \(\mathrm{Ca}_{\beta}\) is different from that for \(\mathrm{Ca}_{\alpha}\) (Figures 2a, b and c) and becomes more complicated with more possible reaction pathways. When water molecules come to the \(\mathrm{Ca_3SiO_5}\) surface, the first stable state is A(5, 1). Albeit the number of water molecule is less than that for \(\mathrm{Ca}_{\alpha}\) , the total coordination number is same with six. However, the reaction pathway for the next step is more complex. The state A has two potential reaction paths to adsorb more water molecules. The first one adsorbs one more water molecule directly without breaking the \(\mathrm{Ca - O_5}\) bond (A- B). While the other one does it by breaking two \(\mathrm{Ca - O_5}\) bonds at the same time (A- D). From a thermodynamic point of view, it is more energetically favourable to pass through state B first due to the larger free energy change between the state A and the second free energy minimum. Nevertheless, from a kinetic point of view, it is rapider to react along the pathway of A- D because of its lower free energy barriers \((\Delta \mathrm{A}^+ (\mathrm{A} - \mathrm{D}) = 6.49 \mathrm{kJ / mol}\) and \(\Delta \mathrm{A}^+ (\mathrm{A} - \mathrm{B}) = 7.01 \mathrm{kJ / mol}\) ). The reaction pathway of A- B- C- D is similar to the B- C- D- E for the dissolution of \(\mathrm{Ca}_{\alpha}\) , in which the start of \(\mathrm{Ca - O_5}\) bond cleavage is from the sevenfold coordination state. Noticeably, the seven- coordinated species is an essential intermediate in the dissolution of \(\mathrm{Ca}\) , which is similar to the decomposition of \(\gamma\) - \(\mathrm{Al}_2\mathrm{O}_3^{30}\) . But the free energy barriers along the B- C- D \((\Delta \mathrm{A}^+ (\mathrm{B} - \mathrm{C}) = 7.38 \mathrm{kJ / mol}\) , \(\Delta \mathrm{A}^+ (\mathrm{C} - \mathrm{D}) = 4.24 \mathrm{kJ / mol}\) ) is much smaller than that along C- D- E for \(\mathrm{Ca}_{\alpha}\) . The state D(3, 2) is the most stable state with the same coordinate of the start point, A, on the FES of \(\mathrm{Ca}_{\alpha}\) and is also a new outset for other two different reaction pathways towards breaking one more \(\mathrm{Ca - O_5}\) bond. The reaction would be more possible to proceed in the direction from D to E(2, 2) due to the lower free energy barrier \((\Delta \mathrm{A}^+ (\mathrm{D} - \mathrm{E}) = 19.60 \mathrm{kJ / mol}\) ) compared to the route from D to G(2, 3) \((\Delta \mathrm{A}^+ (\mathrm{D} - \mathrm{E})\)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 883, 234]]<|/det|>
+\(= 29.02 \mathrm{kJ / mol}\) , \(\Delta \mathrm{A}^{*}(\mathrm{E - F}) = 26.11 \mathrm{kJ / mol})\) ) and it is the rate- controlling step among all the reactions. It should be noted that the further \(\mathrm{Ca - O}_5\) bond cleavage from 2 to 0 is not accessible during this simulation due to the large free energy barriers required. Therefore, to uncover the subsequent dissolution mechanism of \(\mathrm{Ca}\) with less than two- and even zero- coordinated \(\mathrm{O}_5\) , it is necessary to add a 'wall' to constrain the CVs in the region of interest.  
+
+<|ref|>image<|/ref|><|det|>[[130, 254, 888, 860]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 875, 895, 894]]<|/det|>
+<center>Figure 2. Dissolution mechanism of \(\mathrm{Ca}_{\beta}\) from the \(\mathrm{Ca}_3\mathrm{SiO}_5\) surface with \(\mathrm{CN}(\mathrm{Ca - O}_5)\) from 5 to </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[125, 83, 896, 366]]<|/det|>
+2. a, c The one-dimensional free energy profiles with respect to \(\mathrm{CN(Ca - O_s)}\) and \(\mathrm{CN(Ca - O_w)}\) , respectively. b The two-dimensional free energy surface with variables of \(\mathrm{CN(Ca - O_s)}\) and \(\mathrm{CN(Ca - O_w)}\) . d The overhead sketches for the first layer of the \(\mathrm{Ca_3SiO_5}\) surface of the free energy minimum states on the FES (the all-atom configurations is presented in Supplementary Figure 1) and the corresponding reaction pathways. The state number, coordinates on the FES and the Helmholtz free energy values (kJ/mol) relative to state A are presented around the corresponding structure. The values (kJ/mol) in red are free energy barriers and the values under the arrows in black are overall changes in free energies between two states. The biased \(\mathrm{Ca_B}\) is bolded, and the \(\mathrm{O_s}\) and \(\mathrm{O_w}\) bonded with \(\mathrm{Ca_B}\) were highlighted in red and blue, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[115, 408, 885, 892]]<|/det|>
+Complete dissolution of \(\mathrm{Ca_B}\) with \(\mathrm{CN(Ca - O_s)}\) from 2 to 0. For the further detachment of \(\mathrm{Ca_B}\) with \(\mathrm{CN(Ca - O_s)}\) from \(2\) to \(0\) , we assume that the start point is the most stable state \(\mathrm{H}(1,4)\) on the FES (Figure 3a) as it is most possible to exist in reality. To further dissolve, \(\mathrm{Ca_B}\) needs to guest a water molecule first to achieve an octahedral structure crossing over a \(26.99\mathrm{kJ / mol}\) free energy barrier and coming to the state \(\mathrm{I}(1,5)\) (Figure 3b). The next step of welcoming one more water molecule from the state \(\mathrm{I}\) to \(\mathrm{K}(1,6)\) is the rate- controlling step due to the highest free energy barrier of \(28.08\mathrm{kJ / mol}\) . After that, \(\mathrm{Ca_B}\) detaches from the original position progressively and finally gets rid of the confinement of \(\mathrm{O_s}\) network totally, which is hydrated by the surrounding water molecules to the six- or seven- coordinated solute ion. The five- , six- and seven- coordinated \(\mathrm{Ca_B}\) ion can be transformed to each other. However, the sixfold coordination state processes the greatest possibility, not only because the reactions from the fivefold state \(\mathrm{M}(0,5)\) and sevenfold state \(\mathrm{L}(0,7)\) to the sixfold state \(\mathrm{K}(0,6)\) are spontaneous \((\Delta \mathrm{A}(\mathrm{M - K}) = - 2.09\mathrm{kJ / mol}\) and \(\Delta \mathrm{A}(\mathrm{L - K}) = - 10.30\mathrm{kJ / mol})\) , but also the free energy barriers \((\Delta \mathrm{A}^*(\mathrm{M - K}) = 10.34\mathrm{kJ / mol}\) , \(\Delta \mathrm{A}^*(\mathrm{L - K}) = 6.96\mathrm{kJ / mol})\) are lower than those of the reverse reactions, which is confirmed by a further \(30\mathrm{ps}\) equilibrium AIMD simulations. The AIMD also show that the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 83, 880, 135]]<|/det|>
+coordinated \(\mathrm{O_w}\) with \(\mathrm{Ca}_{\beta}\) increased slightly and the dissolved Ca forms a more regular octahedral structure with water and hydroxyl with the evolution of time.  
+
+<|ref|>image<|/ref|><|det|>[[137, 191, 880, 710]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 724, 895, 876]]<|/det|>
+<center>Figure 3. The further dissolution mechanism of \(\mathrm{Ca}_{\beta}\) from the \(\mathrm{Ca}_3\mathrm{SiO}_5\) surface with \(\mathrm{CN}(\mathrm{Ca - O}_s)\) from 2 to 0. a The two-dimensional free energy surface with variables of \(\mathrm{CN}(\mathrm{Ca - O}_s)\) and \(\mathrm{CN}(\mathrm{Ca - O_w})\) . b The configurations of the free energy minimum states on the FES and the corresponding reaction pathways. The state number, coordinates on the FES and the Helmholtz free energy values (kJ/mol) relative to state H are at the upper right. The values (kJ/mol) in red </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[125, 83, 892, 135]]<|/det|>
+are free energy barriers and the values under the arrows in black are overall changes in free energies between two states.  
+
+<|ref|>text<|/ref|><|det|>[[115, 177, 883, 895]]<|/det|>
+Structural and dynamic analysis of the water/Ca \(_3\) SiO \(_5\) interface after the dissolution of the Ca ion. According to the atomic density and atomic excess profiles (Figure 4a), we define six regions for the water structure along the z direction on the Ca \(_3\) SiO \(_5\) surface. It is obvious that the H ion penetrates into the second layer of the surface ( \(z = 6 \mathring{\mathrm{A}}\) ) due to the escape of the Ca ion and the proton exchange, which is the first time for such a phenomenon in the hydration of Ca \(_3\) SiO \(_5\) to be observed by the AIMD simulation. The region II is the chemisorbed water molecule and III is a mixture of the physisorbed water molecule for the surface and the first hydration shell of the dissolved Ca, which expands the destruction area of the layered water structure compared to the surface before dissolution. The region IV, V and VI are the transition layer, bulk layer and water/vacuum layer, respectively, which are similar to those on the perfect Ca \(_3\) SiO \(_5\) surface. The radial distribution function (RDF) (Figure 3b) shows that the first and second hydration shells of the dissolved Ca are more than that of the surface Ca. It also presents that the Ca- O \(_w\) bond length for the dissolved Ca is \(2.39 \mathring{\mathrm{A}}\) , which is shorter than that for surface Ca with \(2.50 \mathring{\mathrm{A}}\) , indicating a stronger interaction between Ca and the water molecule after dissolving. The mean square displacement (MSD) (Figure 3c) presents a more dynamic property and stronger diffusion ability for the dissolved Ca compared to the previous surface state. We also calculated the infrared (IR) spectra for the system and extract the parts for Si- OH and Ca- OH. It clearly shows one band at \(916 \mathrm{cm}^{- 1}\) (Figure 4d), which is characteristic for the Si- OH bond raised in experimental IR spectra upon Ca \(_3\) SiO \(_5\) hydration \(^{19}\) . In addition, two bands arising at \(700 - 1000 \mathrm{cm}^{- 1}\) and \(3640 \mathrm{cm}^{- 1}\) (Figure 4e) shows the formation of Ca- OH during the dissolution of Ca ion as indicated by the experimental results \(^{19,31}\) . From the transmission electron microscopy (TEM) photograph (figure 4f), we can clear see when Ca \(_3\) SiO \(_5\) encounters
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 883, 300]]<|/det|>
+with water, the Ca cations are released from the \(\mathrm{Ca_3SiO_5}\) surface and form the \(\mathrm{Ca(OH)_2}\) in the aqueous solution. An atomistic interpretation of this natural phenomenon can be obtained from our AIMD simulation, which shows the Ca ion coordinated with five water molecules and one hydroxyl group releases from the \(\mathrm{Ca_3SiO_5}\) surface following the proton transfer from the water molecule to the second layer of the interstitial oxygen ion and the occupation of the initial Ca site by one water molecule. This water molecule is parallel to the surface and form hydrogen bond with the new formed O- H and \(\mathrm{O_s}\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[140, 108, 870, 911]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[125, 83, 896, 300]]<|/det|>
+Figure 4. a Atomic density for \(\mathrm{O_w}\) and H, and atomic excess profiles as a function of the height beginning at \(6\mathrm{\AA}\) from the bottom of the \(\mathrm{Ca_3SiO_5}\) surface. b Radial distribution function (RDF) between the dissolved Ca ion ( \(\mathrm{Ca_d}\) ) and \(\mathrm{O_w}\) as well as between the surface Ca ion in the same site ( \(\mathrm{Ca_s}\) ) and \(\mathrm{O_w}\) . c The mean square displacement (MSD) evolution of \(\mathrm{Ca_d}\) and \(\mathrm{Ca_s}\) . d, e IR spectrums for Si- OH group and Ca- OH group, respectively. f TEM image of the dissolution of Ca from the \(\mathrm{Ca_3SiO_5}\) surface and the snapshot of the equilibrium AIMD at \(30\mathrm{ps}\) . The composition and the electron diffraction patterns of \(\mathrm{Ca(OH)_2}\) refers to \(^{32}\)  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 346, 211, 363]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[115, 378, 884, 758]]<|/det|>
+The calcium sites with different coordination environments lead to different reaction pathways and free energy barriers. The dissolution of three- coordinated \(\mathrm{Ca_d}\) is easier than the five- coordinated \(\mathrm{Ca_{\beta}}\) not only because of its initial less restraint from the \(\mathrm{Ca_3SiO_5}\) surface, but also the smaller free energy barriers along the reaction path. In addition, The free energy barriers between the two stable states on either FES of \(\mathrm{Ca_d}\) and \(\mathrm{Ca_{\beta}}\) tends to be larger as the \(\mathrm{Ca - O_s}\) bonds become less, which means the water adsorption on the \(\mathrm{Ca_3SiO_5}\) surface is easier than the detachment of Ca and the kinetic rate decreases gradually as this process proceeds. Nonetheless even the highest free energy barrier is only \(29.02\mathrm{kJ / mol}\) , which is easy to be crossed, suggesting the dissolution is an auto- catalytic process. It should be noted that there is no human intervention during the simulation of crossing over the free energy barriers, which ensures the reliability of our results. Thus, the results obtained in this work are general and applicable to other kinds of calcium silicate species.  
+
+<|ref|>text<|/ref|><|det|>[[117, 771, 882, 888]]<|/det|>
+Our ab initio WT- MetaD simulations with explicit water solvation highlight the importance of the water molecule on the detachment of Ca during the \(\mathrm{Ca_3SiO_5}\) hydration. In short, the dissolution of Ca can be explained in terms of a ligand exchange process. It initially stimulated by the water adsorption, raising the total coordination number with oxygen ions to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 883, 465]]<|/det|>
+a high level (five to seven). The adsorbed water molecules can reduce the free energy barriers for breaking the Ca- O<sub>s</sub> bond, and thus provide an opportunity for Ca to break the Ca- O<sub>s</sub> bonds with the optimal six coordination number unchanged. In fact, the breakage of Ca- O<sub>s</sub> bonds is a multi- step and multi- orientation chemical reaction, and every step needs a relatively high free energy barrier, which is straight to cross for traditional AIMD simulations within 100 ps but easy in reality. In addition, the dissolution of Ca is further promoted by the proton exchange and the diffusion of water molecule from the chemisorbed layer into the second surface layer. On the one hand, the H ion penetrate into the second layer of the surface and bonds to the interstitial oxygen ion previously bonded to the dissolved Ca ion, leading to a repulsive force pushing Ca out of the surface. At the same time, this diffused water molecule resides in the position of the Ca ion before dissolution and forms the hydron bond network with O<sub>s</sub>, O<sub>w</sub> and H, which undermines the attractive force to this Ca.  
+
+<|ref|>text<|/ref|><|det|>[[115, 478, 883, 890]]<|/det|>
+In summary, an atomistic and mechanistic picture of Ca dissolution from the Ca<sub>3</sub>SiO<sub>5</sub> surface in water solution at the initial stage of Ca<sub>3</sub>SiO<sub>5</sub> hydration is investigated using the ab initio molecular dynamics and metadynamics simulations. We find that the Ca dissolution from the Ca<sub>3</sub>SiO<sub>5</sub> are multi- step and multi- orientation chemical reactions accompanied by the water adsorption, proton exchange, breakage of Ca- O<sub>s</sub> bonds and water diffusion. The thermodynamic and kinetic analyses show that the detachment of Ca is spontaneous when the Ca is not fully dissolved and unspontaneous when Ca is no more coordinated with O<sub>s</sub>. Additionally, the Ca dissolution is an auto- catalytic process with the highest free energy barriers of only 29 kJ/mol. The reaction pathways for Ca in different coordination environments are different and the less coordinated Ca is easier to leach from the surface. Besides, we find the water molecules provide not only the attractive forces pulling Ca out of the surface, but also repulsive forces in filling the precious Ca site and pushing it away. The present achievement thus provides an insight into the cement hydration and also predict the evolution
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 546, 101]]<|/det|>
+of other complex geochemical and catalytic systems.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 150, 196, 166]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[128, 199, 271, 215]]<|/det|>
+## Atomistic model  
+
+<|ref|>text<|/ref|><|det|>[[115, 244, 883, 625]]<|/det|>
+The \(\mathrm{Ca_3SiO_5}\) (111) surface was cleaved from the M3 type of \(\mathrm{Ca_3SiO_5}\) (obtained from CCSD \(^{33}\) ), which is most frequently observed in industrial clinkers \(^{34}\) . The details of the DFT- based geometry optimization of the bulk crystal and surface slab were listed in Supplementary Methods. Considering the adsorption of water molecule on the \(\mathrm{Ca_3SiO_5}\) surface occurs even before contacting bulk water \(^{35}\) , we firstly adsorbed isolated water molecules to saturate the dangling bond on the \(\mathrm{Ca_3SiO_5}\) surface. Then, we put a \(20\mathrm{\AA}\) thick layer of water with density of \(1\mathrm{g / cm^3}\) on the \(\mathrm{Ca_3SiO_5}\) (111) surface (totally 333 atoms) for the dissolution simulations. The lattice parameters were \(14.21\mathrm{\AA}\times 11.72\mathrm{\AA}\times 36\mathrm{\AA}\) after setting a vacuum of \(15\mathrm{\AA}\) along z direction. After the detachment of the \(\mathrm{Ca_{\beta}}\) ion from its initial position to the state of D, we further constructed a new model ( \(14.21\mathrm{\AA}\times 11.72\mathrm{\AA}\times 48\mathrm{\AA}\) ) by adding another \(10\mathrm{\AA}\) thick layer of water on the previous system (totally 498 atoms) to calculate the structural and dynamic properties of the equilibrium state.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 655, 281, 671]]<|/det|>
+## AIMD simulations  
+
+<|ref|>text<|/ref|><|det|>[[117, 701, 883, 885]]<|/det|>
+All the AIMD simulations reported in this work were performed within the framework of DFT with the generalized gradient approximation (GGA) using the Perdew- Burke- Ernzerhof (PBE) \(^{36}\) functional and Grimme D3 correction \(^{37}\) , which was implemented in the CP2K/Quickstep \(^{38}\) . The Core electrons were described by Goedecker- Teter- Hutter (GTH) pseudopotentials \(^{39,40}\) and the valence electrons were described by a mixed Gaussian and plane waves basis (GPW) \(^{41}\) . The wave functions were expanded on a double- \(\zeta\) valence polarized
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 883, 465]]<|/det|>
+(DZVP) basis set along with an auxiliary plane wave basis set at a cutoff energy of 500 Ry. The Brillouin zone was sampled by the gamma approximation. During AIMD, the nuclei were treated within the Born–Oppenheimer approximation with a timestep of 0.5 fs for equilibrium simulation, while 1 fs for metadynamics simulations with the replacement of hydrogen by deuterium to accelerate the structural evolution without energy drifts<sup>27,42</sup>. The temperature was maintained at 300 K using a Nosé-Hoover thermostat<sup>43, 44</sup> coupled to the system with a time constant of 1000 fs in the Canonical ensemble (NVT). The convergence criterion for energy was set to \(10^{- 12}\) Hartree and for self- consistent field was \(10^{- 6}\) Hartree. All the system were first optimized to a stable state and then thermalized for at least 2.5 ps before the production run for statistical analysis. The duration of the AIMD simulations for dissolution of \(\mathrm{Ca}_{\alpha}\) , dissolution of \(\mathrm{Ca}_{\beta}\) with \(\mathrm{CN}(\mathrm{Ca - O}_s)\) from 5 to 2, further dissolution of \(\mathrm{Ca}_{\beta}\) with \(\mathrm{CN}(\mathrm{Ca - Os})\) from 2 to 0, equilibrium of final state of \(\mathrm{Ca}_{\beta}\) were 100 ps, 63 ps, 46ps, and 30ps, respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 492, 352, 510]]<|/det|>
+## Metadynamics simulations  
+
+<|ref|>text<|/ref|><|det|>[[115, 538, 883, 722]]<|/det|>
+In the well- tempered ab initio metadynamics<sup>45</sup> (WT- MetaD) simulations, we utilized a two- dimensional collective variables (CVs) characterized by the coordination number (CN) to monitor the dissolution process. The \(\mathrm{CN}(\mathrm{Ca - Os})\) is the coordination number of the Ca ion with all oxygen ions from the surface slab, while \(\mathrm{CN}(\mathrm{Ca - Os})\) is the coordination number of the Ca ion. with all oxygen ions from water molecules. As defined in the PLUMED code<sup>46</sup>, the CN have the expression as follows:  
+
+<|ref|>equation<|/ref|><|det|>[[228, 736, 870, 812]]<|/det|>
+\[CN(Ca,O_{w / s}) = \sum_{j\in O_{w / s}}s_{ij}(r_{ij}) = \sum_{j\in O_{w / s}}\frac{1 - \left(\frac{r_{ij} - d_0}{r_0}\right)^n}{1 - \left(\frac{r_{ij} - d_0}{r_0}\frac{m}{n}\right)^m} \quad (1)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 825, 881, 910]]<|/det|>
+where \(\mathrm{r_{ij}}\) is the distance between atom i and atom j. \(\mathrm{s_{ij}(r_{ij})}\) is a rational type of switching function describing the coordination between atom i and j. \(\mathrm{d_0}\) is the central value of the function. \(\mathrm{r_0}\) is the acceptance distance of the switching function, where the function well be \(\mathrm{n / m}\) at \(\mathrm{d_0 + r_0}\) .
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 82, 880, 168]]<|/det|>
+Here, we define do is 2.42 Å, which is the equilibrium bond length between the Ca and O ions \(^{47}\) ; ro is 0.4 Å, which is around half of the full width at half maximum of the radial distribution function of Ca- O \(^{48}\) and n and m are 6 and 12, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[116, 181, 883, 431]]<|/det|>
+The Gaussian hills was deposited every 30 timesteps with the initial height of \(3.5 \mathrm{kJ / mol}\) and width of 0.15 for both CVs. The biasfactor were 15 for simulations of \(\mathrm{Ca}_{\alpha}\) and 24 for simulation of \(\mathrm{Ca}_{\beta}\) . In addition, to further investigation of the dissolution process to a larger extent, we added a quadratic wall with the force constant of \(500 \mathrm{kJ / mol}\) at the position of \(\mathrm{CN}(\mathrm{Ca - O}_s)\) equals to 1.5 to restrict the simulation of further dissolution of \(\mathrm{Ca}_{\beta}\) on the regions of free energy surface with \(\mathrm{CN}(\mathrm{Ca - O}_s)\) less than 1.5. The time evolutions of the CV1 and CV2, the convergence tests for the free energy surfaces and the errors between the free energy minima were shown in Supplementary Figures 2- 7.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 460, 470, 479]]<|/det|>
+## Structural and spectroscopy calculations  
+
+<|ref|>text<|/ref|><|det|>[[118, 508, 418, 526]]<|/det|>
+The atomic excess (ae) is defined as:  
+
+<|ref|>equation<|/ref|><|det|>[[360, 537, 869, 578]]<|/det|>
+\[a e = \frac{2[O w] - 2[H]}{2[O w] + 2[H]} \quad (2)\]  
+
+<|ref|>text<|/ref|><|det|>[[117, 592, 880, 644]]<|/det|>
+where \([O w]\) and \([H]\) are atomic density for Ow and H, respectively. The negative value for ae indicates an excess of Ow, while the positive one indicates an excess of H.  
+
+<|ref|>text<|/ref|><|det|>[[117, 658, 881, 809]]<|/det|>
+For the vibrational spectra, we use the last \(30 \mathrm{ps}\) equilibrium AIMD trajectory to calculate the infrared (IR) spectrum with the TRAVIS program \(^{46}\) . The molecular electric properties were calculated every 4 fs (8 timesteps) using the Voronoi integration approach \(^{49}\) . The IR spectrum of particular components of a system were computed though the Fourier transform of the molecular dipole autocorrelation function as follows:  
+
+<|ref|>equation<|/ref|><|det|>[[303, 822, 869, 858]]<|/det|>
+\[A(\omega) \propto \int \langle \dot{\mu} (\tau) \dot{\mu} (t + \tau) \rangle_{\tau} e^{-i\omega t} dt \quad (3)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 872, 880, 892]]<|/det|>
+where A is the absorption cross section, \(\omega\) is frequency, and \(\dot{\mu}\) is the time derivative of the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 688, 101]]<|/det|>
+dipole moment leading to the dipole- velocity autocorrelation function.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 133, 699, 151]]<|/det|>
+## Synthesis of \(\mathrm{Ca_3SiO_5}\) and characterization of dissolution of \(\mathrm{Ca_3SiO_5}\)  
+
+<|ref|>text<|/ref|><|det|>[[117, 180, 883, 399]]<|/det|>
+The synthesis of \(\mathrm{Ca_3SiO_5}\) was conducted through a traditional method \(^{50}\) . Calcium carbonate and silica with molar ratio of 3:1 was first ground to pass through a 63- \(\mu \mathrm{m}\) sieve and then evenly mixed for \(2\mathrm{h}\) . The obtained fine powder was compressed into pancakes in a lab press and put into a Pt crucible for calcination at \(1500^{\circ}\mathrm{C}\) for \(5\mathrm{h}\) , and quickly cooled down to ambient temperature within \(10\mathrm{min}\) . The obtained product was ground into fine powder, compressed and reclaimed for four repeating cycles before the production of pure \(\mathrm{Ca_3SiO_5}\) . The synthetic \(\mathrm{Ca_3SiO_5}\) was then dispersed in the water solution for hydrolysis.  
+
+<|ref|>text<|/ref|><|det|>[[117, 412, 881, 497]]<|/det|>
+The morphology of dissolution of \(\mathrm{Ca_3SiO_5}\) was characterized by transmission electron microscopy (JEOL, JEM- 2100, \(200\mathrm{kV}\) ), equipped with an energy dispersive spectroscopy system.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 100, 215, 116]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[113, 144, 884, 911]]<|/det|>
+1. Schliemann R, Churakov SV. Atomic scale mechanism of clay minerals dissolution revealed by ab initio simulations. Geochimica et Cosmochimica Acta 2021, 293: 438-460.  
+2. Ye S, Feng P, Liu Y, Liu J, Bullard JW. In situ nano-scale observation of C3A dissolution in water. Cement and Concrete Research 2020, 132: 106044.  
+3. Sato H, Miyagawa Y, Okabe T, Miyajima M, Sunada H. Dissolution mechanism of diclofenac sodium from wax matrix granules. Journal of pharmaceutical sciences 1997, 86(8): 929-934.  
+4. Pirkanniemi K, Sillanpää M. Heterogeneous water phase catalysis as an environmental application: a review. Chemosphere 2002, 48(10): 1047-1060.  
+5. Ravenelle RM, Copeland JR, Kim W-G, Crittenden JC, Sievers C. Structural changes of γ-Al2O3-supported catalysts in hot liquid water. Acs Catalysis 2011, 1(5): 552-561.  
+6. Taylor HF. Cement chemistry, vol. 2. Thomas Telford London, 1997.  
+7. Vogel W. Glass chemistry. Springer Science & Business Media, 2012.  
+8. Rodriguez J, Rodriguez M, De Aza S, Pena P. Reaction sintering of zircon-dolomite mixtures. Journal of the European Ceramic Society 2001, 21(3): 343-354.  
+9. Zheng Q, Wang W. Calcium silicate based high efficiency thermal insulation. British ceramic transactions 2000, 99(4): 187-190.  
+10. Kinoshita M, Baba K, Nagayasu A, Yamabe K, Shimooka T, Takeichi Yi, et al. Improvement of solubility and oral bioavailability of a poorly water-soluble drug, TAS-301, by its melt-adsorption on a porous calcium silicate. Journal of pharmaceutical sciences 2002, 91(2): 362-370.  
+11. Zhu Y-J, Guo X-X, Sham T-K. Calcium silicate-based drug delivery systems. Expert opinion on drug delivery 2017, 14(2): 215-228.  
+12. Jain SK, Awasthi A, Jain N, Agrawal G. Calcium silicate based microspheres of repaglinide for gastroenterite floating drug delivery: Preparation and in vitro characterization. Journal of controlled release 2005, 107(2): 300-309.  
+13. Prati C, Gandolfi MG. Calcium silicate bioactive cements: Biological perspectives and clinical applications. Dental materials 2015, 31(4): 351-370.  
+14. Xu S, Lin K, Wang Z, Chang J, Wang L, Lu J, et al. Reconstruction of calvarial defect of rabbits using porous calcium silicate bioactive ceramics. Biomaterials 2008, 29(17): 2588-2596.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 102, 881, 157]]<|/det|>
+15. Ding S-J, Shie M-Y, Wang C-Y. Novel fast-setting calcium silicate bone cements with high bioactivity and enhanced osteogenesis in vitro. Journal of Materials Chemistry 2009, 19(8): 1183-1190.  
+
+<|ref|>text<|/ref|><|det|>[[115, 177, 861, 195]]<|/det|>
+16. Zongjin L. Advanced concrete technology. Hoboken, New Jersey: John Wiley & Sons, Inc 2011.  
+
+<|ref|>text<|/ref|><|det|>[[115, 215, 880, 250]]<|/det|>
+17. Juilland P, Gallucci E, Flatt R, Scrivener K. Dissolution theory applied to the induction period in alite hydration. Cement and Concrete Research 2010, 40(6): 831-844.  
+
+<|ref|>text<|/ref|><|det|>[[117, 270, 881, 324]]<|/det|>
+18. Pustovgar E, Sangodkar RP, Andreev AS, Palacios M, Chmelka BF, Flatt RJ, et al. Understanding silicate hydration from quantitative analyses of hydrating tricalcium silicates. Nature communications 2016, 7(1): 1-9.  
+
+<|ref|>text<|/ref|><|det|>[[117, 344, 881, 398]]<|/det|>
+19. Higl J, Hinder D, Rathgeber C, Ramming B, Lindén M. Detailed in situ ATR-FTIR spectroscopy study of the early stages of CSH formation during hydration of monoclinic C3S. Cement and Concrete Research 2021, 142: 106367.  
+
+<|ref|>text<|/ref|><|det|>[[117, 418, 880, 453]]<|/det|>
+20. Bellmann F, Sowoidnich T, Ludwig H-M, Damidot D. Dissolution rates during the early hydration of tricalcium silicate. Cement and Concrete Research 2015, 72: 108-116.  
+
+<|ref|>text<|/ref|><|det|>[[117, 473, 880, 509]]<|/det|>
+21. Naber C, Bellmann F, Neubauer J. Influence of w/s ratio on alite dissolution and CSH precipitation rates during hydration. Cement and Concrete Research 2020, 134: 106087.  
+
+<|ref|>text<|/ref|><|det|>[[117, 528, 883, 564]]<|/det|>
+22. Nicoleau L, Bertolim MA. Analytical model for the alite (C3S) dissolution topography. Journal of the American Ceramic Society 2016, 99(3): 773-786.  
+
+<|ref|>text<|/ref|><|det|>[[117, 584, 880, 620]]<|/det|>
+23. Lasaga AC, Luttge A. Variation of crystal dissolution rate based on a dissolution stepwave model. Science 2001, 291(5512): 2400-2404.  
+
+<|ref|>text<|/ref|><|det|>[[117, 640, 881, 675]]<|/det|>
+24. Tadros M, Skalny J, Kalyoncu R. Early hydration of tricalcium silicate. Journal of the American Ceramic Society 1976, 59(7-8): 344-347.  
+
+<|ref|>text<|/ref|><|det|>[[117, 695, 880, 730]]<|/det|>
+25. Zhang Z, Han F, Yan P. Modelling the dissolution and precipitation process of the early hydration of C3S. Cement and Concrete Research 2020, 136: 106174.  
+
+<|ref|>text<|/ref|><|det|>[[117, 751, 880, 786]]<|/det|>
+26. Qi C, Spagnoli D, Fourie A. DFT-D study of single water adsorption on low-index surfaces of calcium silicate phases in cement. Applied Surface Science 2020, 518: 146255.  
+
+<|ref|>text<|/ref|><|det|>[[117, 806, 880, 860]]<|/det|>
+27. Claverie J, Bernard F, Cordeiro JMM, Kamali-Bernard S. Ab initio molecular dynamics description of proton transfer at water-tricalcium silicate interface. Cement and Concrete Research 2020, 136: 106162.  
+
+<|ref|>text<|/ref|><|det|>[[115, 880, 880, 897]]<|/det|>
+28. Claverie J, Bernard F, Cordeiro JMM, Kamali-Bernard S. Water's behaviour on Ca-rich tricalcium
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[175, 84, 883, 120]]<|/det|>
+silicate surfaces for various degrees of hydration: A molecular dynamics investigation. Journal of Physics and Chemistry of Solids 2019, 132: 48- 55.  
+
+<|ref|>text<|/ref|><|det|>[[117, 140, 881, 194]]<|/det|>
+29. Qi C, Manzano H, Spagnoli D, Chen Q, Fourie A. Initial hydration process of calcium silicates in Portland cement: A comprehensive comparison from molecular dynamics simulations. Cement and Concrete Research 2021, 149: 106576.  
+
+<|ref|>text<|/ref|><|det|>[[117, 214, 880, 268]]<|/det|>
+30. Reocreux R, Girel É, Clabaut P, Tuel A, Besson M, Chaumonnot A, et al. Reactivity of shape-controlled crystals and metadynamics simulations locate the weak spots of alumina in water. Nature communications 2019, 10(1): 1-8.  
+
+<|ref|>text<|/ref|><|det|>[[117, 288, 880, 342]]<|/det|>
+31. Khachani M, El Hamidi A, Halim M, Arsalan S. Non-isothermal kinetic and thermodynamic studies of the dehydroxylation process of synthetic calcium hydroxide Ca (OH) 2. J Mater Environ Sci 2014, 5(2): 615-624.  
+
+<|ref|>text<|/ref|><|det|>[[117, 362, 880, 398]]<|/det|>
+32. Sun G, Li Z, Liang R, Weng L-T, Zhang L. Super stretchable hydrogel achieved by non-aggregated spherulites with diameters < 5 nm. Nature communications 2016, 7(1): 1-8.  
+
+<|ref|>text<|/ref|><|det|>[[117, 417, 881, 472]]<|/det|>
+33. Mumme WG. CRYSTAL-STRUCTURE OF TRICALCIUM SILICATE FROM A PORTLAND-CEMENT CLINKER AND ITS APPLICATION TO QUANTITATIVE XRD ANALYSIS. Neues Jahrbuch Fur Mineralogie-Monatshefte 1995(4): 145-160.  
+
+<|ref|>text<|/ref|><|det|>[[117, 492, 880, 546]]<|/det|>
+34. De Noirfontaine M-N, Dunstetter F, Courtial M, Gasecki G, Signes-Frehel M. Polymorphism of tricalcium silicate, the major compound of Portland cement clinker: 2. Modelling alite for Rietveld analysis, an industrial challenge. Cement and concrete research 2006, 36(1): 54-64.  
+
+<|ref|>text<|/ref|><|det|>[[117, 566, 880, 602]]<|/det|>
+35. Thissen P, Natzeck C, Giraudo N, Weidler P, Woll C. Hydration of Concrete: The First Steps. Chemistry-A European Journal 2018, 24(34): 8603-8608.  
+
+<|ref|>text<|/ref|><|det|>[[117, 622, 881, 657]]<|/det|>
+36. Perdew JP, Burke K, Ernzerhof M. Generalized gradient approximation made simple. Physical review letters 1996, 77(18): 3865.  
+
+<|ref|>text<|/ref|><|det|>[[117, 678, 881, 732]]<|/det|>
+37. 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. The Journal of chemical physics 2010, 132(15): 154104.  
+
+<|ref|>text<|/ref|><|det|>[[117, 753, 880, 807]]<|/det|>
+38. VandeVondele J, Krack M, Mohamed F, Parrinello M, Chassaing T, Hutter J. Quickstep: Fast and accurate density functional calculations using a mixed Gaussian and plane waves approach. Computer Physics Communications 2005, 167(2): 103-128.  
+
+<|ref|>text<|/ref|><|det|>[[117, 826, 881, 861]]<|/det|>
+39. Goedecker S, Teter M, Hutter J. Separable dual-space Gaussian pseudopotentials. Physical Review B 1996, 54(3): 1703.  
+
+<|ref|>text<|/ref|><|det|>[[113, 881, 880, 898]]<|/det|>
+40. Hartwigsen C, Goedecker S, Hutter J. Relativistic separable dual-space Gaussian pseudopotentials
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 85, 542, 100]]<|/det|>
+from H to Rn. Physical Review B 1998, 58(7): 3641.  
+
+<|ref|>text<|/ref|><|det|>[[115, 121, 881, 157]]<|/det|>
+41. VandeVondele J, Hutter J. Gaussian basis sets for accurate calculations on molecular systems in gas and condensed phases. The Journal of chemical physics 2007, 127(11): 114105.  
+
+<|ref|>text<|/ref|><|det|>[[115, 177, 881, 213]]<|/det|>
+42. Leung K, Rempe SB. Ab initio rigid water: Effect on water structure, ion hydration, and thermodynamics. Physical Chemistry Chemical Physics 2006, 8(18): 2153-2162.  
+
+<|ref|>text<|/ref|><|det|>[[115, 233, 881, 268]]<|/det|>
+43. Martyna GJ, Klein ML, Tuckerman M. Nose-Hoover chains: The canonical ensemble via continuous dynamics. The Journal of chemical physics 1992, 97(4): 2635-2643.  
+
+<|ref|>text<|/ref|><|det|>[[115, 289, 881, 324]]<|/det|>
+44. Nose S. A molecular dynamics method for simulations in the canonical ensemble. Molecular physics 1984, 52(2): 255-268.  
+
+<|ref|>text<|/ref|><|det|>[[115, 344, 881, 379]]<|/det|>
+45. Bonomi M, Parrinello M. Enhanced sampling in the well-tempered ensemble. Physical review letters 2010, 104(19): 190601.  
+
+<|ref|>text<|/ref|><|det|>[[115, 400, 881, 435]]<|/det|>
+46. Brehm M, Thomas M, Gehrke S, Kirchner B. TRAVIS—A free analyzer for trajectories from molecular simulation. The Journal of chemical physics 2020, 152(16): 164105.  
+
+<|ref|>text<|/ref|><|det|>[[115, 455, 881, 509]]<|/det|>
+47. Jalilehvand F, Spangberg D, Lindqvist-Reis P, Hermansson K, Persson I, Sandström M. Hydration of the calcium ion. An EXAFS, large-angle X-ray scattering, and molecular dynamics simulation study. Journal of the American Chemical Society 2001, 123(3): 431-441.  
+
+<|ref|>text<|/ref|><|det|>[[115, 529, 880, 564]]<|/det|>
+48. Zhang W, Li S, Hou D, Geng Y, Zhang S, Yin B, et al. Study on unsaturated transport of cement-based silane sol coating materials. Coatings 2019, 9(7): 427.  
+
+<|ref|>text<|/ref|><|det|>[[115, 585, 881, 620]]<|/det|>
+49. Thomas M, Brehm M, Kirchner B. Voronoi dipole moments for the simulation of bulk phase vibrational spectra. Physical Chemistry Chemical Physics 2015, 17(5): 3207-3213.  
+
+<|ref|>text<|/ref|><|det|>[[115, 641, 881, 675]]<|/det|>
+50. Wesselsky A, Jensen OM. Synthesis of pure Portland cement phases. Cement and concrete research 2009, 39(11): 973-980.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 366, 176]]<|/det|>
+Supplementaryinformation.docx VideoS1.mp4
+
+<--- Page Split --->

preprint/preprint__7eb66b91436f3975fdb9236cf3899e9c784f5e9c7b00f5e0aac9122010266d35/images_list.json

@@ -0,0 +1,77 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1. Schematic of the formation of polyolefin reweaved ultra-micropore membrane.",
+    "footnote": [],
+    "bbox": [
+      [
+        167,
+        323,
+        850,
+        660
+      ]
+    ],
+    "page_idx": 3
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2. Structure characterizations of PIM-1/pGMA-x PRUMs. (a) Comparative \\(^1\\mathrm{H}\\) NMR spectra of PIM-1 and PIM-1/pGMA-x PRUMs. SEM EDS mapping of TB/pGMA PRUMs (b) TB/pGMA-7% PRUMs, (c) TB/pGMA-14% PRUMs, (d) TB/pGMA-25% PRUMs, (e) TB/pGMA-42% PRUMs, (f) XRD of PIM-1/pGMA-x PRUMs, (g) CO₂ adsorption isotherms at 273 K, (h) pore size distribution from CO₂ adsorption.",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        82,
+        833,
+        460
+      ]
+    ],
+    "page_idx": 5
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3. Simulated amorphous cells. The green area represents FFV (a) repetitive unit, (b) PIM-1, (c) PIM-1/pGMA-7%, (d) PIM-1/pGMA-15%, (e) PIM-1/pGMA-27%, (f) PIM-1/pGMA-40% (g) simulated FFV and density of polyolefin reweaved membranes.",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        198,
+        800,
+        565
+      ]
+    ],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4. Gas separation performance and diffusion properties of the pristine PIM-1 and obtained PIM-1/pGMA-x polyolefin reweaved membranes. (a) \\(\\mathrm{CH_4}\\) permeability and \\(\\mathrm{CO_2 / CH_4}\\) selectivity, (b) \\(\\mathrm{CO_2}\\) permeability and \\(\\mathrm{CO_2 / N_2}\\) selectivity, (c) \\(\\mathrm{O_2}\\) permeability and \\(\\mathrm{O_2 / N_2}\\) selectivity vary with loading of pGMA, (d) \\(\\mathrm{CO_2}\\) and \\(\\mathrm{CH_4}\\) diffusivity, (e) \\(\\mathrm{CO_2}\\) and \\(\\mathrm{CH_4}\\) solubility, (f) sorption and diffusion selectivity of \\(\\mathrm{CO_2 / CH_4}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        135,
+        850,
+        664
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5. Gas separation performances of pure gas and mixed gas, anti-plasticization and aging performance. (a) the upper bond of PIM-1/pGMA-x polyolefin reweaved membranes of \\(\\mathrm{CO_2 / CH_4}\\) , (b) \\(\\mathrm{CO_2 / CH_4}\\) mixed-gas of the upper bond, (c) anti-plasticization properties of permeability, (d) anti-plasticization properties of relative changes in selectivity, (e) aging properties of relative changes in permeability, (f) aging properties of selectivity, (g) the universality of the electron beam irradiation induced strategy, (h) the \\(^1\\mathrm{H}\\) NMR spectra of PI/pGMA PRUMs, (i) the \\(^1\\mathrm{H}\\) NMR spectra of TB/pGMA PRUMs.",
+    "footnote": [],
+    "bbox": [
+      [
+        154,
+        90,
+        848,
+        465
+      ]
+    ],
+    "page_idx": 12
+  }
+]

preprint/preprint__7eb66b91436f3975fdb9236cf3899e9c784f5e9c7b00f5e0aac9122010266d35/preprint__7eb66b91436f3975fdb9236cf3899e9c784f5e9c7b00f5e0aac9122010266d35.mmd

@@ -0,0 +1,354 @@
+
+# Polyolefin Reweaved Ultra-micropore Membrane for CO2 Capture  
+
+Gongping Liu gpliu@njtech.edu.cn  
+
+Nanjing Tech University https://orcid.org/0000- 0002- 3859- 1278  
+
+Xiuling Chen  
+
+Institute of Coal Chemistry, Chinese Academy of Sciences  
+
+Guining Chen  
+
+Nanjing Tech University  
+
+Lei Wu  
+
+Chinese Academy of Sciences  
+
+Nanwen Li  
+
+Institute of Coal Chemistry, Chinese Academy of Sciences https://orcid.org/0000- 0002- 2191- 8123  
+
+Wanqin Jin  
+
+Nanjing Tech University https://orcid.org/0000- 0001- 8103- 4883  
+
+Cong Xie  
+
+Hubei University of Science and Technology  
+
+Article  
+
+Keywords:  
+
+Posted Date: August 14th, 2024  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 4620538/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 2nd, 2025. See the published version at https://doi.org/10.1038/s41467- 024- 55540- z.
+
+<--- Page Split --->
+
+# Polyolefin Reweaved Ultra-micropore Membrane for \(\mathbf{CO}_2\) Capture  
+
+Xiuling Chen \(^{a,b*}\) , Guining Chen \(^{b}\) , Lei Wu \(^{c}\) , Cong Xie \(^{a}\) , Gongping Liu \(^{b*}\) , Nanwen Li \(^{c}\) , Wanqin Jin \(^{b}\)  
+
+\(^{a}\) Hubei Key Laboratory of Radiation Chemistry and Functional Materials, Hubei University of Science and Technology, Xianning 43780, China \(^{b}\) State Key Laboratory of Materials- Oriented Chemical Engineering, College of Chemical Engineering, Nanjing Tech University, Nanjing, 211816, China \(^{c}\) State Key Laboratory of Coal Conversion, Institute of Coal Chemistry, Chinese Academy of Sciences, Taiyuan 030001, China \(^{*}\) Corresponding authors: cxl828800@163. com (X. Chen); gpliu@njtech.edu.cn (G. Liu)  
+
+## ABSTRACT  
+
+High- performance gas separation membranes have potential in industrial separation applications, while overcoming the permeability- selectivity trade- off via regulable aperture distribution remains challenging. Here, we report a novel strategy to fabricate Polyolefin Reweaved Ultra- micropore Membrane (PRUM) to acquire regulable microporous channel. Specifically, olefin monomers are dispersed uniformly into a pristine membrane (e.g., PIM- 1) via solution diffusion method. Upon controlled electron beam irradiation, the olefin undergoes a free radical polymerization, resulting in the formation of olefin polymer in- situ reweaved in the membrane. The deliberately regulated and contracted pore- aperture size of the membrane can be accomplished by varying the olefin loading to achieve efficient gas separation. For instance, PIM- 1 PRUM containing 27wt% poly- methyl methacrylate demonstrate \(\mathrm{CO}_2\) permeability of 1976 Barrer, combined with \(\mathrm{CO}_2 / \mathrm{CH}_4\) and \(\mathrm{CO}_2 / \mathrm{N}_2\) selectivities of 58.4 and 48.3 respectively, transcending the performance upper bounds. This controllable and high efficiency- design strategy provides a general approach to create sub- nanometre- sized pore- apertures of gas separation membranes with wide universality.
+
+<--- Page Split --->
+
+Keywords: \(\mathrm{CO_2}\) capture; Polyolefin Reweaved Ultra-micropore Membrane; Electron Beam Irradiation.  
+
+## 1. Introduction  
+
+Membrane- based separation technologies have attracted great interest in recent years to address the need for energy- efficient separation processes and are widely utilized in fields such as natural gas sweetening, \(\mathrm{CO_2}\) capture and storage \(^{1 - 3}\) . The key to the future of membrane- based \(\mathrm{CO_2}\) capture and storage lies in highly permeable and selective membrane materials. Typically, commercial membrane materials such as polyimide, polysulfone, and cellulose acetate exhibit low permeability with acceptable selectivity for \(\mathrm{CO_2}\) removal from natural gas, which fails to meet the requirements for cost- effective process \(^{4 - 7}\) . So far, polymeric membranes are suffered from the permeability- selectivity trade- off (also known as Robeson's upper bound) \(^{8 - 9}\) .  
+
+Continuous efforts have been paid to enhance the gas separation performance of polymeric membranes, including (1) molecular design of polymers of intrinsic microporosity (PIMs), Tröger's base (TB), thermally rearranged polymers (TR), porous organic frameworks (POFs) \(^{10 - 14}\) ; (2) incorporating nanofillers such as nanocrystals and nanosheets into polymer \(^{15 - 20}\) ; (3) post- treatment such as molecular chain functionalization, thermal- or ultraviolet- crosslinking \(^{21 - 32}\) . These strategies are effective to tighten chain- to- chain spacing to form micropores and ultra- fine pores, while, the membranes are too brittle to meet the large- scale separation via the cross- linking methods, in addition, a single polymer membrane material resulted in limited performance improvements. Incorporation of nanoparticles into polymers for fabrication of mixed matrix membranes generally improve either the gas permeability or the selectivity to a certain extent, the filler agglomeration and poor interfacial compatibility between fillers and matrices. So, developing a facile strategy to fabricate polymeric membranes with both high permeability and selectivity for \(\mathrm{CO_2}\) capture is a long- standing challenge.  
+
+Here, we present a novel strategy to fabricating polyolefin reweaved ultramicropore membranes (PRUM) for effective gas separation, where olefin monomers
+
+<--- Page Split --->
+
+are in- situ self- polymerized through electron beam irradiation and reweaved in a pristine membrane. Distinct from existing methods, the pristine membrane acts as a scaffold that uniformly dissolves and immobilizes monomer molecule via monomer solution diffusion to form monomer@membrane precursor. The monomer in the membrane undergoes an in- situ free radical polymerization via electron beam irradiation. The resulting rigid polymer segments tightly attach to the pristine membrane micropores to form a defect- free PRUMs with controllable nanochannels for gas separation.  
+
+![](images/Figure_1.jpg)
+
+<center>Fig. 1. Schematic of the formation of polyolefin reweaved ultra-micropore membrane. </center>  
+
+To demonstrate the strategy, glycidyl methacrylate (GMA) monomer and PIM- 1 membrane were selected as precursors (Fig. 1). PIM, a class of high- performing gas separation membrane material with their unique rigidity and molecular packing, has high fractional free volume (FFV) and permeability to make it potentially tunable<sup>30- 32</sup>. Initially, GMA is dispersed uniformly into the PIM- 1 membrane via solution diffusion method, pGMA is synthesized via electron beam irradiation and in- situ reweaved to the PIM- 1 membrane. This process effectively reduces the spacing between PIM chains,
+
+<--- Page Split --->
+
+resulting in a PIM- 1 PRUMs (Fig. 1). The high- energy electron beam irradiation induces two effects<sup>33- 34</sup>, (1) creation of point defects in the carbon sheets and strands, (2) generation of free radical on sp2- structures of GMA. These carbon defects that are vulnerable to interstitial junctions in membrane and prompted the in- situ deposition of pGMA within the interlayer channels of the PIM- 1 membrane. The micropores are precisely regulated by the loading of pGMA in the PIM- 1 membrane. The introduction of pGMA leads to a reduction in interlayer spacing and manipulation of the microstructure of PIM- 1, resulting in a significant enhancement in the perm- selectivity of \(\mathrm{CO_2 / CH_4}\) , transcending the 2019 upper bound.  
+
+## Results  
+
+## Synthesis and characterizations of PIM-1 PRUMs  
+
+The synthesis of PIM- 1 polymer was shown in supporting information (Fig.S1). To fabricate defect- free PIM- 1 PRUM, the PIM- 1 membranes were prepared with thickness of \(50 - 60\mu \mathrm{m}\) and infiltrated in \(10\%\) GMA methanol solution in PET bags. The irradiation dosage ranged from 20 to \(80\mathrm{kGy}\) by a high- energy electron- beam accelerator in ambient air, the GMA monomer was polymerized to pGMA and the in- situ deposited in the PIM- 1 membrane. The resulting membranes were designated as PIM- 1/pGMA- x (the images of PIM- 1/pGMA- x were showed in Fig. S2), where x refers the load of pGMA (7%, 15%, 27%, 40%). To validate the successful synthesis of PIM- 1/pGMA- x PRUMs, \(^1\mathrm{H}\) NMR and ATR- FTIR spectra were used (Fig. 2a). The chemical shift at 5.61 and 6.16 ppm corresponding to the alkene hydrogen in GMA was not detected in the \(^1\mathrm{H}\) NMR spectra of pGMA polymers (Fig. S3), indicating the complete polymerization of the GMA monomer to pGMA polymers via electron beam irradiation. Further analysis of \(^1\mathrm{H}\) NMR spectra of PIM- 1/pGMA- x PRUMs revealed that the pGMA loading increased from 7% to 40% via the peak area ratio of \(\mathrm{CH_3}\) in pGMA polymers with the protons of the aromatic ring in PIM- 1 polymer varying from 1:2.23 to 1:0.75. As ATR- FTIR results showing, the distinct peaks in PIM- 1/pGMA- x PRUMs shifted and the epoxy characteristic peak of \(906\mathrm{cm}^{- 1}\) in PIM- 1/pGMA- x PRUMs became more and more obvious as the load of pGMA increased (Fig. S4).
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Fig. 2. Structure characterizations of PIM-1/pGMA-x PRUMs. (a) Comparative \(^1\mathrm{H}\) NMR spectra of PIM-1 and PIM-1/pGMA-x PRUMs. SEM EDS mapping of TB/pGMA PRUMs (b) TB/pGMA-7% PRUMs, (c) TB/pGMA-14% PRUMs, (d) TB/pGMA-25% PRUMs, (e) TB/pGMA-42% PRUMs, (f) XRD of PIM-1/pGMA-x PRUMs, (g) CO₂ adsorption isotherms at 273 K, (h) pore size distribution from CO₂ adsorption. </center>  
+
+TGA results further demonstrated that incorporating pGMA to the PIM- 1 membrane altered the degradation temperature of the main chain (Fig. S4). The pristine PIM- 1 membrane exhibited thermally stability with the degradation temperature of main chain occurring at \(460^{\circ}\mathrm{C}\) . In contrast, the PIM- 1/pGMA- x PRUMs showed the higher temperature of main chain degradation. The degradation temperature of main chain in PIM- 1/pGMA- x PRUMs elevated from \(460^{\circ}\mathrm{C}\) to \(511^{\circ}\mathrm{C}\) as the load of pGMA increased from \(7\%\) to \(40\%\) . The XPS results of \(\mathrm{C_{1s}}\) , \(\mathrm{N_{1s}}\) and \(\mathrm{O_{1s}}\) showed that the binding energy of atoms in PIM- 1/pGMA- x PRUMs gradually approaches to pGMA with the load of pGMA increasing (Fig. S5). These results confirmed that the pGMA are intercalated into interlayer channels of PIM- 1 membrane successfully and do not significantly
+
+<--- Page Split --->
+
+change the chemical structure of PIM- 1.  
+
+The results of characterizations confirm two critical points: (i) GMA monomer is undetectable in the PIM- 1/pGMA- x membrane, signifying their complete transformation to pGMA; (ii) the chemical shifts of \(\mathrm{- CH_2}\) or \(\mathrm{- CH_3}\) on the benzene ring remain unaffected, confirming that pGMA is physically interaction while not chemical reactions with PIM- 1. According to literature \(^{34}\) and the \(^1\mathrm{H}\) NMR results, we propose a possible mechanism of the electron beam irradiation: the olefin hydrogen sites in GMA monomers underwent break up during electron beam irradiation, leading to the formation of a radical- ion, then pGMA was formed through radical- ion coupling, and the generated pGMA was reweaved in situ with PIM- 1 membrane (containing point defects in the carbon sheets and strands).  
+
+To obtain the distribution features of pGMA in the membrane, we chose another typical microporous polymer Tröger's base (TB) as the pristine membrane to avoid the cross impact of oxygen elements between GMA and polymer (Fig. S1). TB is an optimal candidate due to its similar microporous structure and gas separation property to PIM, while contains only C and N elements. Following the same protocol of synthetizing PIM- 1 PRUM, we obtained TB/pGMA- x PRUMs (The \(^1\mathrm{H}\) NMR spectra of pGMA, TB and TB/pGMA- x PRUMs are shown in Fig. S6). Scanning electron microscope (SEM) images of the surface and cross- section confirmed that TB/pGMA- x PRUMs possessed a homogeneous morphological structure (Fig. S7). The pGMA distribution and concentration on TB/pGMA PRUMs were further analyzed by oxygen element EDS mapping. The cross- sectional SEM image indicated that oxygen element was dispersed uniformly in TB/pGMA- x PRUMs interior with the lower loading of pGMA (Fig. 2b- 2d), demonstrating that the generated pGMA was incorporated in interlayer channels of PIM- 1 membrane. However, as the pGMA loading increased to 40% (Fig. 2e), pGMA agglomeration became evident, which could potentially block a large number of micropores in the membrane.  
+
+## Pore structure characterization  
+
+To analyze the evaluation of interstitial space of the PIM- 1/pGMA- x PRUMs, powder X- ray diffraction (XRD) was used. XRD spectra of PIM- 1/pGMA- x PRUMs
+
+<--- Page Split --->
+
+revealed that all polymers are amorphous (Fig. 2f). Referring to pristine PIM- 1 membrane, three diffraction peaks were evident<sup>36</sup>, corresponding to the \(d\) - spacings of 6.58 Å, 4.88 Å and 4.06 Å, respectively. With increasing load of pGMA, \(d\) - spacing of 6.58 Å gradually became weak or even disappeared, which contribute to gas permeability and the results are consistent with the gas separation performance. The \(d\) - spacing of 4.88 Å and 4.06 Å attribution to the chain- to- chain distance and aromatic systems shifted to 4.56 Å and 3.95 Å respectively, implying that the increasing of pGMA tightens the chain spacing and potentially enhances the membrane molecular sieving properties. The XRD results showed that packing of PIM- 1 chains was changed, where the pGMA insertion could decrease the FFV and micro- pore size of PIM- 1 membrane.  
+
+As evidenced by the XRD results, the pGMA insertion into PIM- 1 membrane tends to tighten the polymer chains, leading to a reduced \(d\) - spacing. To gain further insights of microstructures, the pore size distribution of membranes was characterized using \(\mathrm{CO_2}\) adsorption experiments combined with molecular dynamics simulation. As shown in Fig. 2g- 2h, the \(\mathrm{CO_2}\) derived BET surface areas gradually decreased from 498.5 to 486.9, 432.9, and 322.5 \(\mathrm{m^2 / g}\) for PIM- 1, PIM- 1/pGMA- 7%, PIM- 1/pGMA- 27%, and PIM- 1/pGMA- 40%, respectively. The \(\mathrm{CO_2}\) uptake decreased by 55% along with a continuous reduction in the micropore surface area with the load of pGMA increasing from 7% to 40%, indicating effective regulated micropores with pGMA. The pore size distributions of pristine PIM- 1 membrane and electron beam irradiated PIM- 1/pGMA- x PRUMs derived from \(\mathrm{CO_2}\) sorption isotherms at 273 K were plotted using the nonlocal density functional theory (NLDFT) method (Fig. 2h). The peak of PIM- 1/pGMA- x PRUMs shifted to the left with the loading of pGMA increasing from 7% to 40%. The PIM- 1/pGMA- 27% PRUMs revealed a shrinkage of large microcores and a decrease of ultra- micropore region from 5 Å to 4.5 Å, which are associated with enhanced gas selectivity. A comparison between PIM- 1 membrane and PIM- 1/pGMA- 40% PRUMs revealed a decrease of the micropore from 8 Å to 7 Å, while the ultra- micropore of 5- 4 Å. Indeed, the insertion of pGMA into the PIM- 1 membrane not only regulation the micro- pores but also tailors the width of ultramicropores connecting neighboring
+
+<--- Page Split --->
+
+cavities, allowing selective diffusion of smaller gas molecules such as \(\mathrm{CO_2}\) , while excluding larger gas molecules like \(\mathrm{CH_4}\) , and \(\mathrm{N}_2\) , the discussions were shown in the section below.  
+
+![](images/Figure_3.jpg)
+
+<center>Fig. 3. Simulated amorphous cells. The green area represents FFV (a) repetitive unit, (b) PIM-1, (c) PIM-1/pGMA-7%, (d) PIM-1/pGMA-15%, (e) PIM-1/pGMA-27%, (f) PIM-1/pGMA-40% (g) simulated FFV and density of polyolefin reweaved membranes. </center>  
+
+## Molecular dynamics simulation of PIM-1/pGMA-x PRUMs  
+
+Molecular dynamics simulation was used to simulate the chain packing structure and calculated FFV of PIM- 1 and PIM- 1/pGMA- x PRUMs (Fig. S8). PIM- 1 membrane showed extraordinarily high fractional free volumes (FFV) owing to their ladder polymers and contorted chain structure. The free volume distributions of the pristine and the PIM- 1/pGMA- x PRUMs were depicted in Fig. 3. As revealed, the increase in pGMA resulted in a relatively narrower free volume distribution. The simulated FFV revealed a considerable drop from 0.3246 to 0.2217 between PIM- 1 and PIM- 1/pGMA- 40%. The reduced PIM- 1/pGMA- x FFV can be attributed to the decrease of micro
+
+<--- Page Split --->
+
+pores and chain packing.  
+
+## Gas transport properties of pure gas  
+
+The pGMA tuned micropores of the membrane was assessed through gas permeability measurements. The single gas separation performance of both pristine PIM- 1 and PIM- 1/pGMA- x PRUMs were evaluated sequentially for \(\mathrm{N}_2\) , \(\mathrm{CH}_4\) , \(\mathrm{O}_2\) , \(\mathrm{H}_2\) , and \(\mathrm{CO}_2\) at \(35^{\circ}\mathrm{C}\) and \(100\mathrm{psi}\) . The permeabilities and ideal selectivities are listed in illustrated in Fig. 4, Fig. S9 and Table S1. The pristine PIM- 1 membrane exhibited high gas permeability and moderate selectivity for all gases, which are consistent with our previous work \(^{25}\) . The slight difference is due to variations in polymer batches. Referring to post- treatment PIM- 1/pGMA- x PRUMs, a trend emerged wherein gas permeability decreased and selectivity increased with higher pGMA loading, consistent with the XRD and FFV findings. In comparing the gas permeability, a noteworthy observation was that the pristine PIM- 1 membrane exhibited a sequence of \(\mathrm{P_{CO2} > P_{H2} > P_{O2} > P_{CH4} > P_{N2}}\) , whereas in PIM- 1/pGMA- x PRUMs, the trend reversed to \(\mathrm{P_{H2} > P_{CO2}}\) and \(\mathrm{P_{N2} > P_{CH4}}\) . Such trend is in accordance with the order of gas kinetic diameters, suggesting the molecular sieving property of these membranes.  
+
+The variations in gas permeability and selectivity as a function of pGMA loading can be attributed to variations in pore distribution and simulated FFV (Fig. 3, Fig. S8). Integration of pGMA into interlayer channels of the PIM- 1 membrane served to mitigate FFV, thereby establishing ultra- selective gas channels conductive to a robust molecular sieving effect. Such as, PIM- 1/pGMA- 27% PRUMs presented a \(\mathrm{CO}_2 / \mathrm{N}_2\) selectivity 48.3 and \(\mathrm{CO}_2\) permeability of 1976 Barrer, showing a 3.6- fold increase over the pristine PIM- 1 membrane. The \(\mathrm{CO}_2 / \mathrm{CH}_4\) selectivity reached to 58.4, representing a 6.4- fold increase over the pristine PIM- 1 membrane. Compared to \(\mathrm{CO}_2 / \mathrm{CH}_4\) , \(\mathrm{O}_2 / \mathrm{N}_2\) separation is more challenging due to the closer kinetic diameters of the gas pairs. The PIM- 1/pGMA- 27% PRUMs exhibited superior \(\mathrm{O}_2 / \mathrm{N}_2\) selectivity of 6.9 and 282 Barrer of \(\mathrm{O}_2\) permeability to surpass the latest Robeson upper bound. However, as the loading of pGMA increasing to 40%, both the permeability and selectivity decreased simultaneously, the results furtherly prove that the loading of pGMA diminishes FFV and pore size, impeding the diffusion of larger gas molecules, and consequently
+
+<--- Page Split --->
+
+reducing gas permeability (Fig. 4).  
+
+![](images/Figure_4.jpg)
+
+<center>Fig. 4. Gas separation performance and diffusion properties of the pristine PIM-1 and obtained PIM-1/pGMA-x polyolefin reweaved membranes. (a) \(\mathrm{CH_4}\) permeability and \(\mathrm{CO_2 / CH_4}\) selectivity, (b) \(\mathrm{CO_2}\) permeability and \(\mathrm{CO_2 / N_2}\) selectivity, (c) \(\mathrm{O_2}\) permeability and \(\mathrm{O_2 / N_2}\) selectivity vary with loading of pGMA, (d) \(\mathrm{CO_2}\) and \(\mathrm{CH_4}\) diffusivity, (e) \(\mathrm{CO_2}\) and \(\mathrm{CH_4}\) solubility, (f) sorption and diffusion selectivity of \(\mathrm{CO_2 / CH_4}\) . </center>  
+
+To understand the critical role of electron beam irradiation on the PRUM formation, we did the following control experiments. By replacing the electron beam irradiation with UV radiation for the membrane preparation, the separation performance of PIM- 1
+
+<--- Page Split --->
+
+was not enhanced, due to the energy of UV is not sufficient to polymerize of GMA (Table S1). We also physically blended pGMA with PIM- 1 membrane to form a pGMA/PIM- 1 mixed- matrix membrane. The results showed that the gas permeability decreased significantly, and the selectivity was not significantly increased (Table S1). To exclude the possible effect of electron beam irradiation on the gas separation performance of PIM- 1, we tested the gas separation performance of PIM- 1 membrane irradiated by the identical electron beam condition. The results proved that gas permeability was decreased after electron beam irradiation, while the selectivity was maintained almost. These results confirmed that pGMA could enhance the gas separation performance of PIM- 1 membrane only via electron beam irradiation.  
+
+## Transport mechanism of PIM-1/pGMA-x PRUMs  
+
+Generally, selectivity is affected by both solubility and diffusivity selectivity, we further calculated the diffusion and solution coefficients of \(\mathrm{CO_2}\) and \(\mathrm{CH_4}\) based on the corresponding the time- lag of the constant- volume/variable- pressure method for PIM- 1 and PIM- 1/pGMA- x membranes. As shown in Fig. 4e- 4f and Table S2, compared with PIM- 1 membranes, PIM- 1/pGMA- x PRUMs showed lower solution and diffusion coefficients for both \(\mathrm{CO_2}\) and \(\mathrm{CH_4}\) molecules. While, the \(\mathrm{CO_2 / CH_4}\) diffusivity selectivity of PIM- 1/pGMA- x PRUMs were improved from 1.06 for PIM- 1 to 2.2, 3.3, 5.5, 6.9 respectively for PIM- 1/pGMA- x PRUMs. Particularly, PIM- 1/pGMA- 40% PRUMs exhibited the highest diffusivity selectivity of 6.9, almost 4.3 times as that of PIM- 1 membranes. The solubility selectivity of PIM- 1/pGMA- x membranes were also significantly enhanced compared with that of PIM- 1 membranes except for PIM- 1/pGMA- 40%. The enhanced solubility selectivity can be attributed to the reduced micropore sizes and narrow pore size distribution. The results showed that both the diffusion and solution process affect the efficient \(\mathrm{CO_2 / CH_4}\) separation in PIM- 1/pGMA- x PRUMs, consider the influence of diffusivity and solubility selectivity, the PIM- 1/pGMA- 27% PRUMs showed the best \(\mathrm{CO_2 / CH_4}\) selectivity.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Fig. 5. Gas separation performances of pure gas and mixed gas, anti-plasticization and aging performance. (a) the upper bond of PIM-1/pGMA-x polyolefin reweaved membranes of \(\mathrm{CO_2 / CH_4}\) , (b) \(\mathrm{CO_2 / CH_4}\) mixed-gas of the upper bond, (c) anti-plasticization properties of permeability, (d) anti-plasticization properties of relative changes in selectivity, (e) aging properties of relative changes in permeability, (f) aging properties of selectivity, (g) the universality of the electron beam irradiation induced strategy, (h) the \(^1\mathrm{H}\) NMR spectra of PI/pGMA PRUMs, (i) the \(^1\mathrm{H}\) NMR spectra of TB/pGMA PRUMs. </center>  
+
+## Separation performance comparison with literature  
+
+The gas separation performance of the pristine PIM- 1 and PRUMs in comparison with the latest Robeson's upper bound are shown in Fig. 5a and Fig. S10. The \(\mathrm{CO_2 / CH_4}\) , \(\mathrm{CO_2 / N_2}\) and \(\mathrm{O_2 / N_2}\) gas separation performance of PIM- 1 membrane are below the 2008 upper bounds<sup>37</sup>. It is noteworthy that PIM- 1/pGMA- 27 PRUMs demonstrated exceptional gas separation performance for critical gas pairs, such as \(\mathrm{CO_2 / CH_4}\) and \(\mathrm{CO_2 / N_2}\) selectivity transcended the 2019 upper bounds, and the \(\mathrm{O_2 / N_2}\) selectivity exceeded the 2015 upper bound<sup>8- 9</sup>. The results suggest that the PIM- 1/pGMA- x PRUMs
+
+<--- Page Split --->
+
+can be potentially used in industry for \(\mathrm{CO_2}\) separation and \(\mathrm{O_2}\) enrichment. The comparison of gas separation results of the present PIM- 1/pGMA- x PRUMs with the other reported PIM membranes is compiled in Table S3. The performance comparisons suggest that the electron beam irradiation induced strategy provides a new approach to developing high- performance gas separation membrane through combining electron beam irradiation and olefin monomer to precisely regulate the micropores, thus enhancing the molecular sieving effect.  
+
+The \(\mathrm{CO_2 / CH_4}\) mixed- gas testing was conducted for PIM- 1membrane and PIM- 1/pGMA- 27% PRUMs at \(35^{\circ}\mathrm{C}\) , with \(\mathrm{CO_2}\) upstream pressures ranging from 100 to 300 psi, and the results are summarized in Fig. 5b. The \(\mathrm{CO_2}\) permeability of PIM- 1 decreased with increasing upstream pressure to 300 psi with a lower \(\mathrm{CO_2 / CH_4}\) selectivity, and the performance was below the 2018 mixed- gas trade- off curve. Conversely, the \(\mathrm{CO_2}\) permeability of PIM- 1/pGMA- 27% showed only a slight decrease by increasing the upstream pressure to 300 psi, the \(\mathrm{CO_2 / CH_4}\) mixed- gas selectivity reached as high as 44.1 combining with a \(\mathrm{CO_2}\) permeability of 1702 Barrer, the performance notably exceeded the 2018 mixed- gas trade- off curve for \(\mathrm{CO_2 / CH_4}\) .  
+
+## Anti-plasticizing and aging behavior  
+
+Plasticization is a common phenomenon for polymer membranes in gas separation under aggressive gas feed conditions, plasticization reduces polymer membrane separation performance, so develop plasticization resistance of a polymer membrane is urgently needed. To investigate the membranes properties concerning \(\mathrm{CO_2}\) anti- plasticizing, we conducted permeation test on PIM- 1 membrane and PIM- 1/pGMA- 27% PRUMs under varying \(\mathrm{CO_2}\) feed pressures. As shown in Fig. 5c- d and Table S4, In the case of the PIM- 1 membranes showed a noticeable plasticization behavior near 100 psi, which was reflected by an upswing in \(\mathrm{CO_2}\) permeability. This result is commonly observed in previously reported plasticization of PIM- 1<sup>36</sup>. Conversely, the PIM- 1/pGMA- 27% PRUMs showed a noticeable plasticization response at \(\sim 350\) psi, indicating highly enhanced plasticizing resistance than PIM- 1. This enhancement of \(\mathrm{CO_2}\) anti- plasticization can be attributed to the lower \(d\) - space of PIM- 1/pGMA- 27% PRUMs and reduce soluble gases \(\mathrm{CO_2}\) increases polymer free volume.
+
+<--- Page Split --->
+
+PIM- 1 presents pronounced physical aging, which is one of the major obstacles for its application as a commercial membrane material for gas separation. We evaluated the stability of PIM- 1/pGMA- 27% PRUMs over a 240- day period using four pure gases: \(\mathrm{CO_2}\) , \(\mathrm{CH_4}\) , \(\mathrm{O_2}\) and \(\mathrm{N_2}\) (Fig. 5e- f and Table S5). Throughout the aging process, the membrane was kept in ambient air condition except during the gas permeation tests. It is known that the PIM- 1 membrane exhibited poor aging stability. For example, pristine PIM- 1 membrane experienced a \(64\%\) reduction in \(\mathrm{CO_2}\) permeability within just \(30\sim 90\) days<sup>37</sup>. In contrast, the \(\mathrm{CO_2}\) permeability of PIM- 1/pGMA- 27% membrane decreased only by \(11\%\) over the same duration. Notably, the anti- aging behavior of PIM- 1/pGMA- 27% PRUMs were significantly higher than that of the pristine PIM- 1. This improvement may be due to the insertion of pGMA in the PIM- 1 membrane to rigidify the polymer main chain and decreases the membrane micropore shrinkage, as evidenced by the XRD results.  
+
+## Universality of the electron beam irradiation induced strategy  
+
+We explored other typical olefin molecules such as acrylic acid (AA) and styrene (SE) to fabricate PIM- 1 based PRUMs through controlled irradiation dosage. The results proved that pAA and pSE also improved the separation performance of PIM- 1 membrane, as shown in Fig. 5g- i and Table S6. We extended our investigations to other typical polymeric membrane materials: 6FDA- DAM and \(\mathrm{TB}^{35 - 36, 38 - 39}\) . The results, detailed in Fig. 5g- i and Table S7- S8, showed that this strategy effectively improved the gas separation performance of these membranes as well. Similar to PIM- 1/pGMA- x PRUMs, the increasing load of pGMA shown higher selectivity compared with fresh 6FDA- DAM and TB membranes. For example, the \(\mathrm{CO_2 / CH_4}\) selectivity of 6FDA- DAM/pGMA- 21.7% PRUMs improved by \(352\%\) . The \(\mathrm{CO_2 / CH_4}\) and \(\mathrm{H_2 / CH_4}\) selectivities of TB/PGMA- 25.1% improved by \(254\%\) and \(361\%\) respectively. Given a very large scope of available polymer membranes, as well as broad small molecules to optimize the doping process, we envision that the proposed electron beam irradiation holds tremendous potential for precisely fine- tuning polyolefin reweaved membranes and facilitating scale- up efforts.
+
+<--- Page Split --->
+
+## Discussion  
+
+In summary, we demonstrated a new kind of polyolefin reweaved ultramicropor membranes (PRUMs) through electron beam irradiation strategy that leverages electron beam irradiated in situ polymerization of small molecules to precisely regulate the microporous structure of membrane materials. By varying the load of pGMA polymer, we sought to strike a balance between separation performance and polymer loading, experimental data and molecular simulation revealed that the increase of pGMA reduced both FFV and gas permeability. Ultramicropor distribution was shifted to be more concentrated sub- nanometer range, which fits well for precisely discriminating small gas molecules such as \(\mathrm{CO_2O_2}\) , \(\mathrm{N_2}\) and \(\mathrm{CH_4}\) . Typically, the \(\mathrm{CO_2 / CH_4}\) selectivity of PIM- 1/pGMA- 27%, 6FDA- DAM/pGMA- 21.7%, TB/pGMA- 25.1% improved by 642%, 352% and 254% respectively. Our electron beam irradiation strategy to PRUM is not only applicable to a wide range of polymers and small molecules, but also offers multiple operational advantages such as time- independent thermal treatment, controllable operations and processes, high efficiency, and lack of catalysts or initiators.  
+
+## Methods  
+
+## Materials  
+
+4,4'- Diamino- 3,3'- dimethylbiphenyl, diethoxyethane, glycidyl methacrylate (GMA) were purchased from Aladdin and used as applied without further purification. 5,5'6,6'- Tetrahydroxy- 3,3,3',3'- tetramethyl- 1,1'- spirobisindane (TTSBI), 4,4'- diamino- 3,3'- dimethylbiphenyl and diethoxyethane were bought from Energy Chemical, 2,3,5,6- Tetrafluoroterephthalonitrile (TFTPN) was bought from Sigma- Aldrich. TTSBI was purified by crystallization with tetrahydrofuran and \(n\) - hexane. 2,3,5,6- TFTPN was purified by vacuum sublimation at \(150^{\circ}\mathrm{C}\) , anhydrous potassium carbonate \((\mathrm{K}_2\mathrm{CO}_3\) , Aladdin company) was dried in a vacuum oven at \(120^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) and stored in a desiccator before use. \(N\) - methyl- 2- pyrrolidone (NMP, Aladdin company), and toluene (Aladdin company) were dried using metallic sodium to remove moisture before use. All test gases (>99.9999%) employed for the evaluation of membraneseparation performance were supplied by Air Liquide (Taiyuan).  
+
+Preparation of PIM- 1, 6FDA- DAM, TB and electron beam irradiated polyolefin reweaved membranes
+
+<--- Page Split --->
+
+PIM- 1 6FDA- DAM, and TB polymers were synthesized according to previously reported procedures \(^{35,38}\) , as shown in Fig. S1- 3. Dense polymer membranes with a thickness of \(50 \mu \mathrm{m}\) (±10 um) were prepared from 7 wt.% polymer solutions in chloroform. Polymer solutions were filtered through \(0.45 \mu \mathrm{m}\) polypropylene filters and then cast into Teflon Petri dishes in a glovebox and allowed to evaporate slowly for \(48 \mathrm{~h}\) . The dry PIM- 1, 6FDA- DAM, TB membranes were placed into PET bags ( \(8 \mathrm{~cm} \times 10 \mathrm{~cm}\) ) equipped \(10\%\) GMA methanol solution ( \(10 \mathrm{wt} \% \mathrm{GMA} / 90 \mathrm{wt} \%\) methanol) overnight to ensure the GMA solution was dispersed evenly throughout the membrane. Afterwards, these PET bags were irradiated by \(1 \mathrm{MeV}\) electronic accelerator at room temperature with different absorbed doses. The reaction parameters were varied to desired ranges ( \(20 - 80 \mathrm{kGy}\) absorbed dose in irradiation ETFE films for \(10 \mathrm{min}\) ). At the end of the irradiation reaction, the irradiated PET bags were put into a water bath at \(40^{\circ} \mathrm{C}\) for \(3 \mathrm{~h}\) to make GMA polymerization more complete. The membranes were removed ultrasonically with lots of methanol to eliminate the homopolymer and the GMA monomer, then dried under vacuum at \(80^{\circ} \mathrm{C}\) for \(24 \mathrm{~h}\) to acquire PIM- 1/pGMA- x polyolefin reweaved membranes, 6FDA- DAM/pGMA- x polyolefin reweaved membranes, TB/pGMA- x polyolefin reweaved membranes (x refers to the load of pGMA). The load of pGMA (DG) on the membranes was calculated as follows equation:  
+
+DG \((\%) = (\mathrm{W}1 - \mathrm{W}0) / \mathrm{W}0\times 100\%\) (1)  
+
+where Wo and W1 represent the weights of the initial and grafted membranes, respectively.  
+
+The PIM- 1/pGMA- x ( \(x = 7\%\) , \(15\%\) , \(27\%\) , \(40\%\) ) polyolefin reweaved membranes were obtained via radiation doses \(20 \mathrm{kGy}\) , \(40 \mathrm{kGy}\) , \(60 \mathrm{kGy}\) or \(80 \mathrm{kGy}\) or \(80 \mathrm{kGy}\) respectively.  
+
+## Characterization  
+
+1 MeV electronic accelerator was purchased from USA (Wasik Associates Inc., USA). The \(^1 \mathrm{H}\) NMR spectrums were measured at a spectrometer of \(400 \mathrm{MHz}\) in \(\mathrm{CDCl}_3\) (Bruker AVANCE- III), and the chemical shifts were recorded in ppm. Thermogravimetric analysis (TGA) was used to investigate the thermal stabilities of polymer membranes using a TA SDT Q600. Attenuated total reflection Fourier- transform infrared (ATR- IR) spectra (AVATAR 360, Thermo Nicolet, USA) were recorded from \(500 \mathrm{to} 4000 \mathrm{cm}^{- 1}\) with \(32 \mathrm{scans}\) and \(4 \mathrm{cm}^{- 1}\) resolutions for membranesamples. Scanning electronic microscopy/Energy- dispersive X- ray spectroscopy mapping (SEM/EDS) was applied to analysis the surface morphology and elemental information and density of TB/pGMA- x polyolefin reweaved membranes by a Hitachi S5500 microscope. \(\mathrm{CO}_2\) adsorption- desorption was performed
+
+<--- Page Split --->
+
+using a Micromeritics ASAP 2020 instrument. X- ray diffraction (XRD) was used to study the \(d\) - spacing changes of the PIM- 1/pGMA- x polyolefin reweaved membranes and the results were recorded on a Bruker AXS GADDS apparatus using Cu radiation with a wavelength of 1.54 Å. The testing angle range was \(5 - 40^{\circ}\) at a rate of \(4^{\circ} / \mathrm{min}\) and \(d\) - spacing was computed following Bragg's law \((d = \lambda /2\sin \theta)\) .  
+
+## Gas permeation measurements  
+
+The gas separation performance of PIM- 1/pGMA- x, 6FDA- DAM/pGMA- x, TB/pGMA- x membranes were tested at \(35^{\circ}\mathrm{C}\) using a constant- volume/variable- pressure method. The downstream pressure was measured using a transducer down to \(10^{- 6}\) torr, and the steady state pressure changing with time (dp/dt) was selected to calculate the permeability (P). Pure gas was tested in the sequence of \(\mathrm{N}_2\) , \(\mathrm{CH}_4\) , \(\mathrm{O}_2\) , \(\mathrm{CO}_2\) and \(\mathrm{H}_2\) at 80 psi. PIM- 1/pGMA- x and TB/pGMA- x polyolefin reweaved membranes were tested for three different samples and the deviation was less than \(5\%\) . Permeability (P) was calculated using the following equation 2:  
+
+\[P = 10^{10}\times \frac{V_{d}\times l}{p_{up}\times T\times R\times A}\times \frac{dp}{dt} \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \quad (2)\]  
+
+Where \(P\) is the permeability (Barrer), 1 Barrer \(= 10^{- 10}\mathrm{cm}^3 (\mathrm{STP})\mathrm{cm}\mathrm{cm}^{- 2}\mathrm{s}^{- 1}\mathrm{cmHg}^{- 1}\) , \(V_{d}\) is the calibrated permeate volume \((\mathrm{cm}^3)\) , \(l\) is the membrane thickness \((\mathrm{cm})\) , \(p_{up}\) is the upstream pressure \((\mathrm{cmHg})\) , \(A\) is the effective membrane area \((\mathrm{cm}^2)\) , \(T\) is the operating temperature \((\mathrm{K})\) , \(R\) is the gas constant \((0.278\mathrm{cm}^3\mathrm{cmHg}\mathrm{cm}^{- 3}(\mathrm{STP})\mathrm{K}^{- 1})\) and \(dp / dt\) is the steady- state downstream pressure increase rate \((\mathrm{cmHg}\mathrm{s}^{- 1})\) .  
+
+The ideal selectivity \((\alpha_{x / y})\) for components x and y was defined as the ratio of gas permeability of the two components via equation 3.  
+
+\[\alpha_{x / y} = \frac{p_{x}}{p_{y}} \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \quad (3)\]  
+
+The diffusion coefficient D was calculated from the time- lag apparatus using equation 4:  
+
+\(\mathrm{D} = 1^{2} / 6\epsilon\) (4)  
+
+\(\epsilon\) denotes the time lag. solubility coefficient (S) was obtained indirectly via equation 5:  
+
+\(\mathrm{S} = \mathrm{P} / \mathrm{D}\) (5)  
+
+## Molecular simulation  
+
+To verify the packing change in the structural after the electron beam irradiation process, PIM- 1/pGMA- 7%- 40% polyolefin reweaved membranes were built by the amorphous cell based on
+
+<--- Page Split --->
+
+lamps calculations. PIM- 1 amorphous cell module consists of 6 polymer chains with 20 repeating units each, and pGMA amorphous cell module consists of 6 polymer chains with 20 repeating units each. The force field is pcf, and all processes are carried out under the NPT system, first relaxation at high temperature and pressure (800 K, 500 bar) 2ns, the equilibrium stage temperature was set to 300 K and 1.5 bar. The actual load of pGMA is slightly different from the calculated load, the load of the calculation pGMA is PIM- 1/pGMA- 7.17%, PIM- 1/pGMA- 15.63%, PIM- 1/pGMA- 26.08%, PIM- 1/pGMA- 40.45%.  
+
+## Acknowledgements  
+
+We acknowledge the financial support from National Natural Science Foundation of China (22378102), special fund for the Key Laboratory of Hubei Province (2022ZX02 and 2022ZX04).  
+
+## Author contributions  
+
+All authors contributed to the scientific discussion and manuscript preparation. Xiuling Chen and Lei Wu led the experimental design, data curation, and writing of the pristine manuscript. Guining Chen performed the transport and sorption analyses of membranes. Cong Xie performed the electron beam irradiation. Nanwen Li conceived the concept of the research, Gongping Liu and Wanqin Jin helped to revise the manuscript.  
+
+## Competing interests  
+
+The authors declare no competing interests.  
+
+## Data availability  
+
+Data will be made available on request.  
+
+## References  
+
+[1] Rochelle, G. T. Amine scrubbing for \(\mathrm{CO_2}\) capture. Science 325, 1652- 1654 (2009). [2] Dai, Y. et al. A review on the recent advances in composite membranes for \(\mathrm{CO_2}\) capture processes. Sep. Purif. Technol. 307, 122752 (2023). [3] Koros, W. J. & Zhang, C. Materials for next- generation molecularly selective synthetic membranes. Nat. Mater. 16, 289- 297 (2017).
+
+<--- Page Split --->
+
+[4] Ockwig, N. W. & Nenoff, T. M. Membranes for hydrogen separation. Chem. Rev. 107, 4078- 4110 (2007).  
+
+[5] Nikolaeva, D. et al. The performance of affordable and stable cellulose- based poly- ionic membranes in \(\mathrm{CO_2 / N_2}\) and \(\mathrm{CO_2 / CH_4}\) gas separation. J. Membr. Sci. 564, 552- 561 (2018).  
+
+[6] Murphy, T. M. et al. Physical aging of polystyrene films tracked by gas permeability. Polymer 54, 873- 880 (2013).  
+
+[7] Chanem, B. S. et al. High- performance membranes from polyimides with intrinsic microporosity. Adv. Mater. 20, 2766- 2771 (2008).  
+
+[8] Swaidan, R., Ghanem, B. & Pinnau, I. Fine- tuned intrinsically ultramicroporous polymers redefine the permeability/selectivity upper bounds of membrane- based air and hydrogen separations. ACS Macro Lett. 4, 947- 951 (2015).  
+
+[9] Comesaña- Gandara, B. et al. Redefining the Robeson upper bounds for \(\mathrm{CO_2 / CH_4}\) and \(\mathrm{CO_2 / N_2}\) separations using a series of ultrapemable benzotriptycene- based polymers of intrinsic microporosity. Energy Environ. Sci. 12, 2733- 2740 (2019).  
+
+[10] He, S. et al. Recent progress in PIM- 1 based membranes for sustainable \(\mathrm{CO_2}\) separations: Polymer structure manipulation and mixed matrix membrane design. Sep. Purif. Technol. 284, 120277 (2022).  
+
+[11] Hazazi, K. et al. Precise molecular sieving of ethylene from ethane using triptycene- derived submicroporous carbon membranes. Nat. Mater. 22, 1218- 1226 (2023).  
+
+[12] Bandehali, S. et al. Polymers of intrinsic microporosity and thermally rearranged
+
+<--- Page Split --->
+
+polymer membranes for highly efficient gas separation. Sep. Purif. Technol. 278, 119513 (2022).  
+
+[13] Balcik, M. et al. Exploring the effect of intra- chain rigidity on mixed- gas separation performance of a Triptycene- Tröger's base ladder polymer (PIM- Trip- TB) by atomistic simulations. J. Membr. Sci. 677, 121614 (2023).  
+
+[14] Yu, G. et al. Mixed matrix membranes derived from nanoscale porous organic frameworks for permeable and selective \(\mathrm{CO_2}\) separation. J. Membr. Sci. 591, 117343 (2019).  
+
+[15] Chen, X. et al. Tailoring the microporosity of polymers of intrinsic microporosity for advanced gas separation by atomic layer deposition. Angew. Chem. Int. Ed. 60, 17875- 17880 (2021).  
+
+[16] Terrones, M. et al. Coalescence of single- walled carbon nanotubes. Science 288, 1226- 1229 (2000).  
+
+[17] Bushell, A. F. et al. Gas permeation parameters of mixed matrix membranes based on the polymer of intrinsic microporosity PIM- 1 and the zeolitic imidazolate framework ZIF- 8. J. Membr. Sci. 427, 48- 62 (2013).  
+
+[18] Ghalei, B. et al. Enhanced selectivity in mixed matrix membranes for \(\mathrm{CO_2}\) capture through efficient dispersion of amine- functionalized MOF nanoparticles. Nature, Energy 2, 17086 (2017).  
+
+[19] He, G. et al. Synergistic \(\mathrm{CO_2}\) - sieving from polymer with intrinsic microporosity masking nanoporous single- layer graphene. Adv. Funct. Mater. 30, 2003979 (2020).  
+
+[20] Ma, C. & Urban, J. J. Hydrogen- bonded polyimide/metal- organic framework
+
+<--- Page Split --->
+
+hybrid membranes for ultrafast separations of multiple gas pairs. Adv. Funct. Mater. 29, 1903243 (2019).  
+
+[21] Ji, W. et al. Facile tailoring molecular sieving effect of PIM- 1 by in- situ \(\mathrm{O}_3\) treatment for high performance hydrogen separation. J. Membr. Sci. 662, 120971 (2022). [22] Satilmis, B. et al. Hydroxyalkylaminoalkylamide PIMs: selective adsorption by ethanolamine- and diethanolamine- modified PIM- 1. Macromolecules 48, 5663- 5669 (2015).  
+
+[23] Rodriguez, K.M. et al. Leveraging free volume manipulation to improve the membrane separation performance of amine- functionalized PIM- 1. Angew. Chem. Int. Ed. 60, 6593- 6599 (2021).  
+
+[24] Jung, D. et al. Postsynthetically modified polymers of intrinsic microporosity (PIMs) for capturing toxic gases. ACS Appl. Mater. Interf. 13, 10409- 10415 (2021).  
+
+[25] Pinnau, I. et al. Recent Progress on Polymers of Intrinsic Microporosity and Thermally Modified Analogue Materials for Membrane- Based Fluid Separations. Small Struct. 2, 2100049 (2021).  
+
+[26] Li, F.Y. et al. High- performance thermally self- cross- linked polymer of intrinsic microporosity (PIM- 1) membranes for energy development. Macromolecules 45, 1427- 1437 (2012).  
+
+[27] He, Y. et al. Polymers with side chain porosity for ultrapermeable and plasticization resistant materials for gas separations. Adv. Mater. 31, 1807871 (2019).  
+
+[28] Du, N. et al. Polymer nanosieve membranes for \(\mathrm{CO}_2\) - capture applications. Nat. Mater. 10, 372- 375 (2011).
+
+<--- Page Split --->
+
+[29] Tien- Binh, N. et al. In- situ cross interface linking of PIM- 1 polymer and UiO- 66- NH2 for outstanding gas separation and physical aging control. J. Membr. Sci. 548, 429- 438 (2018).  
+
+[30] Song, Q. et al. Controlled thermal oxidative crosslinking of polymers of intrinsic microporosity towards tunable molecular sieve membranes. Nat. Commun. 4, 1918 (2013).  
+
+[31] Song, Q. & Sivaniah, E. Photo- oxidative enhancement of polymeric molecular sieve membranes. Nat. Commun. 5, 4813 (2014).  
+
+[32] Li, F. et al. UV- rearranged PIM- 1 polymeric membranes for advanced hydrogen purification and production. Adv. Energy Mater. 2, 1456- 1466 (2012).  
+
+[33] Oh, B. et al. Electron- mediated control of nanoporosity for targeted molecular separation in carbon membranes. Nat. Commun. 13, 4972 (2022).  
+
+[34] Dong, Z. et al. Recent progress in environmental applications of functional adsorbent prepared by radiation techniques: A review. J. Hazard. Mater. 424, 126887 (2022).  
+
+[35] Wu, L. et al. Enhanced molecular selectivity and plasticization resistance in ring- opened Tröger's base polymer membranes. J. Membr. Sci. 634, 119399 (2021).  
+
+[36] Zhao, S. et al. Blending of compatible polymer of intrinsic microporosity (PIM- 1) with Tröger's Base polymer for gas separation membranes. J. Membr. Sci. 566, 77- 86 (2018).  
+
+[37] Alberto, M. et al. Impeded physical aging in PIM- 1 membranes containing graphene like fillers. J. Membr. Sci. 563, 513- 520 (2018).
+
+<--- Page Split --->
+
+[38] Chen, X. et al. Ultra- selective molecular- sieving gas separation membranes enabled by multi- covalent- crosslinking of microporous polymer blends. Nat. Commun. 12, 6140 (2021).  
+
+[39] Chen, X. et al. Hydrogen bonding- induced 6FDA- DABA/TB polymer blends for high performance gas separation membranes. J. Membr. Sci. 655, 120575 (2022).
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- Sl.docx
+
+<--- Page Split --->

preprint/preprint__7eb66b91436f3975fdb9236cf3899e9c784f5e9c7b00f5e0aac9122010266d35/preprint__7eb66b91436f3975fdb9236cf3899e9c784f5e9c7b00f5e0aac9122010266d35_det.mmd

@@ -0,0 +1,474 @@
+<|ref|>title<|/ref|><|det|>[[44, 106, 943, 175]]<|/det|>
+# Polyolefin Reweaved Ultra-micropore Membrane for CO2 Capture  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 268, 240]]<|/det|>
+Gongping Liu gpliu@njtech.edu.cn  
+
+<|ref|>text<|/ref|><|det|>[[44, 268, 620, 309]]<|/det|>
+Nanjing Tech University https://orcid.org/0000- 0002- 3859- 1278  
+
+<|ref|>text<|/ref|><|det|>[[44, 312, 560, 331]]<|/det|>
+Xiuling Chen  
+
+<|ref|>text<|/ref|><|det|>[[44, 315, 560, 336]]<|/det|>
+Institute of Coal Chemistry, Chinese Academy of Sciences  
+
+<|ref|>text<|/ref|><|det|>[[44, 341, 163, 359]]<|/det|>
+Guining Chen  
+
+<|ref|>text<|/ref|><|det|>[[52, 363, 261, 381]]<|/det|>
+Nanjing Tech University  
+
+<|ref|>text<|/ref|><|det|>[[44, 387, 99, 404]]<|/det|>
+Lei Wu  
+
+<|ref|>text<|/ref|><|det|>[[52, 408, 320, 427]]<|/det|>
+Chinese Academy of Sciences  
+
+<|ref|>text<|/ref|><|det|>[[44, 432, 139, 450]]<|/det|>
+Nanwen Li  
+
+<|ref|>text<|/ref|><|det|>[[44, 454, 920, 475]]<|/det|>
+Institute of Coal Chemistry, Chinese Academy of Sciences https://orcid.org/0000- 0002- 2191- 8123  
+
+<|ref|>text<|/ref|><|det|>[[44, 479, 143, 497]]<|/det|>
+Wanqin Jin  
+
+<|ref|>text<|/ref|><|det|>[[52, 500, 620, 520]]<|/det|>
+Nanjing Tech University https://orcid.org/0000- 0001- 8103- 4883  
+
+<|ref|>text<|/ref|><|det|>[[44, 525, 125, 544]]<|/det|>
+Cong Xie  
+
+<|ref|>text<|/ref|><|det|>[[52, 547, 438, 567]]<|/det|>
+Hubei University of Science and Technology  
+
+<|ref|>text<|/ref|><|det|>[[44, 608, 105, 625]]<|/det|>
+Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 645, 134, 664]]<|/det|>
+Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 683, 321, 703]]<|/det|>
+Posted Date: August 14th, 2024  
+
+<|ref|>text<|/ref|><|det|>[[44, 721, 475, 741]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 4620538/v1  
+
+<|ref|>text<|/ref|><|det|>[[42, 758, 914, 802]]<|/det|>
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 819, 535, 840]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 875, 936, 919]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on January 2nd, 2025. See the published version at https://doi.org/10.1038/s41467- 024- 55540- z.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[163, 93, 835, 115]]<|/det|>
+# Polyolefin Reweaved Ultra-micropore Membrane for \(\mathbf{CO}_2\) Capture  
+
+<|ref|>text<|/ref|><|det|>[[165, 141, 850, 190]]<|/det|>
+Xiuling Chen \(^{a,b*}\) , Guining Chen \(^{b}\) , Lei Wu \(^{c}\) , Cong Xie \(^{a}\) , Gongping Liu \(^{b*}\) , Nanwen Li \(^{c}\) , Wanqin Jin \(^{b}\)  
+
+<|ref|>text<|/ref|><|det|>[[163, 214, 853, 430]]<|/det|>
+\(^{a}\) Hubei Key Laboratory of Radiation Chemistry and Functional Materials, Hubei University of Science and Technology, Xianning 43780, China \(^{b}\) State Key Laboratory of Materials- Oriented Chemical Engineering, College of Chemical Engineering, Nanjing Tech University, Nanjing, 211816, China \(^{c}\) State Key Laboratory of Coal Conversion, Institute of Coal Chemistry, Chinese Academy of Sciences, Taiyuan 030001, China \(^{*}\) Corresponding authors: cxl828800@163. com (X. Chen); gpliu@njtech.edu.cn (G. Liu)  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 454, 260, 471]]<|/det|>
+## ABSTRACT  
+
+<|ref|>text<|/ref|><|det|>[[147, 497, 853, 908]]<|/det|>
+High- performance gas separation membranes have potential in industrial separation applications, while overcoming the permeability- selectivity trade- off via regulable aperture distribution remains challenging. Here, we report a novel strategy to fabricate Polyolefin Reweaved Ultra- micropore Membrane (PRUM) to acquire regulable microporous channel. Specifically, olefin monomers are dispersed uniformly into a pristine membrane (e.g., PIM- 1) via solution diffusion method. Upon controlled electron beam irradiation, the olefin undergoes a free radical polymerization, resulting in the formation of olefin polymer in- situ reweaved in the membrane. The deliberately regulated and contracted pore- aperture size of the membrane can be accomplished by varying the olefin loading to achieve efficient gas separation. For instance, PIM- 1 PRUM containing 27wt% poly- methyl methacrylate demonstrate \(\mathrm{CO}_2\) permeability of 1976 Barrer, combined with \(\mathrm{CO}_2 / \mathrm{CH}_4\) and \(\mathrm{CO}_2 / \mathrm{N}_2\) selectivities of 58.4 and 48.3 respectively, transcending the performance upper bounds. This controllable and high efficiency- design strategy provides a general approach to create sub- nanometre- sized pore- apertures of gas separation membranes with wide universality.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 90, 850, 135]]<|/det|>
+Keywords: \(\mathrm{CO_2}\) capture; Polyolefin Reweaved Ultra-micropore Membrane; Electron Beam Irradiation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 165, 280, 181]]<|/det|>
+## 1. Introduction  
+
+<|ref|>text<|/ref|><|det|>[[147, 191, 852, 432]]<|/det|>
+Membrane- based separation technologies have attracted great interest in recent years to address the need for energy- efficient separation processes and are widely utilized in fields such as natural gas sweetening, \(\mathrm{CO_2}\) capture and storage \(^{1 - 3}\) . The key to the future of membrane- based \(\mathrm{CO_2}\) capture and storage lies in highly permeable and selective membrane materials. Typically, commercial membrane materials such as polyimide, polysulfone, and cellulose acetate exhibit low permeability with acceptable selectivity for \(\mathrm{CO_2}\) removal from natural gas, which fails to meet the requirements for cost- effective process \(^{4 - 7}\) . So far, polymeric membranes are suffered from the permeability- selectivity trade- off (also known as Robeson's upper bound) \(^{8 - 9}\) .  
+
+<|ref|>text<|/ref|><|det|>[[147, 441, 852, 850]]<|/det|>
+Continuous efforts have been paid to enhance the gas separation performance of polymeric membranes, including (1) molecular design of polymers of intrinsic microporosity (PIMs), Tröger's base (TB), thermally rearranged polymers (TR), porous organic frameworks (POFs) \(^{10 - 14}\) ; (2) incorporating nanofillers such as nanocrystals and nanosheets into polymer \(^{15 - 20}\) ; (3) post- treatment such as molecular chain functionalization, thermal- or ultraviolet- crosslinking \(^{21 - 32}\) . These strategies are effective to tighten chain- to- chain spacing to form micropores and ultra- fine pores, while, the membranes are too brittle to meet the large- scale separation via the cross- linking methods, in addition, a single polymer membrane material resulted in limited performance improvements. Incorporation of nanoparticles into polymers for fabrication of mixed matrix membranes generally improve either the gas permeability or the selectivity to a certain extent, the filler agglomeration and poor interfacial compatibility between fillers and matrices. So, developing a facile strategy to fabricate polymeric membranes with both high permeability and selectivity for \(\mathrm{CO_2}\) capture is a long- standing challenge.  
+
+<|ref|>text<|/ref|><|det|>[[148, 858, 850, 904]]<|/det|>
+Here, we present a novel strategy to fabricating polyolefin reweaved ultramicropore membranes (PRUM) for effective gas separation, where olefin monomers
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 853, 302]]<|/det|>
+are in- situ self- polymerized through electron beam irradiation and reweaved in a pristine membrane. Distinct from existing methods, the pristine membrane acts as a scaffold that uniformly dissolves and immobilizes monomer molecule via monomer solution diffusion to form monomer@membrane precursor. The monomer in the membrane undergoes an in- situ free radical polymerization via electron beam irradiation. The resulting rigid polymer segments tightly attach to the pristine membrane micropores to form a defect- free PRUMs with controllable nanochannels for gas separation.  
+
+<|ref|>image<|/ref|><|det|>[[167, 323, 850, 660]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 684, 794, 701]]<|/det|>
+<center>Fig. 1. Schematic of the formation of polyolefin reweaved ultra-micropore membrane. </center>  
+
+<|ref|>text<|/ref|><|det|>[[147, 729, 852, 914]]<|/det|>
+To demonstrate the strategy, glycidyl methacrylate (GMA) monomer and PIM- 1 membrane were selected as precursors (Fig. 1). PIM, a class of high- performing gas separation membrane material with their unique rigidity and molecular packing, has high fractional free volume (FFV) and permeability to make it potentially tunable<sup>30- 32</sup>. Initially, GMA is dispersed uniformly into the PIM- 1 membrane via solution diffusion method, pGMA is synthesized via electron beam irradiation and in- situ reweaved to the PIM- 1 membrane. This process effectively reduces the spacing between PIM chains,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 854, 330]]<|/det|>
+resulting in a PIM- 1 PRUMs (Fig. 1). The high- energy electron beam irradiation induces two effects<sup>33- 34</sup>, (1) creation of point defects in the carbon sheets and strands, (2) generation of free radical on sp2- structures of GMA. These carbon defects that are vulnerable to interstitial junctions in membrane and prompted the in- situ deposition of pGMA within the interlayer channels of the PIM- 1 membrane. The micropores are precisely regulated by the loading of pGMA in the PIM- 1 membrane. The introduction of pGMA leads to a reduction in interlayer spacing and manipulation of the microstructure of PIM- 1, resulting in a significant enhancement in the perm- selectivity of \(\mathrm{CO_2 / CH_4}\) , transcending the 2019 upper bound.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 340, 214, 356]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 367, 580, 385]]<|/det|>
+## Synthesis and characterizations of PIM-1 PRUMs  
+
+<|ref|>text<|/ref|><|det|>[[147, 394, 854, 888]]<|/det|>
+The synthesis of PIM- 1 polymer was shown in supporting information (Fig.S1). To fabricate defect- free PIM- 1 PRUM, the PIM- 1 membranes were prepared with thickness of \(50 - 60\mu \mathrm{m}\) and infiltrated in \(10\%\) GMA methanol solution in PET bags. The irradiation dosage ranged from 20 to \(80\mathrm{kGy}\) by a high- energy electron- beam accelerator in ambient air, the GMA monomer was polymerized to pGMA and the in- situ deposited in the PIM- 1 membrane. The resulting membranes were designated as PIM- 1/pGMA- x (the images of PIM- 1/pGMA- x were showed in Fig. S2), where x refers the load of pGMA (7%, 15%, 27%, 40%). To validate the successful synthesis of PIM- 1/pGMA- x PRUMs, \(^1\mathrm{H}\) NMR and ATR- FTIR spectra were used (Fig. 2a). The chemical shift at 5.61 and 6.16 ppm corresponding to the alkene hydrogen in GMA was not detected in the \(^1\mathrm{H}\) NMR spectra of pGMA polymers (Fig. S3), indicating the complete polymerization of the GMA monomer to pGMA polymers via electron beam irradiation. Further analysis of \(^1\mathrm{H}\) NMR spectra of PIM- 1/pGMA- x PRUMs revealed that the pGMA loading increased from 7% to 40% via the peak area ratio of \(\mathrm{CH_3}\) in pGMA polymers with the protons of the aromatic ring in PIM- 1 polymer varying from 1:2.23 to 1:0.75. As ATR- FTIR results showing, the distinct peaks in PIM- 1/pGMA- x PRUMs shifted and the epoxy characteristic peak of \(906\mathrm{cm}^{- 1}\) in PIM- 1/pGMA- x PRUMs became more and more obvious as the load of pGMA increased (Fig. S4).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[156, 82, 833, 460]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 479, 852, 608]]<|/det|>
+<center>Fig. 2. Structure characterizations of PIM-1/pGMA-x PRUMs. (a) Comparative \(^1\mathrm{H}\) NMR spectra of PIM-1 and PIM-1/pGMA-x PRUMs. SEM EDS mapping of TB/pGMA PRUMs (b) TB/pGMA-7% PRUMs, (c) TB/pGMA-14% PRUMs, (d) TB/pGMA-25% PRUMs, (e) TB/pGMA-42% PRUMs, (f) XRD of PIM-1/pGMA-x PRUMs, (g) CO₂ adsorption isotherms at 273 K, (h) pore size distribution from CO₂ adsorption. </center>  
+
+<|ref|>text<|/ref|><|det|>[[147, 635, 853, 905]]<|/det|>
+TGA results further demonstrated that incorporating pGMA to the PIM- 1 membrane altered the degradation temperature of the main chain (Fig. S4). The pristine PIM- 1 membrane exhibited thermally stability with the degradation temperature of main chain occurring at \(460^{\circ}\mathrm{C}\) . In contrast, the PIM- 1/pGMA- x PRUMs showed the higher temperature of main chain degradation. The degradation temperature of main chain in PIM- 1/pGMA- x PRUMs elevated from \(460^{\circ}\mathrm{C}\) to \(511^{\circ}\mathrm{C}\) as the load of pGMA increased from \(7\%\) to \(40\%\) . The XPS results of \(\mathrm{C_{1s}}\) , \(\mathrm{N_{1s}}\) and \(\mathrm{O_{1s}}\) showed that the binding energy of atoms in PIM- 1/pGMA- x PRUMs gradually approaches to pGMA with the load of pGMA increasing (Fig. S5). These results confirmed that the pGMA are intercalated into interlayer channels of PIM- 1 membrane successfully and do not significantly
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 90, 474, 107]]<|/det|>
+change the chemical structure of PIM- 1.  
+
+<|ref|>text<|/ref|><|det|>[[147, 117, 853, 386]]<|/det|>
+The results of characterizations confirm two critical points: (i) GMA monomer is undetectable in the PIM- 1/pGMA- x membrane, signifying their complete transformation to pGMA; (ii) the chemical shifts of \(\mathrm{- CH_2}\) or \(\mathrm{- CH_3}\) on the benzene ring remain unaffected, confirming that pGMA is physically interaction while not chemical reactions with PIM- 1. According to literature \(^{34}\) and the \(^1\mathrm{H}\) NMR results, we propose a possible mechanism of the electron beam irradiation: the olefin hydrogen sites in GMA monomers underwent break up during electron beam irradiation, leading to the formation of a radical- ion, then pGMA was formed through radical- ion coupling, and the generated pGMA was reweaved in situ with PIM- 1 membrane (containing point defects in the carbon sheets and strands).  
+
+<|ref|>text<|/ref|><|det|>[[147, 395, 853, 841]]<|/det|>
+To obtain the distribution features of pGMA in the membrane, we chose another typical microporous polymer Tröger's base (TB) as the pristine membrane to avoid the cross impact of oxygen elements between GMA and polymer (Fig. S1). TB is an optimal candidate due to its similar microporous structure and gas separation property to PIM, while contains only C and N elements. Following the same protocol of synthetizing PIM- 1 PRUM, we obtained TB/pGMA- x PRUMs (The \(^1\mathrm{H}\) NMR spectra of pGMA, TB and TB/pGMA- x PRUMs are shown in Fig. S6). Scanning electron microscope (SEM) images of the surface and cross- section confirmed that TB/pGMA- x PRUMs possessed a homogeneous morphological structure (Fig. S7). The pGMA distribution and concentration on TB/pGMA PRUMs were further analyzed by oxygen element EDS mapping. The cross- sectional SEM image indicated that oxygen element was dispersed uniformly in TB/pGMA- x PRUMs interior with the lower loading of pGMA (Fig. 2b- 2d), demonstrating that the generated pGMA was incorporated in interlayer channels of PIM- 1 membrane. However, as the pGMA loading increased to 40% (Fig. 2e), pGMA agglomeration became evident, which could potentially block a large number of micropores in the membrane.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 846, 421, 862]]<|/det|>
+## Pore structure characterization  
+
+<|ref|>text<|/ref|><|det|>[[148, 868, 850, 914]]<|/det|>
+To analyze the evaluation of interstitial space of the PIM- 1/pGMA- x PRUMs, powder X- ray diffraction (XRD) was used. XRD spectra of PIM- 1/pGMA- x PRUMs
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 854, 386]]<|/det|>
+revealed that all polymers are amorphous (Fig. 2f). Referring to pristine PIM- 1 membrane, three diffraction peaks were evident<sup>36</sup>, corresponding to the \(d\) - spacings of 6.58 Å, 4.88 Å and 4.06 Å, respectively. With increasing load of pGMA, \(d\) - spacing of 6.58 Å gradually became weak or even disappeared, which contribute to gas permeability and the results are consistent with the gas separation performance. The \(d\) - spacing of 4.88 Å and 4.06 Å attribution to the chain- to- chain distance and aromatic systems shifted to 4.56 Å and 3.95 Å respectively, implying that the increasing of pGMA tightens the chain spacing and potentially enhances the membrane molecular sieving properties. The XRD results showed that packing of PIM- 1 chains was changed, where the pGMA insertion could decrease the FFV and micro- pore size of PIM- 1 membrane.  
+
+<|ref|>text<|/ref|><|det|>[[147, 394, 853, 916]]<|/det|>
+As evidenced by the XRD results, the pGMA insertion into PIM- 1 membrane tends to tighten the polymer chains, leading to a reduced \(d\) - spacing. To gain further insights of microstructures, the pore size distribution of membranes was characterized using \(\mathrm{CO_2}\) adsorption experiments combined with molecular dynamics simulation. As shown in Fig. 2g- 2h, the \(\mathrm{CO_2}\) derived BET surface areas gradually decreased from 498.5 to 486.9, 432.9, and 322.5 \(\mathrm{m^2 / g}\) for PIM- 1, PIM- 1/pGMA- 7%, PIM- 1/pGMA- 27%, and PIM- 1/pGMA- 40%, respectively. The \(\mathrm{CO_2}\) uptake decreased by 55% along with a continuous reduction in the micropore surface area with the load of pGMA increasing from 7% to 40%, indicating effective regulated micropores with pGMA. The pore size distributions of pristine PIM- 1 membrane and electron beam irradiated PIM- 1/pGMA- x PRUMs derived from \(\mathrm{CO_2}\) sorption isotherms at 273 K were plotted using the nonlocal density functional theory (NLDFT) method (Fig. 2h). The peak of PIM- 1/pGMA- x PRUMs shifted to the left with the loading of pGMA increasing from 7% to 40%. The PIM- 1/pGMA- 27% PRUMs revealed a shrinkage of large microcores and a decrease of ultra- micropore region from 5 Å to 4.5 Å, which are associated with enhanced gas selectivity. A comparison between PIM- 1 membrane and PIM- 1/pGMA- 40% PRUMs revealed a decrease of the micropore from 8 Å to 7 Å, while the ultra- micropore of 5- 4 Å. Indeed, the insertion of pGMA into the PIM- 1 membrane not only regulation the micro- pores but also tailors the width of ultramicropores connecting neighboring
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 89, 852, 163]]<|/det|>
+cavities, allowing selective diffusion of smaller gas molecules such as \(\mathrm{CO_2}\) , while excluding larger gas molecules like \(\mathrm{CH_4}\) , and \(\mathrm{N}_2\) , the discussions were shown in the section below.  
+
+<|ref|>image<|/ref|><|det|>[[155, 198, 800, 565]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 581, 850, 653]]<|/det|>
+<center>Fig. 3. Simulated amorphous cells. The green area represents FFV (a) repetitive unit, (b) PIM-1, (c) PIM-1/pGMA-7%, (d) PIM-1/pGMA-15%, (e) PIM-1/pGMA-27%, (f) PIM-1/pGMA-40% (g) simulated FFV and density of polyolefin reweaved membranes. </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 678, 661, 696]]<|/det|>
+## Molecular dynamics simulation of PIM-1/pGMA-x PRUMs  
+
+<|ref|>text<|/ref|><|det|>[[147, 701, 852, 914]]<|/det|>
+Molecular dynamics simulation was used to simulate the chain packing structure and calculated FFV of PIM- 1 and PIM- 1/pGMA- x PRUMs (Fig. S8). PIM- 1 membrane showed extraordinarily high fractional free volumes (FFV) owing to their ladder polymers and contorted chain structure. The free volume distributions of the pristine and the PIM- 1/pGMA- x PRUMs were depicted in Fig. 3. As revealed, the increase in pGMA resulted in a relatively narrower free volume distribution. The simulated FFV revealed a considerable drop from 0.3246 to 0.2217 between PIM- 1 and PIM- 1/pGMA- 40%. The reduced PIM- 1/pGMA- x FFV can be attributed to the decrease of micro
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 91, 350, 108]]<|/det|>
+pores and chain packing.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 118, 465, 136]]<|/det|>
+## Gas transport properties of pure gas  
+
+<|ref|>text<|/ref|><|det|>[[147, 145, 853, 526]]<|/det|>
+The pGMA tuned micropores of the membrane was assessed through gas permeability measurements. The single gas separation performance of both pristine PIM- 1 and PIM- 1/pGMA- x PRUMs were evaluated sequentially for \(\mathrm{N}_2\) , \(\mathrm{CH}_4\) , \(\mathrm{O}_2\) , \(\mathrm{H}_2\) , and \(\mathrm{CO}_2\) at \(35^{\circ}\mathrm{C}\) and \(100\mathrm{psi}\) . The permeabilities and ideal selectivities are listed in illustrated in Fig. 4, Fig. S9 and Table S1. The pristine PIM- 1 membrane exhibited high gas permeability and moderate selectivity for all gases, which are consistent with our previous work \(^{25}\) . The slight difference is due to variations in polymer batches. Referring to post- treatment PIM- 1/pGMA- x PRUMs, a trend emerged wherein gas permeability decreased and selectivity increased with higher pGMA loading, consistent with the XRD and FFV findings. In comparing the gas permeability, a noteworthy observation was that the pristine PIM- 1 membrane exhibited a sequence of \(\mathrm{P_{CO2} > P_{H2} > P_{O2} > P_{CH4} > P_{N2}}\) , whereas in PIM- 1/pGMA- x PRUMs, the trend reversed to \(\mathrm{P_{H2} > P_{CO2}}\) and \(\mathrm{P_{N2} > P_{CH4}}\) . Such trend is in accordance with the order of gas kinetic diameters, suggesting the molecular sieving property of these membranes.  
+
+<|ref|>text<|/ref|><|det|>[[147, 534, 853, 914]]<|/det|>
+The variations in gas permeability and selectivity as a function of pGMA loading can be attributed to variations in pore distribution and simulated FFV (Fig. 3, Fig. S8). Integration of pGMA into interlayer channels of the PIM- 1 membrane served to mitigate FFV, thereby establishing ultra- selective gas channels conductive to a robust molecular sieving effect. Such as, PIM- 1/pGMA- 27% PRUMs presented a \(\mathrm{CO}_2 / \mathrm{N}_2\) selectivity 48.3 and \(\mathrm{CO}_2\) permeability of 1976 Barrer, showing a 3.6- fold increase over the pristine PIM- 1 membrane. The \(\mathrm{CO}_2 / \mathrm{CH}_4\) selectivity reached to 58.4, representing a 6.4- fold increase over the pristine PIM- 1 membrane. Compared to \(\mathrm{CO}_2 / \mathrm{CH}_4\) , \(\mathrm{O}_2 / \mathrm{N}_2\) separation is more challenging due to the closer kinetic diameters of the gas pairs. The PIM- 1/pGMA- 27% PRUMs exhibited superior \(\mathrm{O}_2 / \mathrm{N}_2\) selectivity of 6.9 and 282 Barrer of \(\mathrm{O}_2\) permeability to surpass the latest Robeson upper bound. However, as the loading of pGMA increasing to 40%, both the permeability and selectivity decreased simultaneously, the results furtherly prove that the loading of pGMA diminishes FFV and pore size, impeding the diffusion of larger gas molecules, and consequently
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 90, 432, 107]]<|/det|>
+reducing gas permeability (Fig. 4).  
+
+<|ref|>image<|/ref|><|det|>[[155, 135, 850, 664]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 673, 852, 802]]<|/det|>
+<center>Fig. 4. Gas separation performance and diffusion properties of the pristine PIM-1 and obtained PIM-1/pGMA-x polyolefin reweaved membranes. (a) \(\mathrm{CH_4}\) permeability and \(\mathrm{CO_2 / CH_4}\) selectivity, (b) \(\mathrm{CO_2}\) permeability and \(\mathrm{CO_2 / N_2}\) selectivity, (c) \(\mathrm{O_2}\) permeability and \(\mathrm{O_2 / N_2}\) selectivity vary with loading of pGMA, (d) \(\mathrm{CO_2}\) and \(\mathrm{CH_4}\) diffusivity, (e) \(\mathrm{CO_2}\) and \(\mathrm{CH_4}\) solubility, (f) sorption and diffusion selectivity of \(\mathrm{CO_2 / CH_4}\) . </center>  
+
+<|ref|>text<|/ref|><|det|>[[148, 830, 850, 905]]<|/det|>
+To understand the critical role of electron beam irradiation on the PRUM formation, we did the following control experiments. By replacing the electron beam irradiation with UV radiation for the membrane preparation, the separation performance of PIM- 1
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 853, 358]]<|/det|>
+was not enhanced, due to the energy of UV is not sufficient to polymerize of GMA (Table S1). We also physically blended pGMA with PIM- 1 membrane to form a pGMA/PIM- 1 mixed- matrix membrane. The results showed that the gas permeability decreased significantly, and the selectivity was not significantly increased (Table S1). To exclude the possible effect of electron beam irradiation on the gas separation performance of PIM- 1, we tested the gas separation performance of PIM- 1 membrane irradiated by the identical electron beam condition. The results proved that gas permeability was decreased after electron beam irradiation, while the selectivity was maintained almost. These results confirmed that pGMA could enhance the gas separation performance of PIM- 1 membrane only via electron beam irradiation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 376, 580, 395]]<|/det|>
+## Transport mechanism of PIM-1/pGMA-x PRUMs  
+
+<|ref|>text<|/ref|><|det|>[[147, 402, 853, 840]]<|/det|>
+Generally, selectivity is affected by both solubility and diffusivity selectivity, we further calculated the diffusion and solution coefficients of \(\mathrm{CO_2}\) and \(\mathrm{CH_4}\) based on the corresponding the time- lag of the constant- volume/variable- pressure method for PIM- 1 and PIM- 1/pGMA- x membranes. As shown in Fig. 4e- 4f and Table S2, compared with PIM- 1 membranes, PIM- 1/pGMA- x PRUMs showed lower solution and diffusion coefficients for both \(\mathrm{CO_2}\) and \(\mathrm{CH_4}\) molecules. While, the \(\mathrm{CO_2 / CH_4}\) diffusivity selectivity of PIM- 1/pGMA- x PRUMs were improved from 1.06 for PIM- 1 to 2.2, 3.3, 5.5, 6.9 respectively for PIM- 1/pGMA- x PRUMs. Particularly, PIM- 1/pGMA- 40% PRUMs exhibited the highest diffusivity selectivity of 6.9, almost 4.3 times as that of PIM- 1 membranes. The solubility selectivity of PIM- 1/pGMA- x membranes were also significantly enhanced compared with that of PIM- 1 membranes except for PIM- 1/pGMA- 40%. The enhanced solubility selectivity can be attributed to the reduced micropore sizes and narrow pore size distribution. The results showed that both the diffusion and solution process affect the efficient \(\mathrm{CO_2 / CH_4}\) separation in PIM- 1/pGMA- x PRUMs, consider the influence of diffusivity and solubility selectivity, the PIM- 1/pGMA- 27% PRUMs showed the best \(\mathrm{CO_2 / CH_4}\) selectivity.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[154, 90, 848, 465]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 479, 852, 662]]<|/det|>
+<center>Fig. 5. Gas separation performances of pure gas and mixed gas, anti-plasticization and aging performance. (a) the upper bond of PIM-1/pGMA-x polyolefin reweaved membranes of \(\mathrm{CO_2 / CH_4}\) , (b) \(\mathrm{CO_2 / CH_4}\) mixed-gas of the upper bond, (c) anti-plasticization properties of permeability, (d) anti-plasticization properties of relative changes in selectivity, (e) aging properties of relative changes in permeability, (f) aging properties of selectivity, (g) the universality of the electron beam irradiation induced strategy, (h) the \(^1\mathrm{H}\) NMR spectra of PI/pGMA PRUMs, (i) the \(^1\mathrm{H}\) NMR spectra of TB/pGMA PRUMs. </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 692, 597, 710]]<|/det|>
+## Separation performance comparison with literature  
+
+<|ref|>text<|/ref|><|det|>[[147, 719, 853, 905]]<|/det|>
+The gas separation performance of the pristine PIM- 1 and PRUMs in comparison with the latest Robeson's upper bound are shown in Fig. 5a and Fig. S10. The \(\mathrm{CO_2 / CH_4}\) , \(\mathrm{CO_2 / N_2}\) and \(\mathrm{O_2 / N_2}\) gas separation performance of PIM- 1 membrane are below the 2008 upper bounds<sup>37</sup>. It is noteworthy that PIM- 1/pGMA- 27 PRUMs demonstrated exceptional gas separation performance for critical gas pairs, such as \(\mathrm{CO_2 / CH_4}\) and \(\mathrm{CO_2 / N_2}\) selectivity transcended the 2019 upper bounds, and the \(\mathrm{O_2 / N_2}\) selectivity exceeded the 2015 upper bound<sup>8- 9</sup>. The results suggest that the PIM- 1/pGMA- x PRUMs
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 852, 275]]<|/det|>
+can be potentially used in industry for \(\mathrm{CO_2}\) separation and \(\mathrm{O_2}\) enrichment. The comparison of gas separation results of the present PIM- 1/pGMA- x PRUMs with the other reported PIM membranes is compiled in Table S3. The performance comparisons suggest that the electron beam irradiation induced strategy provides a new approach to developing high- performance gas separation membrane through combining electron beam irradiation and olefin monomer to precisely regulate the micropores, thus enhancing the molecular sieving effect.  
+
+<|ref|>text<|/ref|><|det|>[[147, 283, 852, 526]]<|/det|>
+The \(\mathrm{CO_2 / CH_4}\) mixed- gas testing was conducted for PIM- 1membrane and PIM- 1/pGMA- 27% PRUMs at \(35^{\circ}\mathrm{C}\) , with \(\mathrm{CO_2}\) upstream pressures ranging from 100 to 300 psi, and the results are summarized in Fig. 5b. The \(\mathrm{CO_2}\) permeability of PIM- 1 decreased with increasing upstream pressure to 300 psi with a lower \(\mathrm{CO_2 / CH_4}\) selectivity, and the performance was below the 2018 mixed- gas trade- off curve. Conversely, the \(\mathrm{CO_2}\) permeability of PIM- 1/pGMA- 27% showed only a slight decrease by increasing the upstream pressure to 300 psi, the \(\mathrm{CO_2 / CH_4}\) mixed- gas selectivity reached as high as 44.1 combining with a \(\mathrm{CO_2}\) permeability of 1702 Barrer, the performance notably exceeded the 2018 mixed- gas trade- off curve for \(\mathrm{CO_2 / CH_4}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 535, 464, 553]]<|/det|>
+## Anti-plasticizing and aging behavior  
+
+<|ref|>text<|/ref|><|det|>[[147, 562, 864, 914]]<|/det|>
+Plasticization is a common phenomenon for polymer membranes in gas separation under aggressive gas feed conditions, plasticization reduces polymer membrane separation performance, so develop plasticization resistance of a polymer membrane is urgently needed. To investigate the membranes properties concerning \(\mathrm{CO_2}\) anti- plasticizing, we conducted permeation test on PIM- 1 membrane and PIM- 1/pGMA- 27% PRUMs under varying \(\mathrm{CO_2}\) feed pressures. As shown in Fig. 5c- d and Table S4, In the case of the PIM- 1 membranes showed a noticeable plasticization behavior near 100 psi, which was reflected by an upswing in \(\mathrm{CO_2}\) permeability. This result is commonly observed in previously reported plasticization of PIM- 1<sup>36</sup>. Conversely, the PIM- 1/pGMA- 27% PRUMs showed a noticeable plasticization response at \(\sim 350\) psi, indicating highly enhanced plasticizing resistance than PIM- 1. This enhancement of \(\mathrm{CO_2}\) anti- plasticization can be attributed to the lower \(d\) - space of PIM- 1/pGMA- 27% PRUMs and reduce soluble gases \(\mathrm{CO_2}\) increases polymer free volume.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 853, 442]]<|/det|>
+PIM- 1 presents pronounced physical aging, which is one of the major obstacles for its application as a commercial membrane material for gas separation. We evaluated the stability of PIM- 1/pGMA- 27% PRUMs over a 240- day period using four pure gases: \(\mathrm{CO_2}\) , \(\mathrm{CH_4}\) , \(\mathrm{O_2}\) and \(\mathrm{N_2}\) (Fig. 5e- f and Table S5). Throughout the aging process, the membrane was kept in ambient air condition except during the gas permeation tests. It is known that the PIM- 1 membrane exhibited poor aging stability. For example, pristine PIM- 1 membrane experienced a \(64\%\) reduction in \(\mathrm{CO_2}\) permeability within just \(30\sim 90\) days<sup>37</sup>. In contrast, the \(\mathrm{CO_2}\) permeability of PIM- 1/pGMA- 27% membrane decreased only by \(11\%\) over the same duration. Notably, the anti- aging behavior of PIM- 1/pGMA- 27% PRUMs were significantly higher than that of the pristine PIM- 1. This improvement may be due to the insertion of pGMA in the PIM- 1 membrane to rigidify the polymer main chain and decreases the membrane micropore shrinkage, as evidenced by the XRD results.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 451, 682, 470]]<|/det|>
+## Universality of the electron beam irradiation induced strategy  
+
+<|ref|>text<|/ref|><|det|>[[147, 479, 853, 887]]<|/det|>
+We explored other typical olefin molecules such as acrylic acid (AA) and styrene (SE) to fabricate PIM- 1 based PRUMs through controlled irradiation dosage. The results proved that pAA and pSE also improved the separation performance of PIM- 1 membrane, as shown in Fig. 5g- i and Table S6. We extended our investigations to other typical polymeric membrane materials: 6FDA- DAM and \(\mathrm{TB}^{35 - 36, 38 - 39}\) . The results, detailed in Fig. 5g- i and Table S7- S8, showed that this strategy effectively improved the gas separation performance of these membranes as well. Similar to PIM- 1/pGMA- x PRUMs, the increasing load of pGMA shown higher selectivity compared with fresh 6FDA- DAM and TB membranes. For example, the \(\mathrm{CO_2 / CH_4}\) selectivity of 6FDA- DAM/pGMA- 21.7% PRUMs improved by \(352\%\) . The \(\mathrm{CO_2 / CH_4}\) and \(\mathrm{H_2 / CH_4}\) selectivities of TB/PGMA- 25.1% improved by \(254\%\) and \(361\%\) respectively. Given a very large scope of available polymer membranes, as well as broad small molecules to optimize the doping process, we envision that the proposed electron beam irradiation holds tremendous potential for precisely fine- tuning polyolefin reweaved membranes and facilitating scale- up efforts.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[148, 91, 242, 107]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[147, 117, 853, 500]]<|/det|>
+In summary, we demonstrated a new kind of polyolefin reweaved ultramicropor membranes (PRUMs) through electron beam irradiation strategy that leverages electron beam irradiated in situ polymerization of small molecules to precisely regulate the microporous structure of membrane materials. By varying the load of pGMA polymer, we sought to strike a balance between separation performance and polymer loading, experimental data and molecular simulation revealed that the increase of pGMA reduced both FFV and gas permeability. Ultramicropor distribution was shifted to be more concentrated sub- nanometer range, which fits well for precisely discriminating small gas molecules such as \(\mathrm{CO_2O_2}\) , \(\mathrm{N_2}\) and \(\mathrm{CH_4}\) . Typically, the \(\mathrm{CO_2 / CH_4}\) selectivity of PIM- 1/pGMA- 27%, 6FDA- DAM/pGMA- 21.7%, TB/pGMA- 25.1% improved by 642%, 352% and 254% respectively. Our electron beam irradiation strategy to PRUM is not only applicable to a wide range of polymers and small molecules, but also offers multiple operational advantages such as time- independent thermal treatment, controllable operations and processes, high efficiency, and lack of catalysts or initiators.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 509, 217, 523]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 536, 223, 550]]<|/det|>
+## Materials  
+
+<|ref|>text<|/ref|><|det|>[[147, 561, 853, 857]]<|/det|>
+4,4'- Diamino- 3,3'- dimethylbiphenyl, diethoxyethane, glycidyl methacrylate (GMA) were purchased from Aladdin and used as applied without further purification. 5,5'6,6'- Tetrahydroxy- 3,3,3',3'- tetramethyl- 1,1'- spirobisindane (TTSBI), 4,4'- diamino- 3,3'- dimethylbiphenyl and diethoxyethane were bought from Energy Chemical, 2,3,5,6- Tetrafluoroterephthalonitrile (TFTPN) was bought from Sigma- Aldrich. TTSBI was purified by crystallization with tetrahydrofuran and \(n\) - hexane. 2,3,5,6- TFTPN was purified by vacuum sublimation at \(150^{\circ}\mathrm{C}\) , anhydrous potassium carbonate \((\mathrm{K}_2\mathrm{CO}_3\) , Aladdin company) was dried in a vacuum oven at \(120^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) and stored in a desiccator before use. \(N\) - methyl- 2- pyrrolidone (NMP, Aladdin company), and toluene (Aladdin company) were dried using metallic sodium to remove moisture before use. All test gases (>99.9999%) employed for the evaluation of membraneseparation performance were supplied by Air Liquide (Taiyuan).  
+
+<|ref|>text<|/ref|><|det|>[[148, 867, 850, 911]]<|/det|>
+Preparation of PIM- 1, 6FDA- DAM, TB and electron beam irradiated polyolefin reweaved membranes
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[146, 90, 853, 528]]<|/det|>
+PIM- 1 6FDA- DAM, and TB polymers were synthesized according to previously reported procedures \(^{35,38}\) , as shown in Fig. S1- 3. Dense polymer membranes with a thickness of \(50 \mu \mathrm{m}\) (±10 um) were prepared from 7 wt.% polymer solutions in chloroform. Polymer solutions were filtered through \(0.45 \mu \mathrm{m}\) polypropylene filters and then cast into Teflon Petri dishes in a glovebox and allowed to evaporate slowly for \(48 \mathrm{~h}\) . The dry PIM- 1, 6FDA- DAM, TB membranes were placed into PET bags ( \(8 \mathrm{~cm} \times 10 \mathrm{~cm}\) ) equipped \(10\%\) GMA methanol solution ( \(10 \mathrm{wt} \% \mathrm{GMA} / 90 \mathrm{wt} \%\) methanol) overnight to ensure the GMA solution was dispersed evenly throughout the membrane. Afterwards, these PET bags were irradiated by \(1 \mathrm{MeV}\) electronic accelerator at room temperature with different absorbed doses. The reaction parameters were varied to desired ranges ( \(20 - 80 \mathrm{kGy}\) absorbed dose in irradiation ETFE films for \(10 \mathrm{min}\) ). At the end of the irradiation reaction, the irradiated PET bags were put into a water bath at \(40^{\circ} \mathrm{C}\) for \(3 \mathrm{~h}\) to make GMA polymerization more complete. The membranes were removed ultrasonically with lots of methanol to eliminate the homopolymer and the GMA monomer, then dried under vacuum at \(80^{\circ} \mathrm{C}\) for \(24 \mathrm{~h}\) to acquire PIM- 1/pGMA- x polyolefin reweaved membranes, 6FDA- DAM/pGMA- x polyolefin reweaved membranes, TB/pGMA- x polyolefin reweaved membranes (x refers to the load of pGMA). The load of pGMA (DG) on the membranes was calculated as follows equation:  
+
+<|ref|>text<|/ref|><|det|>[[147, 536, 761, 554]]<|/det|>
+DG \((\%) = (\mathrm{W}1 - \mathrm{W}0) / \mathrm{W}0\times 100\%\) (1)  
+
+<|ref|>text<|/ref|><|det|>[[147, 564, 794, 582]]<|/det|>
+where Wo and W1 represent the weights of the initial and grafted membranes, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[148, 592, 850, 636]]<|/det|>
+The PIM- 1/pGMA- x ( \(x = 7\%\) , \(15\%\) , \(27\%\) , \(40\%\) ) polyolefin reweaved membranes were obtained via radiation doses \(20 \mathrm{kGy}\) , \(40 \mathrm{kGy}\) , \(60 \mathrm{kGy}\) or \(80 \mathrm{kGy}\) or \(80 \mathrm{kGy}\) respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 648, 279, 662]]<|/det|>
+## Characterization  
+
+<|ref|>text<|/ref|><|det|>[[147, 673, 853, 913]]<|/det|>
+1 MeV electronic accelerator was purchased from USA (Wasik Associates Inc., USA). The \(^1 \mathrm{H}\) NMR spectrums were measured at a spectrometer of \(400 \mathrm{MHz}\) in \(\mathrm{CDCl}_3\) (Bruker AVANCE- III), and the chemical shifts were recorded in ppm. Thermogravimetric analysis (TGA) was used to investigate the thermal stabilities of polymer membranes using a TA SDT Q600. Attenuated total reflection Fourier- transform infrared (ATR- IR) spectra (AVATAR 360, Thermo Nicolet, USA) were recorded from \(500 \mathrm{to} 4000 \mathrm{cm}^{- 1}\) with \(32 \mathrm{scans}\) and \(4 \mathrm{cm}^{- 1}\) resolutions for membranesamples. Scanning electronic microscopy/Energy- dispersive X- ray spectroscopy mapping (SEM/EDS) was applied to analysis the surface morphology and elemental information and density of TB/pGMA- x polyolefin reweaved membranes by a Hitachi S5500 microscope. \(\mathrm{CO}_2\) adsorption- desorption was performed
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 89, 852, 219]]<|/det|>
+using a Micromeritics ASAP 2020 instrument. X- ray diffraction (XRD) was used to study the \(d\) - spacing changes of the PIM- 1/pGMA- x polyolefin reweaved membranes and the results were recorded on a Bruker AXS GADDS apparatus using Cu radiation with a wavelength of 1.54 Å. The testing angle range was \(5 - 40^{\circ}\) at a rate of \(4^{\circ} / \mathrm{min}\) and \(d\) - spacing was computed following Bragg's law \((d = \lambda /2\sin \theta)\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 230, 385, 246]]<|/det|>
+## Gas permeation measurements  
+
+<|ref|>text<|/ref|><|det|>[[147, 256, 853, 441]]<|/det|>
+The gas separation performance of PIM- 1/pGMA- x, 6FDA- DAM/pGMA- x, TB/pGMA- x membranes were tested at \(35^{\circ}\mathrm{C}\) using a constant- volume/variable- pressure method. The downstream pressure was measured using a transducer down to \(10^{- 6}\) torr, and the steady state pressure changing with time (dp/dt) was selected to calculate the permeability (P). Pure gas was tested in the sequence of \(\mathrm{N}_2\) , \(\mathrm{CH}_4\) , \(\mathrm{O}_2\) , \(\mathrm{CO}_2\) and \(\mathrm{H}_2\) at 80 psi. PIM- 1/pGMA- x and TB/pGMA- x polyolefin reweaved membranes were tested for three different samples and the deviation was less than \(5\%\) . Permeability (P) was calculated using the following equation 2:  
+
+<|ref|>equation<|/ref|><|det|>[[163, 444, 825, 479]]<|/det|>
+\[P = 10^{10}\times \frac{V_{d}\times l}{p_{up}\times T\times R\times A}\times \frac{dp}{dt} \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \quad (2)\]  
+
+<|ref|>text<|/ref|><|det|>[[147, 487, 852, 616]]<|/det|>
+Where \(P\) is the permeability (Barrer), 1 Barrer \(= 10^{- 10}\mathrm{cm}^3 (\mathrm{STP})\mathrm{cm}\mathrm{cm}^{- 2}\mathrm{s}^{- 1}\mathrm{cmHg}^{- 1}\) , \(V_{d}\) is the calibrated permeate volume \((\mathrm{cm}^3)\) , \(l\) is the membrane thickness \((\mathrm{cm})\) , \(p_{up}\) is the upstream pressure \((\mathrm{cmHg})\) , \(A\) is the effective membrane area \((\mathrm{cm}^2)\) , \(T\) is the operating temperature \((\mathrm{K})\) , \(R\) is the gas constant \((0.278\mathrm{cm}^3\mathrm{cmHg}\mathrm{cm}^{- 3}(\mathrm{STP})\mathrm{K}^{- 1})\) and \(dp / dt\) is the steady- state downstream pressure increase rate \((\mathrm{cmHg}\mathrm{s}^{- 1})\) .  
+
+<|ref|>text<|/ref|><|det|>[[147, 626, 852, 672]]<|/det|>
+The ideal selectivity \((\alpha_{x / y})\) for components x and y was defined as the ratio of gas permeability of the two components via equation 3.  
+
+<|ref|>equation<|/ref|><|det|>[[147, 680, 822, 710]]<|/det|>
+\[\alpha_{x / y} = \frac{p_{x}}{p_{y}} \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \quad (3)\]  
+
+<|ref|>text<|/ref|><|det|>[[147, 720, 777, 737]]<|/det|>
+The diffusion coefficient D was calculated from the time- lag apparatus using equation 4:  
+
+<|ref|>text<|/ref|><|det|>[[147, 746, 822, 764]]<|/det|>
+\(\mathrm{D} = 1^{2} / 6\epsilon\) (4)  
+
+<|ref|>text<|/ref|><|det|>[[147, 775, 764, 793]]<|/det|>
+\(\epsilon\) denotes the time lag. solubility coefficient (S) was obtained indirectly via equation 5:  
+
+<|ref|>text<|/ref|><|det|>[[147, 802, 824, 820]]<|/det|>
+\(\mathrm{S} = \mathrm{P} / \mathrm{D}\) (5)  
+
+<|ref|>sub_title<|/ref|><|det|>[[147, 831, 312, 847]]<|/det|>
+## Molecular simulation  
+
+<|ref|>text<|/ref|><|det|>[[147, 858, 850, 904]]<|/det|>
+To verify the packing change in the structural after the electron beam irradiation process, PIM- 1/pGMA- 7%- 40% polyolefin reweaved membranes were built by the amorphous cell based on
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 853, 274]]<|/det|>
+lamps calculations. PIM- 1 amorphous cell module consists of 6 polymer chains with 20 repeating units each, and pGMA amorphous cell module consists of 6 polymer chains with 20 repeating units each. The force field is pcf, and all processes are carried out under the NPT system, first relaxation at high temperature and pressure (800 K, 500 bar) 2ns, the equilibrium stage temperature was set to 300 K and 1.5 bar. The actual load of pGMA is slightly different from the calculated load, the load of the calculation pGMA is PIM- 1/pGMA- 7.17%, PIM- 1/pGMA- 15.63%, PIM- 1/pGMA- 26.08%, PIM- 1/pGMA- 40.45%.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 285, 318, 302]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[148, 311, 852, 385]]<|/det|>
+We acknowledge the financial support from National Natural Science Foundation of China (22378102), special fund for the Key Laboratory of Hubei Province (2022ZX02 and 2022ZX04).  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 396, 334, 413]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[147, 422, 853, 552]]<|/det|>
+All authors contributed to the scientific discussion and manuscript preparation. Xiuling Chen and Lei Wu led the experimental design, data curation, and writing of the pristine manuscript. Guining Chen performed the transport and sorption analyses of membranes. Cong Xie performed the electron beam irradiation. Nanwen Li conceived the concept of the research, Gongping Liu and Wanqin Jin helped to revise the manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 562, 325, 579]]<|/det|>
+## Competing interests  
+
+<|ref|>text<|/ref|><|det|>[[149, 590, 503, 607]]<|/det|>
+The authors declare no competing interests.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 618, 293, 634]]<|/det|>
+## Data availability  
+
+<|ref|>text<|/ref|><|det|>[[149, 645, 468, 662]]<|/det|>
+Data will be made available on request.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 677, 245, 694]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[147, 712, 850, 881]]<|/det|>
+[1] Rochelle, G. T. Amine scrubbing for \(\mathrm{CO_2}\) capture. Science 325, 1652- 1654 (2009). [2] Dai, Y. et al. A review on the recent advances in composite membranes for \(\mathrm{CO_2}\) capture processes. Sep. Purif. Technol. 307, 122752 (2023). [3] Koros, W. J. & Zhang, C. Materials for next- generation molecularly selective synthetic membranes. Nat. Mater. 16, 289- 297 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 94, 849, 149]]<|/det|>
+[4] Ockwig, N. W. & Nenoff, T. M. Membranes for hydrogen separation. Chem. Rev. 107, 4078- 4110 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[147, 168, 850, 260]]<|/det|>
+[5] Nikolaeva, D. et al. The performance of affordable and stable cellulose- based poly- ionic membranes in \(\mathrm{CO_2 / N_2}\) and \(\mathrm{CO_2 / CH_4}\) gas separation. J. Membr. Sci. 564, 552- 561 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[147, 280, 849, 335]]<|/det|>
+[6] Murphy, T. M. et al. Physical aging of polystyrene films tracked by gas permeability. Polymer 54, 873- 880 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[147, 354, 850, 409]]<|/det|>
+[7] Chanem, B. S. et al. High- performance membranes from polyimides with intrinsic microporosity. Adv. Mater. 20, 2766- 2771 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[147, 428, 850, 520]]<|/det|>
+[8] Swaidan, R., Ghanem, B. & Pinnau, I. Fine- tuned intrinsically ultramicroporous polymers redefine the permeability/selectivity upper bounds of membrane- based air and hydrogen separations. ACS Macro Lett. 4, 947- 951 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[147, 539, 851, 631]]<|/det|>
+[9] Comesaña- Gandara, B. et al. Redefining the Robeson upper bounds for \(\mathrm{CO_2 / CH_4}\) and \(\mathrm{CO_2 / N_2}\) separations using a series of ultrapemable benzotriptycene- based polymers of intrinsic microporosity. Energy Environ. Sci. 12, 2733- 2740 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[147, 650, 850, 742]]<|/det|>
+[10] He, S. et al. Recent progress in PIM- 1 based membranes for sustainable \(\mathrm{CO_2}\) separations: Polymer structure manipulation and mixed matrix membrane design. Sep. Purif. Technol. 284, 120277 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[147, 761, 850, 854]]<|/det|>
+[11] Hazazi, K. et al. Precise molecular sieving of ethylene from ethane using triptycene- derived submicroporous carbon membranes. Nat. Mater. 22, 1218- 1226 (2023).  
+
+<|ref|>text<|/ref|><|det|>[[147, 873, 849, 891]]<|/det|>
+[12] Bandehali, S. et al. Polymers of intrinsic microporosity and thermally rearranged
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 94, 850, 150]]<|/det|>
+polymer membranes for highly efficient gas separation. Sep. Purif. Technol. 278, 119513 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[147, 168, 851, 261]]<|/det|>
+[13] Balcik, M. et al. Exploring the effect of intra- chain rigidity on mixed- gas separation performance of a Triptycene- Tröger's base ladder polymer (PIM- Trip- TB) by atomistic simulations. J. Membr. Sci. 677, 121614 (2023).  
+
+<|ref|>text<|/ref|><|det|>[[147, 279, 851, 372]]<|/det|>
+[14] Yu, G. et al. Mixed matrix membranes derived from nanoscale porous organic frameworks for permeable and selective \(\mathrm{CO_2}\) separation. J. Membr. Sci. 591, 117343 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[147, 390, 851, 483]]<|/det|>
+[15] Chen, X. et al. Tailoring the microporosity of polymers of intrinsic microporosity for advanced gas separation by atomic layer deposition. Angew. Chem. Int. Ed. 60, 17875- 17880 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[147, 501, 850, 557]]<|/det|>
+[16] Terrones, M. et al. Coalescence of single- walled carbon nanotubes. Science 288, 1226- 1229 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[147, 575, 851, 668]]<|/det|>
+[17] Bushell, A. F. et al. Gas permeation parameters of mixed matrix membranes based on the polymer of intrinsic microporosity PIM- 1 and the zeolitic imidazolate framework ZIF- 8. J. Membr. Sci. 427, 48- 62 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[147, 686, 851, 780]]<|/det|>
+[18] Ghalei, B. et al. Enhanced selectivity in mixed matrix membranes for \(\mathrm{CO_2}\) capture through efficient dispersion of amine- functionalized MOF nanoparticles. Nature, Energy 2, 17086 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[147, 798, 850, 854]]<|/det|>
+[19] He, G. et al. Synergistic \(\mathrm{CO_2}\) - sieving from polymer with intrinsic microporosity masking nanoporous single- layer graphene. Adv. Funct. Mater. 30, 2003979 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[147, 872, 850, 890]]<|/det|>
+[20] Ma, C. & Urban, J. J. Hydrogen- bonded polyimide/metal- organic framework
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 94, 850, 149]]<|/det|>
+hybrid membranes for ultrafast separations of multiple gas pairs. Adv. Funct. Mater. 29, 1903243 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[147, 168, 853, 333]]<|/det|>
+[21] Ji, W. et al. Facile tailoring molecular sieving effect of PIM- 1 by in- situ \(\mathrm{O}_3\) treatment for high performance hydrogen separation. J. Membr. Sci. 662, 120971 (2022). [22] Satilmis, B. et al. Hydroxyalkylaminoalkylamide PIMs: selective adsorption by ethanolamine- and diethanolamine- modified PIM- 1. Macromolecules 48, 5663- 5669 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[147, 353, 851, 446]]<|/det|>
+[23] Rodriguez, K.M. et al. Leveraging free volume manipulation to improve the membrane separation performance of amine- functionalized PIM- 1. Angew. Chem. Int. Ed. 60, 6593- 6599 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[147, 464, 850, 520]]<|/det|>
+[24] Jung, D. et al. Postsynthetically modified polymers of intrinsic microporosity (PIMs) for capturing toxic gases. ACS Appl. Mater. Interf. 13, 10409- 10415 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[147, 539, 851, 631]]<|/det|>
+[25] Pinnau, I. et al. Recent Progress on Polymers of Intrinsic Microporosity and Thermally Modified Analogue Materials for Membrane- Based Fluid Separations. Small Struct. 2, 2100049 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[147, 650, 851, 743]]<|/det|>
+[26] Li, F.Y. et al. High- performance thermally self- cross- linked polymer of intrinsic microporosity (PIM- 1) membranes for energy development. Macromolecules 45, 1427- 1437 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[147, 761, 850, 817]]<|/det|>
+[27] He, Y. et al. Polymers with side chain porosity for ultrapermeable and plasticization resistant materials for gas separations. Adv. Mater. 31, 1807871 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[147, 835, 849, 890]]<|/det|>
+[28] Du, N. et al. Polymer nanosieve membranes for \(\mathrm{CO}_2\) - capture applications. Nat. Mater. 10, 372- 375 (2011).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 94, 850, 188]]<|/det|>
+[29] Tien- Binh, N. et al. In- situ cross interface linking of PIM- 1 polymer and UiO- 66- NH2 for outstanding gas separation and physical aging control. J. Membr. Sci. 548, 429- 438 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[147, 205, 850, 298]]<|/det|>
+[30] Song, Q. et al. Controlled thermal oxidative crosslinking of polymers of intrinsic microporosity towards tunable molecular sieve membranes. Nat. Commun. 4, 1918 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[147, 316, 850, 372]]<|/det|>
+[31] Song, Q. & Sivaniah, E. Photo- oxidative enhancement of polymeric molecular sieve membranes. Nat. Commun. 5, 4813 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[147, 390, 850, 446]]<|/det|>
+[32] Li, F. et al. UV- rearranged PIM- 1 polymeric membranes for advanced hydrogen purification and production. Adv. Energy Mater. 2, 1456- 1466 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[147, 464, 850, 520]]<|/det|>
+[33] Oh, B. et al. Electron- mediated control of nanoporosity for targeted molecular separation in carbon membranes. Nat. Commun. 13, 4972 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[147, 538, 850, 630]]<|/det|>
+[34] Dong, Z. et al. Recent progress in environmental applications of functional adsorbent prepared by radiation techniques: A review. J. Hazard. Mater. 424, 126887 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[147, 649, 850, 705]]<|/det|>
+[35] Wu, L. et al. Enhanced molecular selectivity and plasticization resistance in ring- opened Tröger's base polymer membranes. J. Membr. Sci. 634, 119399 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[147, 723, 850, 817]]<|/det|>
+[36] Zhao, S. et al. Blending of compatible polymer of intrinsic microporosity (PIM- 1) with Tröger's Base polymer for gas separation membranes. J. Membr. Sci. 566, 77- 86 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[147, 835, 850, 891]]<|/det|>
+[37] Alberto, M. et al. Impeded physical aging in PIM- 1 membranes containing graphene like fillers. J. Membr. Sci. 563, 513- 520 (2018).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 94, 850, 191]]<|/det|>
+[38] Chen, X. et al. Ultra- selective molecular- sieving gas separation membranes enabled by multi- covalent- crosslinking of microporous polymer blends. Nat. Commun. 12, 6140 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[147, 206, 850, 261]]<|/det|>
+[39] Chen, X. et al. Hydrogen bonding- induced 6FDA- DABA/TB polymer blends for high performance gas separation membranes. J. Membr. Sci. 655, 120575 (2022).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[42, 92, 768, 112]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 150, 148]]<|/det|>
+- Sl.docx
+
+<--- Page Split --->

preprint/preprint__7ed793eacd675c022e3c4a1a6684d41fe242f54bbdfd7daa74a9cf06d91b8d3e/images_list.json

@@ -0,0 +1,107 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. Engineering the \\(N\\) -hydroxy pipecolic acid (NHP) biosynthetic pathway in Arabidopsis.",
+    "footnote": [],
+    "bbox": [
+      [
+        163,
+        128,
+        840,
+        456
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Proteomics analysis reveals a potential regulator of NHP biosynthesis. (A) Label-free proteomics identified 724 proteins showing differential abundance in pNHP plants as compared to WT ones. A total of 4587 proteins (pink color) were identified and quantified. Of these, 656 (yellow color) and 68 (brown color) proteins were up-regulated or down-regulated in pNHP relative to WT plants. (B) Venn diagram showing overlapping transcripts/proteins between the differentially accumulated proteins found in this study and NHP response genes obtained from available transcriptomics data (16). The blue, orange, and brown colors represent the number of NHP-induced genes (NHP+), differentially accumulated proteins in pNHP plants (pNHP), and NHP-repressed genes (NHP-), respectively. (C) A heatmap displaying 725 differentially accumulated proteins in pNHP plants relative to their expression in WT samples. The color scale represents log2 fold change. Each row in the color heat map represents a single protein. Two different pNHP independent lines are shown (marked as 1 and 2) (D) GO annotation of proteins detected as differentially accumulating in pNHP plants as compared to the WT ones. (E) Relative",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        84,
+        840,
+        422
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. Pathogen-dependent induction of NAC90 is mediated by NHP, SA, and SARD1. (A)",
+    "footnote": [],
+    "bbox": [
+      [
+        220,
+        80,
+        765,
+        751
+      ]
+    ],
+    "page_idx": 38
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4. Characterization of NAC90 mutants and overexpression plants. (A) and (B) Transcript levels of NAC90 in leaves of 3-week-old wild-type (WT), nac90 mutants (nac90-1 and nac90-2) and NAC90 overexpression (NAC90-1 and NAC90-2) plants. (C) Western blot (Anti-Flag) showing NAC90 protein abundance in the leaves of 4-week-old WT and NAC90 overexpression plants (NAC90-1 and NAC90-2). (D) Phenotype of WT, nac90-1, nac90-2, and NAC90",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        83,
+        832,
+        700
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5. NAC90 negatively regulates plant immunity by controlling the NHP and SA biosynthetic pathways. (A-D) Transcript levels of NHP (ALD1, SARD4, and FMO1) and SA (ICS1) biosynthetic genes in the leaves of 4-week-old WT, nac90-1, nac90-2, and NAC90 overexpression (NAC90-1 and NAC90-2) plants. (E) Chromatin immunoprecipitation-qPCR assay indicated that NAC90 directly binds to the upstream region of NHP biosynthetic genes. (F) Dual luciferase promoter activity assays showed that NAC90 represses the expression of NHP biosynthetic genes. (G-H) Abundance of NHP (G) and SA (H) in the leaves of 4-week-old WT, nac90-1, nac90-2,",
+    "footnote": [],
+    "bbox": [
+      [
+        149,
+        80,
+        840,
+        630
+      ]
+    ],
+    "page_idx": 42
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6. FMO1 is required for constitutive activation of defense response in the nac90 mutant background. (A) FMO1 determines the size of the rosette in nac90-1 plant. Four-week-old wild-type (WT), fmo1, nac90-1, nac90-1 fmo1, and nac90-1 sid2 plants was examined for their morphology. Bar = 1 cm. (B) and (C) Abundance of NHP (B) and SA (C) in the leaves of 4-week-old WT, fmo1, nac90-1, nac90-1 fmo1, and nac90-1 sid2 plants. (D) Growth of Psm ES4326 in the leaves of WT, fmo1, nac90-1, nac90-1 fmo1, and nac90-1 sid2 plants. (E) Symptoms detected in leaves of WT, fmo1, nac90-1, nac90-1 fmo1, and nac90-1 sid2 plants treated with Psm ES44326.",
+    "footnote": [],
+    "bbox": [
+      [
+        144,
+        85,
+        844,
+        650
+      ]
+    ],
+    "page_idx": 44
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Figure 7. A triad of NAC transcription factor cooperate in the regulation of NHP biosynthesis. (A)",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        80,
+        825,
+        780
+      ]
+    ],
+    "page_idx": 46
+  }
+]

preprint/preprint__7ed793eacd675c022e3c4a1a6684d41fe242f54bbdfd7daa74a9cf06d91b8d3e/preprint__7ed793eacd675c022e3c4a1a6684d41fe242f54bbdfd7daa74a9cf06d91b8d3e.mmd

@@ -0,0 +1,513 @@
+
+# A NAC Triad Mediates Plant Immunity by Negatively Regulating N-Hydroxy Pipelcolic Acid Biosynthesis  
+
+Asaph Aharoni  
+
+asaph.aharoni@weizmann.ac.il  
+
+Weizmann Institute of Science https://orcid.org/0000- 0002- 6077- 1590  
+
+Jianghua Cai  
+
+Weizmann Institute of Science  
+
+Sayantan Panda  
+
+Weizmann Institute of Science  
+
+Yana Kazachkova  
+
+Weizmann Institute of Science  
+
+Eden Amzallag  
+
+Hebrew university of Jerusalem  
+
+Sagit Meir  
+
+Weizmann Institute of Science https://orcid.org/0000- 0001- 7102- 3152  
+
+Ilana Rogachev  
+
+Weizmann institute of science  
+
+## Article  
+
+Keywords: N- hydroxy pipelcolic acid, Salicylic acid, NAC transcription factor, Plant immunity, Label- free proteomics, Transcriptome  
+
+Posted Date: July 24th, 2023  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 3133457/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 22nd, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 51515- 2.
+
+<--- Page Split --->
+
+# A NAC Triad Mediates Plant Immunity by Negatively Regulating N-Hydroxy  
+
+## 2 Pipelcolic Acid Biosynthesis  
+
+3 Jianghua Cai<sup>1</sup>, Sayantan Panda<sup>1,2</sup>, Yana Kazachkova<sup>1</sup>, Eden Amzallag<sup>3</sup>, Sagit Meir<sup>1</sup>, Ilana Rogachev<sup>1</sup>, Asaph Aharoni<sup>1,\*</sup>  
+
+5 1. Department of Plant and Environmental Sciences, Weizmann Institute of Science, Rehovot 6 761001, Israel  
+
+7 2. Current address: Department of Cell and Metabolic Biology, Leibniz Institute of Plant 8 Biochemistry, Weinberg 3, Halle (Saale), 06120, Germany  
+
+9 3. The Robert H. Smith Institute of Plant Sciences and Genetics in Agriculture, The Hebrew 10 University of Jerusalem, Rehovot, 7610001, Israel  
+
+11 \*Correspondence: Asaph Aharoni (asaph.aharoni@weizmann.ac.il)  
+
+## 13 Abstract  
+
+14 N- hydroxy- pipelcolic acid (NHP) plays an important role in plant immunity. In contrast to its 15 biosynthesis, our current knowledge with respect to the transcriptional regulation of the NHP 16 pathway is limited. This study commenced with the engineering of Arabidopsis plants that 17 constitutively produce high NHP levels and display enhanced immunity. Label- free proteomics 18 revealed a NAC- type transcription factor (NAC90) that is strongly induced in these plants. We 19 found that NAC90 is a target gene of SAR DEFICIENT 1 (SARD1) and induced by pathogen, 20 salicylic acid (SA), and NHP. NAC90 knockout mutants exhibit constitutive immune activation, 21 earlier senescence, higher levels of NHP and SA, as well as increased expression of NHP and SA
+
+<--- Page Split --->
+
+biosynthetic genes. In contrast, NAC90 overexpression lines were compromised in disease resistance and accumulated reduced levels of NHP and SA. NAC90 could interact with NAC61 and NAC36 which are also induced by pathogen, SA, and NHP. We next discovered that this protein triad directly represses expression of the NHP and SA biosynthetic genes AGD2- LIKE DEFENSE RESPONSE PROTEIN 1 (ALDI), FLAVIN MONOOXYGENASE 1 (FMOI), and ISOCHORISMATE SYNTHASE 1 (ICS1). Constitutive immune response in nac90 is abolished once blocking NHP biosynthesis in the fmo1 background, signifying that NAC90 negative regulation of immunity is mediated via NHP biosynthesis. Our findings expand the currently documented NHP regulatory network suggesting a model that together with NHP glycosylation, NAC repressors take part in a 'gas- and- brake' transcriptional mechanism to control NHP production and the plant growth and defense trade- off.  
+
+Keywords: N- hydroxy pipecolic acid, Salicylic acid, NAC transcription factor, Plant immunity, Label- free proteomics, Transcriptome
+
+<--- Page Split --->
+
+## Introduction  
+
+Plant defense responses are initiated by the recognition of pathogen- associated molecular patterns (PAMPs) or effectors molecules, which as a consequence result in activation of PAMP- triggered immunity (PTI) or effector- triggered immunity (ETI), respectively<sup>1</sup>. In recent years, outstanding strides have been made in dissecting the regulatory mechanisms and components necessary for plant defense. Following attack by pathogens, salicylic acid (SA) accumulates in plants<sup>2, 3</sup>. External application of SA or its chemical analogues can enhance the plants resistance to various diseases, whereas plants lacking SA biosynthesis, perception, or signal transduction display severely compromised resistance and failure in systemic acquired resistance (SAR) establishment<sup>4, 5, 6</sup>. Apart from SA, several additional metabolites have been implicated in plant defense response, including methyl salicylate (MeSA)<sup>7</sup>, azelaic acid (AzA)<sup>8</sup>, glycerol- 3- phosphate (G3P)<sup>9</sup>, dehydroabietinal (DA)<sup>10</sup>, \(N\) - hydroxy pipecolic acid (NHP)<sup>11, 12</sup>, \(\beta\) - aminobutyric acid (BABA)<sup>13</sup>, L- glutamate<sup>14</sup>, and melatonin<sup>15</sup>.  
+
+NHP is a non- protein amino acid derived from lysine that systemically accumulates in plants during pathogen infection and is essential for plant defense activation and SAR establishment<sup>11, 12</sup>. Significant progress has been made in understanding the NHP biosynthetic pathway. It starts from L- lysine and involves three enzymes: AGD2- LIKE DEFENSE RESPONSE PROTEIN 1 (ALD1), SAR- DEFICIENT 4 (SARD4), and FLAVIN MONOOXYGENASE 1 (FMO1). Lysine is converted by ALD1 to form 2,3- dehydropipecolic acid, which is then reduced by SARD4 to generate pipecolic acid (Pip)<sup>16</sup>. Next, FMO1 hydroxylates Pip to form NHP<sup>11, 12</sup>. The expression of these three NHP biosynthetic genes was strongly induced in plants in response to pathogen infection, resulting in high levels of NHP accumulation<sup>17</sup>. Plants with ALD1 and FMO1 loss- of- function mutations block NHP biosynthesis, making them more susceptible to pathogens and
+
+<--- Page Split --->
+
+having a weaker SAR response<sup>11, 12, 16</sup>. Moreover, NHP and its biosynthetic genes have been found in a variety of plant species, and external treatment with synthetic NHP or overaccumulation of NHP through transiently overexpressing ALD1 and FMO1 can enhance disease resistance against various pathogens in different plants<sup>18</sup>. Furthermore, exogenous application of NHP in soil can restore the defense response in NHP deficient ald1 and fmo1 Arabidopsis mutants, partially rescue the response in salicylic acid induction- deficient 2 (sid2), a mutant in ISOCHORISMATE SYNTHASE 1 (ICS1), but fails to restore the defense response in the non- expression of pr1 (npr1) mutant<sup>12</sup>. The transcript levels of NHP, SA, phytoalexin and camalexin biosynthetic genes, as well as a series of defense- related genes could be effectively activated by exogenous NHP treatment<sup>19</sup>. Additionally, NHP can be further converted to NHP- glycoside (NHPG) and NHP- glucose ester (NHPGE), both of which are also considerably induced by pathogen infection<sup>20</sup>. The UDP- glycosyltransferase 76B1 (UGT76B1) is the enzyme that catalyzes the glycosylation of NHP to NHPG and is essential for NHP- triggered SAR response in Arabidopsis and tomato<sup>21, 22, 23, 24</sup>. Moreover, UGT76B1- mediated NHP homeostasis is critical to balance growth and defense trade- off in plants<sup>21</sup>. Therefore, NHP is a widespread and important metabolite in the activation of plant defense and SAR.  
+
+Over the last several years, significant progress has been made in understanding the regulation of NHP biosynthesis. Following pathogen infection, NHP levels were found to be much higher in sid2- 1 and npr1 mutants, but much lower in enhanced disease susceptibility (eds1) and phytoalexin- deficient 4 (pad4) mutants<sup>20</sup>. MITOGEN- ACTIVATED PROTEIN KINASE (MAPK) activation was observed to trigger the expression of ALD1 and FMO1, resulting in higher levels of NHP<sup>25</sup>. Constitutive expression of \(Ca^{2 + }\) - DEPENDENT PROTEIN KINASES 5 (CPK5) in Arabidopsis elevated the expression of NHP biosynthetic genes as well as NHP levels<sup>26</sup>. Apart from these, several transcriptional regulators have been demonstrated to play critical roles in
+
+<--- Page Split --->
+
+controlling NHP biosynthesis. For example, pathogen induced expression of ALD1, SARD4, and FMO1 was greatly decreased in sar deficient1 (sard1) and cam- binding protein 60- like g (cbp60g) double mutant (sard1/cbp60g) \(^{27}\) . SARD1 and CBP60g were shown to regulate NHP biosynthesis by direct binding to the promoters of NHP biosynthetic genes ALD1, SARD4, and FMO \(^{127}\) . WRKY33 is another positive regulator of NHP biosynthesis, which acts directly on the ALD1 promoter \(^{25}\) . Pathogen induced NHP as well as the expression of ALD1 and FMO1 were significantly impaired in the wrky33 mutant \(^{25}\) . TGACG- BINDING FACTOR 1 (TGA1), TGA4, WRKY70, and WRKY54 were also found to positively regulate NHP biosynthesis by activating the expression of SARD1 or CBP60g, whereas CALMODULIN- BINDING TRANSCRIPTION ACTIVATOR (CAMTA1/2/3) were proposed to indirectly regulate NHP biosynthesis in a negative manner by repressing the expression of SARD1 and CBP60g \(^{28, 29, 30}\) . However, a direct negative regulator controlling NHP biosynthesis was not identified so far.  
+
+In this study, we identified a NAC- type transcription factor (i.e., NAC90) that negative controls the NHP biosynthetic pathway. NAC90 was previously reported as a negative regulator of leaf senescence by directly inhibiting SA biosynthesis \(^{31}\) . Here, we showed that NAC90 protein abundance was significantly enhanced in Arabidopsis plants constitutively producing high NHP levels. We discovered that through interaction with two other NAC factors (i.e., NAC36 and NAC61) NAC90 negatively regulates plant immunity via direct suppression of both NHP and SA biosynthesis. Blocking NHP or SA biosynthesis suppressed the constitutive activation of immune response in the nac90 mutant background. Activity of this NAC- type triad of negative regulators complements the recently reported role of NHP glycosylation through UGT76B1 in balancing NHP levels and consequently the plant growth and defense trade- off.  
+
+## Results
+
+<--- Page Split --->
+
+## Engineering constitutive and high level NHP production in plants  
+
+We previously reported transient expression of the three NHP biosynthetic genes ALD1, SARD4, and FMO1 from Arabidopsis in N. benthamiana leaves<sup>21</sup>. As a continuation of this previous research, we generated stable transformants of Arabidopsis and tomato (Solanum lycopersicum) plants (termed pNHP and tpNHP, respectively) expressing the three Arabidopsis genes. Both Arabidopsis and tomato transgenic lines displayed severe phenotypes. The pNHP plants developed a smaller rosette (reduced diameter) and showed an early senescence phenotype as compared to wild- type (WT) Arabidopsis plants (Figure 1A and 1B; Supplementary Fig 1A). Trypan blue staining showed that pNHP and tpNHP leaves were stained much darker than WT ones<sup>32</sup>, indicating cell death symptoms. Moreover, even though tpNHP plants developed flowers, they failed to set fruit (data not shown). Liquid chromatography- mass spectrometry (LC- MS) analysis showed significant accumulation of NHP in pNHP and tpNHP leaves while it was not detectable in WT ones. (Figure 1C and Supplementary Figure 1B). The glycosylated form of NHP (NHPG) also accumulated to considerable amount in the transgenic plants (Supplementary Figure 1C and 1D). In our previous study we showed that exogenous application of NHP could enhance SA accumulation in Arabidopsis<sup>21</sup>. To confirm this SA enhancement in vivo (i.e., not through exogenous NHP application), we examined SA production in pNHP plants and found that its level was significantly enhanced (Figure 1D). Next, we assayed the expression of NHP biosynthetic genes (ALD1, SARD4, and FMO1) and the defense marker gene- PATHOGENESIS- RELATED GENE 1 (PR1) and found that all of them were dramatically enhanced in pNHP as compared to WT plants (Figure 1E, 1F, 1G, and 1H). These results suggested that engineering NHP production can constitutively activate immune responses in Arabidopsis and tomato plants.
+
+<--- Page Split --->
+
+## Proteomics of pNHP plants results in the identification of NAC90 as a putative regulator of NHP biosynthesis  
+
+To understand the changes in the proteome of NHP over- producing Arabidopsis, we performed a label- free proteomics analysis of leaves derived from pNHP- 1 and pNHP- 2 (independent transgenic lines) and WT plants (Supplementary Figure 2A). A total of 4587 proteins were identified in this experiment (Figure 2A, Supplementary Table S1); out of which 827 defined as Differentially Accumulated Proteins (DAP; 2- fold significant difference set as threshold, \(p\) value \(< 0.05\) ). Compared to WT samples, 738 proteins were up- regulated and 89 were down- regulated in pNHP- 1, while 807 proteins were up- regulated and 110 were down- regulated in pNHP- 2 (Supplementary Figure 2B). Overall, a total of 656 proteins were significantly up- regulated, and 68 proteins were down- regulated in both pNHP- 1 and pNHP- 2 lines (Figure 2A; see proteins list in Supplementary Table S1).  
+
+To identify DAPs that are transcriptionally altered by NHP, we compared the list of DAPs with the NHP- responsive genes<sup>19</sup>. Notably, 352 proteins out of 724 comprised 329 NHP- induced (NHP+) and 23 NHP- repressed (NHP-) genes (Figure 2B and Supplementary Table S2). Moreover, we also found 477 DAPs that overlapped with SAR responsive genes (407 SAR+ and 70 SAR-) (Supplementary Figure 2C and Supplementary Table S3)<sup>33</sup>. A total of 300 proteins were encoded by both NHP and SAR responsive genes (Supplementary Figure 2D). Next, we explored gene ontology (GO) categories enrichment among the DAPs. The GO analysis results showed that the four largest enriched groups were “Response to stimulus”, “Response to chemical”, “Response to biotic stimulus” and “Defense response” (Figure 2D). Moreover, “systemic acquired resistance”, and “ubiquitin related process” which are important for plant immunity were also highly enriched categories among the DAPs (Figure 2D). In addition, a set of known proteins involved in defense response were also identified as DAPs, including PR1, PR5, ACCELERATED CELL DEATH 6
+
+<--- Page Split --->
+
+(ACD6), DOWNY MILDEW RESISTANCE 6 (DMR6), NUDIX HYDROLASE 6, MITOGEN- ACTIVATED PROTEIN KINASE (MPK11), MITOGEN- ACTIVATED PROTEIN KINASE KINASE (MKK4), and RPM1 INTERACTING PROTEIN 4 (RIN4) (Supplementary Table S1). Furthermore, ENHANCED DISEASE SUSCEPTIBILITY 1 (EDS1), PHYTOALEXIN- DEFICIENT 4 (PAD4), and MPK3 involved in NHP and SA biosynthesis and signal transduction<sup>12, 25</sup>, were also found to be among the DAPs (Supplementary Table S1). In addition to these proteins, several transcription factors, including CAMTA3, CBP60b, and NAC90, were found to be significantly induced in pNHP plants, (Figure 2E, Supplementary Figure 2D and Supplementary Table S1). CAMTA3 and CBP60b have been reported to regulate plant immunity through CBP60g<sup>29, 34, 35</sup>. NAC90 was reported to bind to the promoter of ICS1 and negatively regulate SA biosynthesis during senescence<sup>31</sup>. Here, NAC90 protein abundance was dramatically induced (more than 100 times) in pNHP plants and it also appeared among the list of NHP- and SAR- induced genes (see above). We thus decided to focus on the characterization of the NAC90 transcription factor and its involvement in NHP biosynthesis and pathogen response.  
+
+## Expression of NAC90 is induced by pathogen and directly regulated by SARD1  
+
+Consistent with the changes in protein abundance, NAC90 transcript level was also significantly enhanced in both pNHP- 1 and pNHP- 2 plants as compared to WT leaves (Figure 3A). We next examined the transcript levels of NAC90 in response to bacterial pathogen (Psm ES4326), NHP, and SA. Samples harvested at 24- hour post inoculation (hpi) showed that NAC90 expression was strongly up- regulated following bacterial infection, NHP, and SA treatments (Figure 3B). NAC90 transcript level was also evaluated in a series of mutants defective in NHP and SA defense response pathways including the NHP- deficient fmo1, salicylic acid induction- deficient (sid2), non- expression of PR1 (npr1), enhanced disease susceptibility 1 (eds1), and phytoalexin- deficient
+
+<--- Page Split --->
+
+4 (pad4) mutants. The results showed that the basal (i.e., mock treatment) transcript level of NAC90 in fmo1, eds1, and pad4 was lower than in WT (Figure 3C). Following Psm ES4326 infection, the induction of NAC90 transcript level was severely reduced in fmo1, sid2, npr1, eds1, and pad4 mutants (Figure 3C). These results indicate that FMO1, ICS1, NPR1, EDS1, and PAD4 positively modulate the expression of NAC90 following pathogen infection. The UDP glycosyltransferase UGT76B1 serves as a negative regulator of defense immunity by controlling the levels of NHP and \(\mathrm{SA}^{21, 22, 23, 24}\) . We analyzed NAC90 transcript level in the ug176b1 mutant background that over- accumulates NHP and SA. A substantial increase of NAC90 in the ug176b1 mutant relative to WT suggested that UGT76B1 activity negatively affects the expression of NAC90 (Figure 3C).  
+
+SARD1 and CBP60g are two master regulators that directly control SA and NHP biosynthesis and thus play important roles in plant immunity<sup>27, 29</sup>. We therefore examined NAC90 expression in sard1, cbp60g, and sard1 cbp60g plants. As shown in Figure 3D, following Psm ES4326 infection, NAC90 expression was notably induced in WT, and this induction was significantly reduced in sard1 and sard1 cbp60g but not in cbp60g plants, suggesting that SARD1 may regulate NAC90 expression. Furthermore, examining the 2,000 bp upstream region of the NAC90 gene sequence identified three GAAATT motifs, which are like the previously reported SARD1 binding elements (Sun et al., 2015; Supplementary Figure 3A). In addition, SARD1 chromatin immunoprecipitation sequencing revealed a high signal upstream of the NAC90 translation start site (Supplementary Figure 3B)<sup>27</sup>, suggesting that NAC90 is the direct target gene of SARD1. To further support this assumption, we performed dual- luciferase reporter assays in N. benthamiana leaves by co- expressing a reporter construct of the firefly luciferase (LUC) driven by the NAC90 upstream region (NAC90upsr:LUC) and an effector construct expressing the SARD1 or CBP60g protein (Supplementary Figure 3C). The results suggested that SARD1 could activate
+
+<--- Page Split --->
+
+the NAC90 upstream region (Figure 3E), further supporting that NAC90 is directly regulated by SARD1.  
+
+## Growth phenotypes and immunity-related transcriptome changes of NAC90 altered plants  
+
+To further explore the function of NAC90, we identified two Arabidopsis mutants with T- DNA insertions in its coding region [nac90- 1 (SALK_206203) and nac90- 2 (SALK_203643)]. (Supplementary Figure 4A). We also generated NAC90 constitutive overexpression lines (NAC90- 1 and NAC90- 2) in which the NAC90 coding sequence is fused to a 3×Flag tag. NAC90 transcript was barely detectable in leaves of these two nac90 mutants, whereas both overexpression lines displayed a significantly higher transcript levels as compared to WT ones (Figure 4A and 4B). Western blotting with anti- Flag confirmed the abundance of NAC90 protein expression in both NAC90- 1 and NAC90- 2 lines (Figure 4C). Phenotypically, nac90 mutant plants exhibited small rosettes and early senescence, while NAC90 overexpression lines developed enlarged rosettes (Figure 4D). Consistent with these phenotypes, the nac90 mutants above- ground parts weight and rosettes diameter were considerably reduced as compared to those of WT plants, whereas they were notably bigger in the NAC90 overexpression plants (Supplementary Figure 4B and 4C).  
+
+To investigate the impact of NAC90 knockout on gene expression, we performed transcriptome analysis of WT and nac90- 2 leaves. A total of 1,480 Differentially Expressed Genes (DEGs, 2- fold significant difference set as threshold, \(p\) value \(< 0.05\) ) including 818 up- regulated and 662 down- regulated were found in the nac90- 2 mutant (Supplementary Figure 4D and Supplementary Table S4). We also re- analyzed the transcriptome data that examined global gene expression in the nac90- 1 mutant<sup>31</sup> and found that 716 genes, including 497 up- regulated and 219 down- regulated, were differentially expressed in the nac90- 1 mutant (Supplementary Figure 4D
+
+<--- Page Split --->
+
+and Supplementary Table S5). We compared the transcriptome datasets from the two nac90 mutants and identified 176 common DEGs (including 32 down-regulated and 147 up-regulated genes) in these two mutant alleles relative to the WT plants leaves (Supplementary Figure 4E and Supplementary Table S6). To determine the specific cellular processes associated with these 176 DEGs, we performed enrichment analysis of Gene Ontology (GO) biological processes. We found that the functional classes “Response to stimulus”, “Defense response”, “Response to bacterium”, and “Response to hormone” were the four enriched groups containing the largest number of genes (Figure 4E). In addition, 31 and 11 DEGs were classified as “Response to salicylic acid” and “Leaf senescence” processes, respectively. Among these 42 (i.e., 31 + 11) DEGs, those encoding defense markers (PR1, CHI, and SAG13), NHP biosynthesis (ALD1 and FMO1), SA transport (EDS5), and the two pathogen-induced master regulators of NHP and SA biosynthesis (SARD1 and CBP60g) were substantially up-regulated in both the nac90- 1 and nac90- 2 mutants (Figure 4F). Notably, two other NAC-type transcription factors (NAC61 and NAC36) were up-regulated in both knockout mutant lines (Figure 4F) suggesting that they might play a role in plant immunity.  
+
+## NAC90 is a negative regulator of NHP and SA biosynthesis  
+
+The transcriptome data described above showed that the expression of ALD1, FMO1, and ICS1 was significantly up- regulated in both nac90 mutants (Figure 4F). The results were confirmed by Real Time- PCR (RT- PCR) assays showing that transcript levels of ALD1, SARD4, FMO1, and ICS1 were significantly higher in the nac90 mutants than in WT, but lower in NAC90 overexpression lines (Figure 5A, 5B, 5C and 5D). We next analyzed the upstream region of these genes and found that they contained from 4 and up to 9 NAC transcription factor binding motifs (Supplementary Figure 5A), suggesting that NAC90 may bind to the promoters of these genes. To
+
+<--- Page Split --->
+
+further investigate whether NAC90 could directly bind to the upstream regions of NHP biosynthetic genes in vivo, we performed chromatin immunoprecipitation coupled with quantitative RT- PCR (ChIP- qPCR) analysis for NHP biosynthetic genes. ICS1 was previously reported to be directly repressed by NAC90 31 and hence used as a positive control in this experiment. Compared to the negative control IgG, the putative promoter fragments of ALD1 and FMO1 were significantly enriched by Anti- Flag, indicating that NAC90 can bind to these regulatory regions (Figure 5E and Supplementary Figure 5B). NAC90 does not appear to bind the putative SARD4 promoter although a trend in enrichment was detected (which was not statistically significant). We next performed dual- luciferase reporter assays in N. benthamiana leaves cotransfected with reporter constructs of firefly luciferase driven by the upstream regions of NHP biosynthetic genes (i.e. ALD1upsr::LUC, SARD4upsr::LUC, and FMO1upsr::LUC) and an effector construct expressing the NAC90 protein. After two days, a significant reduction in expression driven by the putative promoter regions of the three NHP biosynthetic genes and ICS1 was observed in the samples with NAC90 (Figure 5F and Supplementary Figure 5D). The results indicated that NAC90 directly binds to the promoters of NHP and SA biosynthetic genes to repress their expression.  
+
+Since NAC90 directly represses the expression of NHP and SA biosynthetic genes, we investigated the levels of Pip, NHP, and SA in leaves of WT, nac90 mutants and NAC90 overexpression lines. LC- MS analysis showed that levels of these three metabolites were significantly elevated in the leaves of nac90 mutants as compared to WT and NAC90 overexpression lines. In contrast, the levels of Pip and SA in the leaves of NAC90 overexpression plants were much lower than those in WT plants (Figure 5G, 5H, and Supplementary 5E). In addition, the levels of NHPG and salicylic acid beta- glucoside (SAG), the glycosylated derivatives
+
+<--- Page Split --->
+
+of NHP and SA, were also much higher in the leaves of nac90 mutants than in WT ones, whereas they were significantly lower in NAC90 overexpression lines (Supplementary Figure 5F and 5G).  
+
+## NAC90 mutants exhibit constitutive activation of immune response  
+
+We then investigated whether alteration of NAC90 expression affects plant immunity. We treated Arabidopsis leaves of WT, nac90 mutants, and NAC90 overexpression lines with Psm ES4326 and analyzed bacterial growth three days later. The Psm bacterial titers in nac90 mutants were significantly lower than those in WT plants leaves and were notably higher in NAC90 overexpression plants (Figure 5I and 5J). RT- PCR assays results showed that the transcript level of the defense marker gene PRI was significantly enhanced in nac90 mutants and suppressed in NAC90 overexpression as compared to WT plants (Figure 5K). The results indicated that nac90 mutants display constitutive activation of immune response. Taken together, the data suggests that NAC90 functions as a negative regulator of plant immunity via suppressing NHP and SA biosynthesis.  
+
+## Knockout of FMO1 suppresses the constitutive activation of immune response in nac90 mutant  
+
+The above results showed that nac90 mutants displayed constitutive activation of immune response and accumulated higher levels of NHP and SA (Figure 5). To examine whether the NHP or SA biosynthesis pathway is responsible for the activation of immune response in nac90 mutants, fmo1 or sid2 mutants were crossed with nac90- 1 to create nac90- 1 fmo1 and nac90- 1 sid2 double mutants. As shown in Figure 6A and Supplementary Figure 6A, the dwarfed morphology of nac90- 1 can be completely abolished and resemble WT plants in the nac90- 1 fmo1 double mutant
+
+<--- Page Split --->
+
+and could be partially reverted to a WT phenotype in the nac90- 1 sid2 double mutants. Next, we examined the Pip, NHP, and SA levels in these plants. The results showed that nac90- 1 accumulated a considerable amount of Pip and NHP while no NHP and lower Pip levels were observed in WT, fmo1, nac90- 1 fmo1, and nac90- 1 sid2 plants leaves (Figure 6B and Supplementary Figure S6B). In terms of SA, lower levels of SA accumulated in fmo1, sid2, nac90- 1 fmo1, and nac90 sid2 plants as compared to WT, whereas nac90- 1 plants accumulated higher levels of SA (Figure 6C). A similar pattern of metabolite accumulation was observed for NHPG and SAG levels (Supplementary Figure 6C and 6D). Furthermore, we investigated disease resistance in these genotypes and found that the enhanced disease resistance against Psm ES4326 in nac90- 1 was abolished in the nac90- 1 fmo1 and nac90- 1 sid2 double mutants (Figure 6D and 6E). In agreement with these findings, RT- PCR analysis showed that the increased transcript levels of PR1 observed in nac90- 1 was also repressed in the nac90- 1 fmo1 and nac90- 1 sid2 double mutants (Figure 6F). In addition, we could not detect the nac90- 1 early senescence phenotype in the nac90- 1 fmo1 and nac90- 1 sid2 double mutant (Data not shown). Thus, we provide several points of evidence that FMO1 and ICS1 are required for the constitutive activation of immune responses and early senescence in the nac90- 1 mutants.  
+
+## The NAC triad cooperates to repress NHP biosynthesis and plant immunity  
+
+To identify additional regulators involved in NHP biosynthesis, we performed gene coexpression analysis using the three NHP biosynthetic genes (ALD1, SARD4, and FMO1) as baits against the expression data of all transcription factors from the previously reported transcriptomics experiments of (i) Psm- treated WT, (ii) Psm- treated ald1 and sid2 mutants<sup>33</sup>, and (iii) Pip- treated fmo1 mutants<sup>12</sup>. Consequently, one hundred transcription factors were co- expressed with all three
+
+<--- Page Split --->
+
+baits with an \(r\) value greater than 0.9. Among these transcription factors, NAC and WRKY were the two largest families (Figure 7A). As WRKY33 was shown to be a positive regulator of NHP biosynthesis (Wang et al., 2018), we focused on NAC- type factors. Phylogenetic analysis of this this family revealed that NAC90, NAC61, and NAC36 (all three co- expressed) belong to the same subgroup (Supplementary Figure 7). In addition, NAC61 and NAC36 were significantly induced by Psm ES4326, NHP, and SA treatments (Supplementary Figure 8A and 8B), as well as up- regulated in both nac90- 1 and nac90- 2 mutants as noted above (Figure 4F).  
+
+To investigate the role of NAC61 and NAC36 in plant immunity, we generated nac61 and nac36 mutants (using CRISPR/Cas9). As shown in Figure 7B and 7C, similar to the nac90 mutants, nac36 mutants (i.e., nac36- 1 and nac36- 2) displayed reduced growth, whereas no differences in morphology were observed in the nac61 mutants (i.e., nac61- 1 and nac61- 2). Moreover, nac36 mutants showed higher resistance to Psm ES4326, whereas no significant difference was observed in disease resistance of the nac61 mutants (Figure 7D and Supplementary Figure 9A). The transcript levels of PR1, ALD1, SARD4, FMO1, and ICS1 were also significantly elevated in both nac61 and nac36 mutants as compared to WT plants (Supplementary Figure 9B, 9C, 10A, and 10B). Consistent with the transcript levels of ALD1, FMO1 and ICS1, the levels of defense- related metabolites (NHP, SA, Pip, NHPG and SAG) were also much higher in both nac61 and nac36 mutants than those detected in WT (Figure 7E, 7F, and Supplementary Figure 9D, 10C), indicating that NAC61 and NAC36 also function as negative regulators of NHP biosynthesis.  
+
+NAC- type transcription factors form homodimer or heterodimers to regulate target gene expression<sup>36, 37</sup>. To examine whether NAC90, NAC61, and NAC36 can interact with each other, we performed a coimmunoprecipitation assay. As shown in Figure 7G, NAC61- HA and NAC36-
+
+<--- Page Split --->
+
+HA could be co- precipitated with NAC90- Flag or NAC36- Flag by an anti- Flag antibody. We next performed a split luciferase complementation imaging assay in N. benthamiana leaves. Strong luminescence was observed in the co- transformations of NAC90 and NAC61, NAC61 and NAC36, as well as NAC36 and NAC90, whereas no signals were detected in the combination of negative controls (Figure 7H, 7I, and 7J). This indicated that the NAC triad can form heterodimers between each other and likely cooperate to repress NHP biosynthesis.  
+
+## Discussion  
+
+Activation of plant defence leads in some responses to simultaneous biosynthesis of NHP and SA, a pair of signalling molecules with essential functions in plant immunity<sup>38</sup>. However, constitutive production of NHP and SA cannot endure infinitely as they activate defence programs that negatively correlate with plant growth. Considerable progress has been made in understanding the mechanisms that govern plant defence and growth. Many of these processes are controlled by hormone signaling pathways and interconnecting regulatory networks. In our previous work, we reported that UGT76B1- mediated NHP glycosylation plays a critical role in regulating the levels of NHP, resulting in balancing the trade- off between plant growth and defence<sup>21</sup>. In the present study, we discovered that a triad of NAC transcription factors functions as negative regulators of NHP biosynthesis and plays a vital role in controlling this trade- off by directly repressing NHP accumulation. Together with the transcriptional activators reported previously by others, this triad joins the mechanism mediated by NHP glycosylation. Yet, the above activities are likely part of a more intricate process controlling NHP levels in time and space that might include, non- coding RNAs, epigenetic, and post- transcriptional, - translational mechanisms.  
+
+NHP and SA are the two key metabolites essential for plant immunity and SAR<sup>12, 34</sup>. Pathogen attack results in a significant increase in the levels of these two metabolites, which in
+
+<--- Page Split --->
+
+turn activate the plant immunity to enhance the resistance against pathogens \(^{17, 38}\) . Substantial evidence indicates the presence of shared co- regulation molecular mechanisms between SA and NHP, especially in biosynthesis, transport, and metabolism. EDS1 and PAD4 are essential components necessary for the expression of SA and NHP biosynthetic genes, accumulation of NHP and SA, as well as the execution of their functions \(^{12, 39}\) . The production of SA and NHP can also be regulated by \(\mathrm{Ca}^{2 + }\) through the modulation of CPKs, CBP60a/CBP60g, and CAMTA1/2/3 activities \(^{38}\) . SARD1 and CBP60g serve as the master regulators in plant immunity as they exhibit direct binding abilities to the promoters of both SA (ICS1, and PBS3) and NHP biosynthetic genes \(^{27}\) . EDS5 is responsible for the transport of both SA and Pip from the chloroplast to the cytoplasm \(^{40}\) . UGT76B1 is the essential enzyme responsible for the glycosylation of both NHP and \(\mathrm{SA}^{21, 22, 23, 24}\) . Apart from the above co- regulated mechanisms, there are also distinct mechanisms that vary between SA and NHP. For instance, despite the fact that the SA receptor NPR1 is essential for NHP- mediated immunity, there was no discernible interaction found between NPR1 and NHP \(^{41}\) . WRKY33 directly binds to the promoter of ALD1 to enhance NHP biosynthesis, but represses SA biosynthesis \(^{25}\) . In this study, we found that a NAC triad plays a significant role as in suppressing the accumulation of both NHP and SA. This is achieved through the direct inhibition of their biosynthetic genes.  
+
+Earlier research reported that NHP application enhances the expression of more than 1,500 genes in Arabidopsis \(^{19}\) . These were associated with a variety of defence responses, including the accumulation of SA, phytoalexins, and branched- chain amino acid biosynthesis. In this study, we performed proteomics analysis of pNHP plants and identified 724 DAPs between pNHP and WT plants. A large number of proteins, including many key immune regulators PR1, PR5, EDS1, PAD4, RIN4, ACD6, DMR6, MPK11, and MPK \(^{342, 43, 44, 45, 46}\) were found to overlap with SAR
+
+<--- Page Split --->
+
+and NHP- induced genes. In our study of plants constitutively producing NHP, many proteins involved in ubiquitination related processes were highly enriched, suggesting significant involvement of such mechanisms in NHP- induced plant immunity. Indeed, protein homeostasis maintained through ubiquitination has been reported to play an important role in plant immunity and many key defence regulators are degraded through this pathway. For example, NPR3 and NPR4 act as the adaptors of Cullin3 (CUL3)- based E3 ligase to mediate the degradation of the master regulators of plant defence NPR1 and EDS1<sup>47, 48, 49</sup>. Apart from ubiquitination processes, numerous proteins associated with other processes were also significantly enriched in our pNHP plants, some of these including transporters and signaling factors require deeper investigation.  
+
+As we found NAC90 to act as a negative regulator, we questioned its hierarchy in the established regulatory network controlling NHP biosynthesis (Figure 7K). The results showed that NAC90 expression following pathogen infection was considerably reduced in the fmo1, sid2, npr1, eds1, pad4, and sard1 mutants, suggesting that these genes are required for NAC90 function. Additionally, ChIP- sequencing and promoter activity assays provided convincing evidence for NAC90 being a direct target of SARD1. Furthermore, FMO1, and ICS1 transcripts level was significantly enhanced in the nac90 mutant background and suppressed in NAC90 overexpression plants, suggesting a regulatory negative feedback loop between NAC90 and these genes. We detected higher transcript levels of NHP and SA biosynthetic genes (ALD1, SARD4, FMO1, and ICS1), as well as higher levels of Pip, NHP, and SA in nac90 mutants, and the opposite in NAC90 overexpression plants. Moreover, we found that NAC90 directly represses the expression of ALD1, FMO1, and ICS1.  
+
+NAC transcription factors have already been reported to act together as a brake in regulating downstream target pathways. NAC019, NAC055, and NAC072 negatively regulate SA
+
+<--- Page Split --->
+
+accumulation by directly suppressing ICS1 and activating SALICYLIC ACID GLUCOSYLTRANSFERASE, a gene encoding enzyme involved in SA glycosylation<sup>50</sup>. A NAC troika (NAC017, NAC082, and NAC90) acts a negative regulator of leaf senescence by coordinately suppressing both SA and reactive oxygen species pathways<sup>31</sup>. Here, we showed that the transcript levels of NAC36 and NAC61 were significantly enhanced in both nac90 mutants, and they can interact with each other. Similarly, we also observed higher levels of NHP and SA, as well as enhanced resistance against Psm ES4326 in nac36 mutants. However, we did not find any differences in the resistance against Psm ES4326 in nac61 mutants. Considering these different findings, it is possible that the NHP and SA levels in nac61 mutants are insufficient to initiate defence response. Thus, these three transcription factors exhibit partial functional redundancy and together act to directly suppress the biosynthesis of NHP and SA.  
+
+Taken together, we propose a model in which a NAC triad plays a significant role in the trade- off between plant growth and defence response. During pathogen attack, the NHP and SA biosynthetic pathways are activated and both metabolites accumulate to high levels. Consequently, defence response is enhanced but also early senescence and growth inhibition is triggered. However, at a certain threshold, the three NAC factors are transcriptionally induced and in turn function as a 'hand brake' to decrease NHP and SA levels via transcriptional repression of the corresponding biosynthetic genes (Figure 7K). This negative control of gene expression prevents harmful effects of excessive NHP and SA levels. Discovery of the regulatory NAC proteins here expands the array of molecular genetic tools currently available for manipulating crop plants to better sustain pathogen attacks with reduced penalty in terms of biomass and yield.  
+
+## Materials and methods  
+
+## Plant Material and Growth Condition
+
+<--- Page Split --->
+
+Arabidopsis (Arabidopsis thaliana) ecotype Columbia- 0 (Col- 0) was used as wild- type (WT). The nac90- 1 and nac90- 2 T- DNA insertion mutants (SALK_206203 and SALK_203643, respectively) were obtained from the Arabidopsis Biological Resource Center (ABRC) and genotyped by PCR using gene- and T- DNA- specific oligonucleotides. Seeds of nac90- 1, nac90- 2, fmo112, sid251, npr151, pad439, and eds139were surface sterilized with 70% ethanol for 4 min and 3% (v/v) bleach and then washed with sterile water. Subsequently, seeds were sown on half MS agar medium with 1 % sucrose and stratified at 4 °C for 2 days. Arabidopsis seedlings with two true leaves were transplanted to soil in a long day growth chamber (16 hours light). Four to five weeks- old adult plants were used for all the experiments.  
+
+## Bacterial strains and Growth  
+
+Pseudomonas syringae pv. maculicola (Psm) strain ES4326 was grown at 28 °C in King B agar medium containing 50 μg.L- 1 rifampicin and 50 μg.L- 1 kanamycin52. For bacterial infection, an overnight log phase culture was collected by centrifugation at 4,000 g, washed three times with 10 mM MgCl2 and diluted to a final optical density of 0.001 (at OD600 nm) for infection53.  
+
+## Vectors Construction and plant transformation  
+
+The open reading frame of NAC90 was amplified from Arabidopsis leaves cDNA by PCR using specific oligonucleotides and cloned into the overexpression vector pALPHA2- NptII- UBQ10pro- CCD- Ter10 using golden gate reaction. For CRISPR plasmid construction, four guide RNAs specifically targeting the NAC61 and NAC36 were designed using CRISPR- P (http://cbi.hzau.edu.cn/crispr/)54. After two rounds of amplification, the four sgRNA expression cassettes were cloned into the pAGM55261- CCD binary vector using the golden gate ligation method55, 56. The overexpression and CRISPR vectors were introduced to the A. tumefaciens
+
+<--- Page Split --->
+
+GV3101 strain using the heat shock method and next transformed to Arabidopsis by the flower dipping method<sup>57</sup>. Primers used for plasmid construction were listed in Supplementary Table S7.  
+
+## Label-free proteomics  
+
+Label- free proteomics was performed as described previously<sup>58</sup>. In brief, leaves were harvested from 28 d old WT, \(pNHP1\) , and \(pNHP2\) plants and milled into fine power in liquid nitrogen. About 100 mg leaf powder was lysed in 300 μl lysis buffer (4% (w/v) sodium dodecyl sulfate, 100 mm Tris/HCl pH 7.6, 0.1 M dithiothreitol), boiled at 95 °C 5 min, and then centrifugated at 16,000 g for 15 min. Subsequently, the supernatants were collected and subjected to protein concentrations determination using Thermo Scientific Pierce 660 nm protein assay kit. Fifty micrograms of protein from each sample were reduced, alkylated and digested on a centrifugal filter unit (10 kDa MWCO) according to the filter- aided sample preparation (FASP) protocol<sup>59</sup>. The tryptic peptides were acidified with trifluoroacetic acid, desalted, and dried using a speed vacuum. The label- free proteomics experiment was performed in three independent biological replicates. The peptides were dissolved in 97:3 H<sub>2</sub>O: acetonitrile + 0.1 % formic acid, and then separated using an HSS T3 nano- column (75 μm internal diameter, 250 mm length, 1.8 μm particle size; Waters) at 0.35 μL min<sup>- 1</sup>. Peptides were then eluted using the following gradient: 4- 35% B in 105 min; 35- 90% B in 5 min; maintained at 90% for 5 min; and then back to initial conditions. The mass spectrometry analysis was performed using a Nano- UPLC system equipped with a quadrupole orbitrap mass spectrometer (Q Exactive Plus, Thermo Scientific). Data was acquired in a data dependent acquisition mode using the Top20 method. MS1 mass range was 300- 1750 m/z, resolution was set to 60 000 (at 400 m/z), AGC target 1e<sup>6</sup> and maximum injection time was set to 120 msec. MS2 resolution was set to 17 500, isolation window of 2 m/z, underfill ratio 1.3%, AGC target of 5e<sup>5</sup> and maximum injection time of 100 msec. Dynamic exclusion was 120
+
+<--- Page Split --->
+
+msec. MS raw data was processed using the Proteome Discoverer 2.4 software (Thermo Fisher Scientific) for protein identification and quantification with \(1\%\) false discovery rate (FDR) against the TAIR 10 proteins dataset. To identify proteins that were significantly different between WT and \(pNHP\) plants, a twofold cut- off was used as a determinant. Go terms were analysed in the TAIR website (https://www.arabidopsis.org/tools/go_term_enrichment.jsp).  
+
+## Metabolite extraction and quantification  
+
+Metabolites were extracted from leaf samples according to previously reported method<sup>21</sup>. Briefly, 100 mg Arabidopsis and tomato leaf tissue were flash frozen in liquid nitrogen immediately after collection, ground to fine power using a ball mill (Retsch MM400), resuspended in \(400 \mu \mathrm{l}\) of \(80\%\) MeOH: \(\mathrm{H}_2\mathrm{O}\) (v/v), sonicated for 20 min at room temperature, and centrifugated at 13,000 g for 15 min. The supernatants were harvested, filtered through a \(0.22 \mu \mathrm{m}\) polyvinylidene fluoride (PVDF) syringe- driven filter (Millipore, Billerica, MA), and subjected to liquid chromatography- tandem mass spectrometry (LC- MS/MS) analysis. The analyses of metabolites were conducted on Xevo TQ- S micro Triple Quadrupole Mass Spectrometry (Waters) connected to UPLC instrument (Premier system, Waters). MS was operating in a Multiple Reaction Monitoring (MRM) mode for relative quantification of NHP and SA derivatives (Pip, NHP, NHPG, SA, and SAG). A \(1.8 \mu \mathrm{m}\) , \(2.1 \mathrm{mm} \times 100 \mathrm{mm}\) ACQUITY UPLC HSS T3 column (Waters) was used for metabolites separation. The mobile phase A consisted of \(0.1\%\) formic acid in \(1\%\) aqueous acetonitrile and mobile phase B was \(0.1\%\) formic acid in \(100\%\) acetonitrile. The following linear gradient was used with a flow rate of \(0.3 \mathrm{ml / min}\) (percentages indicate percentage A): \(0\text{- } 3\mathrm{min}\) \((99.9\%)\) , \(3\text{- } 9\mathrm{min}\) \((99.9\% - 92\%)\) , \(9\text{- } 9.2\mathrm{min}\) \((92\% - 0\%)\) , \(9.2\text{- } 11.8\mathrm{min}\) \((0\% - 0\%)\) , \(11.8\text{- } 12\) min \((0\% - 99.9\%)\) , \(12\text{- } 14\mathrm{min}\) \((99.9\%)\) . Mass spectra were collected within a mass range of 50- 1600 in the positive ionization mode for NHP derivatives and in the negative ionization mode for SA
+
+<--- Page Split --->
+
+derivatives. The following instrument settings were applied: capillary \(= 1.5 \mathrm{kV}\) and cone \(= 15 \mathrm{V}\) for positive ionization (1 kV and 27 V for negative ionization); source temperature, \(140 ^{\circ} \mathrm{C}\) ; desolvation, \(450 ^{\circ} \mathrm{C}\) ; desolvation gas flow, \(800 \mathrm{l / h}\) . Argon was used as the collision gas. NHP was relatively quantified in positive mode using quantification transition (QT) \((146.08 > 110.06)\) and verification transition (VT) \((146.08 > 100.07)\) . For NHPG, QT was \(308.13 > 146.08\) , and VT was \(308.13 > 110.06\) . SA was relatively quantified in negative mode using transition \(137.02 > 93.03\) , and VT \(93.03 > 65.03\) . For SAG, QT was \([299.07 > 137.02]\) , VT was \([299.07 > 93.03]\) . TargetLynx (Water, UK) was used for data analysis and relative quantification. A mixture of Pip, NHP, NHPG, SA, and SAG standards was processed along the samples and served as a positive quality control.  
+
+## Protein Extraction and western blot  
+
+Western blot were conducted following the previously reported protocol<sup>60</sup>. Hundred mg leaf samples were ground and heated at \(95 ^{\circ} \mathrm{C}\) in \(2 \times\) protein loading buffer for 5 min. Protein samples were collected after centrifugation, run on an \(12 \%\) SDS- PAGE gel, transferred to PVDF membranes and blocked with \(5 \%\) BSA in \(1 \times\) TBST for \(1 \mathrm{h}\) at room temperature followed by incubation with the Anti- Flag (Abmart, China) primary antibody in \(5 \%\) BSA for \(1 \mathrm{h}\) . PVDF membranes were washed four times with \(1 \times\) TBST and incubated with horseradish peroxidase- conjugated secondary antibody for \(1 \mathrm{h}\) . SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific) was used for chemiluminescence signal.  
+
+## Dual luciferase promoter activity assays  
+
+Dual luciferase promoter activity assays were performed according to the reported method with several modifications<sup>61</sup>. About \(2 \mathrm{kb}\) upstream from the transcription start site of target genes was amplified from Arabidopsis leaves DNA and introduced upstream of the firefly luciferase
+
+<--- Page Split --->
+
+reporter gene in the pCambia 1302 plasmid harbouring a renilla luciferase gene under the control of the CaMV35S promoter as an internal control. The open reading frame of NAC90, SARD1, and CBP60g were amplified and cloned into the effector plasmid pALPHA2- Npt II- UBQ10- CCDB- Ter10 using the infusion method (Vazyme, China). The empty effector plasmid was used as the negative control. The destination reporters and effectors were co- transformed into N. benthamiana leaves by agroinfiltration and infiltrated leaves were harvested after \(48\mathrm{~h}\) at \(22^{\circ}\mathrm{C}\) and subjected for LUC/REN luciferase activities assay. All the primers used for dual luciferase promoter activity assays are listed in Supplementary Table S7.  
+
+## Gene expression analysis  
+
+Leaves tissues were collected and flash frozen in liquid nitrogen immediately for total RNA isolation, total RNA was extracted from leaves tissues using the hot phenol method<sup>62</sup>, treated with RNase- free DNAase I (Sigma), and then reversed transcribed to cDNA using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA, USA). Quantitative real- time PCR was performed with Fast SYBR Green reagent (Applied Biosystems) in an Applied Biosystems Quantstudio real- time PCR system (Applied Biosystems). Gene- specific primers used for qRT- PCR are listed in Supplementary Table S7. The fold change was calculated according to the formula \(2^{-\Delta \Delta \mathrm{ct}}\) , and actin gene expression was used as a reference for normalization<sup>63</sup>.  
+
+## Chromatin immunoprecipitation (ChIP) analysis  
+
+ChIP assays were performed according to the reported protocol with some modifications<sup>64</sup>. Four- gram leaves of NAC90- 3 × Flag plants at 25 days were harvested, fixed in 1% formaldehyde under a vacuum for 10 min. Glycine (2 M) was added to a final concentration of 0.125 M and the leaves was vacuumed for another 5 min to stop the cross- linking. After that, the leaves tissues
+
+<--- Page Split --->
+
+were washed with 300 ml of cold ddH₂0 and dried with paper. The chromatin complexes were isolated and sonicated to an average size of 0.3 - 1 kb as previously described<sup>64</sup>. The sonicated chromatin complexes were incubated with polyclonal anti- Flag or pre- immune serum IgG (negative control) according to the previously described method (Sun et al., 2015). The eluted chromatin complexes were reversed and the precipitated DNA was purified using a DNA purification kit (Cat #N1073, GDSbio, China). The purified DNA was quantified by real- time PCR using specific primers (Supplementary Table S7).  
+
+## Phylogenetic tree construction  
+
+NAC protein sequences were extracted from the TAIR database (https://www.arabidopsisir.org) using HMM search based on the NAC domain (Pfam: PF02365) and alignment performed in Clustal X (version 2.1) with default multiple parameters and imported into the MEGA (Version X) software<sup>65</sup>. The phylogenetic tree was constructed using the neighbour- joining method with 1000 bootstrap replicates.  
+
+## Split luciferase complementation imaging (LCI) assay  
+
+LCI experiments were performed according to the described method<sup>60</sup>. The coding region of NAC90, NAC61, and NAC36 was amplified from the Arabidopsis DNA and purified with DNA purification kit (Cat #N1073, GDSbio, China). The fragments then were introduced into pCAMBIA1300- Cluc/Nluc to generate NAC90- Nluc, Cluc- NAC90, Cluc- NAC36, and NAC61- Nluc using the in- fusion method. Primers used for the LCI plasmids construction are listed in Supplementary Table S7. The destination plasmids were introduced into A. tumefaciens strain GV3101 and transiently expressed in N. benthamiana leaves through agroinfiltration. Forty- eight hours post infiltration, leaves were sprayed with 1 mM luciferin solution (1% Triton X- 100), kept in the dark for 5 min, and observed for luminescence.
+
+<--- Page Split --->
+
+## Co-Immunoprecipitation (CoIP) assay  
+
+CoIP assays were performed according to the described protocol<sup>60</sup>. Briefly, total proteins were extracted from agroinfiltrated N. benthamiana leaves using native extraction buffer and centrifuged twice at 16,000 g at 4 °C for 20 min. The supernatants were collected for input immunoblot analysis and immunoprecipitation. The supernatants were incubated with antibody anti- Flag and IgG conjugated magnetic beads (Sigma) for 4 hours with agitation. After that, beads were washed four times with native protein extraction buffer. Finally, the \(2 \times\) SDS loading buffer was added to the beads and boiled 5 min to release the proteins from the beads. The eluted proteins were immunoblotted using the anti- HA antibody (Abmart, China).  
+
+## Acknowledgments  
+
+The work is supported by the ERC- 2019- ADG project SIREM (884316). We thank the Adelis Foundation, the Leona M. and Harry B. Helmsley Charitable Trust, the Jeanne and Joseph Nissim Foundation for Life Sciences, the Tom and Sondra Rykoff Family Foundation Research, and the Raymond Burton Plant Genome Research Fund for supporting the Aharoni lab. AA is the incumbent of the Peter J. Cohn Professorial Chair.  
+
+## References  
+
+1. Jones JDG, Dangl JL. The plant immune system. Nature 444, 323-329 (2006).  
+
+2. Malamy J, Carr JP, Klessig DF, Raskin I. Salicylic acid: a likely endogenous signal in the resistance response of tobacco to viral infection. Science 250, 1002-1004 (1990).
+
+<--- Page Split --->
+
+3. Métraux JP, et al. Increase in salicylic acid at the onset of systemic acquired resistance in cucumber. Science 250, 1004-1006 (1990). 
+
+4. Nawrath C, Métraux J. Salicylic acid induction-deficient mutants of *Arabidopsis* express PR-2 and PR-5 and accumulate high levels of camalexin after pathogen inoculation. *Plant Cell* 11, 1393-1404 (1999). 
+
+5. Ward E, et al. Coordinate gene activity in response to agents that induce systemic acquired resistance. *Plant Cell* 3, 1085-1094 (1991). 
+
+6. Wu Y, et al. The *Arabidopsis* NPR1 protein is a receptor for the plant defense hormone salicylic acid. *Cell Rep* 1, 639-647 (2012). 
+
+7. Park S, Kaimoyo E, Kumar D, Mosher S, Klessig DF. Methyl salicylate is a critical mobile signal for plant systemic acquired resistance. *Science* 318, 113-116 (2007). 
+
+8. Jung H, Tschaplinski TJ, Wang L, Glazebrook J, Greenberg JT. Priming in systemic plant immunity. *Science* 324, 89-91 (2009). 
+
+9. Chanda B, et al. Glycerol-3-phosphate is a critical mobile inducer of systemic immunity in plants. *Nat Genet* 43, 421-427 (2011). 
+
+10. Chaturvedi R, et al. An abietane diterpenoid is a potent activator of systemic acquired resistance. *Plant J* 71, 161-172 (2012). 
+
+11. Chen Y, et al. N-hydroxy-pipecolic acid is a mobile metabolite that induces systemic disease resistance in *Arabidopsis*. *Proc Natl Acad Sci USA* 115, E4920-E4929 (2018). 
+
+12. Hartmann M, et al. Flavin monooxygenase-generated N-hydroxypipecolic acid Is a critical element of plant systemic immunity. *Cell* 173, 456-469.e416 (2018).
+
+<--- Page Split --->
+
+13. Schwarzenbacher RE, et al. The IBI1 receptor of \(\beta\) -aminobutyric acid interacts with VOZ transcription factors to regulate abscisic acid signaling and callose-associated defense. Mol Plant 13, 1455-1469 (2020).  
+
+14. Goto Y, et al. Exogenous treatment with glutamate induces immune responses in Arabidopsis. Mol Plant-Microbe Interact 33, 474-487 (2020).  
+
+15. Zeng H, Bai Y, Wei Y, Reiter RJ, Shi H. Phytomelatonin as a central molecule in plant disease resistance. J Exp Bot 73, 5874-5885 (2022).  
+
+16. Návarová H, Bernsdorff F, Döring A-C, Zeier J. Pipelagic acid, an endogenous mediator of defense amplification and priming, is a critical regulator of inducible plant immunity. Plant Cell 24, 5123-5141 (2012).  
+
+17. Tian H, Zhang Y. The emergence of a mobile signal for systemic acquired resistance. Plant Cell 31, 1414-1415 (2019).  
+
+18. Holmes EC, Chen Y, Sattely ES, Mudgett MB. An engineered pathway for \(N\) -hydroxy-pipelagic acid synthesis enhances systemic acquired resistance in tomato. Sci Signaling 12, eaay3066 (2019).  
+
+19. Yildiz I, et al. The mobile SAR signal \(N\) -hydroxy-pipelagic acid induces NPR1-dependent transcriptional reprogramming and immune priming. Plant Physiol 186, 1679-1705 (2021).  
+
+20. Hartmann M, Zeier J. l-lysine metabolism to \(N\) -hydroxy-pipelagic acid: an integral immunreactivating pathway in plants. Plant J 96, 5-21 (2018).  
+
+21. Cai J, et al. Glycosylation of \(N\) -hydroxy-pipelagic acid equilibrates between systemic acquired resistance response and plant growth. Mol Plant 14, 440-455 (2021).
+
+<--- Page Split --->
+
+22. Bauer S, et al. UGT76B1, a promiscuous hub of small molecule-based immune signaling, glucosylates N-hydroxypipecolic acid, and balances plant immunity. Plant Cell 33, 714-734 (2021).  
+
+23. Holmes EC, Chen Y, Mudgett MB, Sattely ES. Arabidopsis UGT76B1 glycosylates N-hydroxy-pipecolic acid and inactivates systemic acquired resistance in tomato. Plant Cell 33, 750-765 (2021).  
+
+24. Mohnike L, et al. The glycosyltransferase UGT76B1 modulates N-hydroxy-pipecolic acid homeostasis and plant immunity. Plant Cell 33, 735-749 (2021).  
+
+25. Wang Y, et al. A MPK3/6-WRKY33-ALD1-pipecolic acid regulatory loop contributes to systemic acquired resistance. Plant Cell 30, 2480-2494 (2018).  
+
+26. Guerra T, et al. Calcium-dependent protein kinase 5 links calcium signaling with N-hydroxy-1-pipecolic acid- and SARD1-dependent immune memory in systemic acquired resistance. New Phytol 225, 310-325 (2020).  
+
+27. Sun T, Zhang Y, Li Y, Zhang Q, Ding Y, Zhang Y. ChIP-seq reveals broad roles of SARD1 and CBP60g in regulating plant immunity. Nat Commun 6, 10159 (2015).  
+
+28. Sun T, Busta L, Zhang Q, Ding P, Jetter R, Zhang Y. TGACG-BINDING FACTOR 1 (TGA1) and TGA4 regulate salicylic acid and pipecolic acid biosynthesis by modulating the expression of SYSTEMIC ACQUIRED RESISTANCE DEFICIENT 1 (SARD1) and CALMODULIN-BINDING PROTEIN 60g (CBP60g). New Phytol 217, 344-354 (2018).  
+
+29. Sun T, et al. Redundant CAMTA transcription factors negatively regulate the biosynthesis of salicylic acid and N-hydroxypipecolic acid by modulating the expression of SARD1 and CBP60g. Mol Plant 13, 144-156 (2020).
+
+<--- Page Split --->
+
+30. Zhou M, et al. WRKY70 prevents axenic activation of plant immunity by direct repression of SARD1. New Phytol 217, 700-712 (2018).  
+
+31. Kim HJ, et al. Time-evolving genetic networks reveal a NAC troika that negatively regulates leaf senescence in Arabidopsis. Proc Natl Acad Sci USA 115, E4930-E4939 (2018).  
+
+32. Cai J, Aharoni A. Amino acids and their derivatives mediating defense priming and growth tradeoff. Curr Opin Plant Biol 69, 102288 (2022).  
+
+33. Bernsdorff F, Döring A, Gruner K, Schuck S, Bräutigam A, Zeier J. Pipeloclic acid orchestrates plant systemic acquired resistance and defense priming via salicylic acid-dependent and -independent pathways. Plant Cell 28, 102-129 (2016).  
+
+34. Huang W, Wu Z, Tian H, Li X, Zhang Y. Arabidopsis CALMODULIN-BINDING PROTEIN 60b plays dual roles in plant immunity. Plant Commun 2, 100213 (2021).  
+
+35. Li L, et al. Arabidopsis CBP60b is a central transcriptional activator of immunity. Plant Physiol 186, 1645-1659 (2021).  
+
+36. Ernst HA, Nina Olsen A, Skriver K, Larsen S, Lo Leggio L. Structure of the conserved domain of ANAC, a member of the NAC family of transcription factors. EMBO Rep 5, 297-303 (2004).  
+
+37. Olsen AN, Ernst HA, Leggio LL, Skriver K. DNA-binding specificity and molecular functions of NAC transcription factors. Plant Sci 169, 785-797 (2005).  
+
+38. Huang W, Wang Y, Li X, Zhang Y. Biosynthesis and regulation of salicylic acid and N-hydroxypipecolic acid in plant immunity. Mol Plant 13, 31-41 (2020).
+
+<--- Page Split --->
+
+39. Cui HT, Gobbato E, Kracher B, Qiu JD, Bautor J, Parker JE. A core function of EDS1 with PAD4 is to protect the salicylic acid defense sector in Arabidopsis immunity. New Phytol 213, 1802-1817 (2017).  
+
+40. Rekhter D, Mohnike L, Feussner K, Zienkiewicz K, Zhang Y, Feussner I. The plastidial exporter Enhanced Disease Susceptibility 5 is required for the biosynthesis of N-hydroxy pipecolic acid. bioRxiv, 630723 (2019).  
+
+41. Liu YN, et al. Diverse roles of the salicylic acid receptors NPR1 and NPR3/NPR4 in plant immunity. Plant Cell 32, 4002-4016 (2020).  
+
+42. Bethke G, et al. Activation of the Arabidopsis thaliana mitogen-activated protein kinase MPK11 by the flagellin-derived elicitor peptide, flg22. Mol Plant-Microbe Interact 25, 471-480 (2012).  
+
+43. Feys BJ, Moisan LJ, Newman M-A, Parker JE. Direct interaction between the Arabidopsis disease resistance signaling proteins, EDS1 and PAD4. EMBO J 20, 5400-5411 (2001).  
+
+44. Lu H, Rate DN, Song JT, Greenberg JT. ACD6, a novel ankyrin protein, is a regulator and an effector of salicylic acid signaling in the Arabidopsis defense response. Plant Cell 15, 2408-2420 (2003).  
+
+45. Mackey D, Belkhadir Y, Alonso JM, Ecker JR, Dangl JL. Arabidopsis RIN4 is a target of the type III virulence effector AvrRpt2 and modulates RPS2-mediated resistance. Cell 112, 379-389 (2003).  
+
+46. Zeilmaker T, et al. DOWNY MILDEW RESISTANT 6 and DMR6-LIKE OXYGENASE 1 are partially redundant but distinct suppressors of immunity in Arabidopsis. Plant J 81, 210-222 (2015).
+
+<--- Page Split --->
+
+47. Chang M, et al. PBS3 protects EDS1 from proteasome-mediated degradation in plant immunity. Mol Plant 12, 678-688 (2019).  
+
+48. Fu Z, et al. NPR3 and NPR4 are receptors for the immune signal salicylic acid in plants. Nature 486, 228-232 (2012).  
+
+49. Spoel S, Mou Z, Tada Y, Spivey NW, Genschik P, Dong X. Proteasome-mediated turnover of the transcription coactivator NPR1 plays dual roles in regulating plant immunity. Cell 137, 860-872 (2009).  
+
+50. Zheng X, et al. Coronatine promotes Pseudomonas syringae virulence in plants by activating a signaling cascade that inhibits salicylic acid accumulation. Cell Host Microbe 11, 587-596 (2012).  
+
+51. Poraty-Gavra L, et al. The Arabidopsis Rho of plants GTPase AtROP6 functions in developmental and pathogen response pathways. Plant Physiol 161, 1172-1188 (2013).  
+
+52. Fan J, Crooks C, Lamb C. High-throughput quantitative luminescence assay of the growth in planta of Pseudomonas syringae chromosomally tagged with Photorhabdus luminescens luxCDABE. Plant J 53, 393-399 (2008).  
+
+53. Rufian J, Rueda-Blanco J, Beuzón CR, Ruiz-Albert J. Protocol: an improved method to quantify activation of systemic acquired resistance (SAR). Plant Methods 15, 16 (2019).  
+
+54. Lei Y, Lu L, Liu H, Li S, Xing F, Chen L-L. CRISPR-P: a web tool for synthetic single-guide RNA design of CRISPR-system in plants. Mol Plant 7, 1494-1496 (2014).  
+
+55. Grützner R, et al. High-efficiency genome editing in plants mediated by a Cas9 gene containing multiple introns. Plant Commun 2, 100135 (2021).
+
+<--- Page Split --->
+
+56. Ma X, et al. A robust CRISPR/Cas9 system for convenient, high-efficiency multiplex genome editing in monocot and dicot plants. Mol Plant 8, 1274-1284 (2015).  
+
+57. Zhang X, Henriques R, Lin S, Niu Q, Chua N. Agrobacterium-mediated transformation of Arabidopsis thaliana using the floral dip method. Nat Protoc 1, 641-646 (2006).  
+
+58. Szymanski J, et al. Label-free deep shotgun proteomics reveals protein dynamics during tomato fruit tissues development. Plant J 90, 396-417 (2017).  
+
+59. Wisniewski JR, Zougman A, Nagaraj N, Mann M. Universal sample preparation method for proteome analysis. Nature Methods 6, 359-362 (2009).  
+
+60. Cai J, Chen T, Wang Y, Qin G, Tian S. SIREM1 triggers cell death by activating an oxidative burst and other regulators 1. Plant Physiol 183, 717-732 (2020).  
+
+61. Hellens RP, et al. Transient expression vectors for functional genomics, quantification of promoter activity and RNA silencing in plants. Plant Methods 1, (2005).  
+
+62. Cai J, Qin G, Chen T, Tian S. The mode of action of remorin1 in regulating fruit ripening at transcriptional and post-transcriptional levels. New Phytol 219, 1406-1420 (2018).  
+
+63. Livak K, Schmittgen TD. Analysis of relative gene expression data using real-time quantitative PCR and the \(2^{-\Delta \Delta \mathrm{CT}}\) method. Methods 25, 402-408 (2001).  
+
+64. Wang Y, Wang W, Cai J, Zhang Y, Qin G, Tian S. Tomato nuclear proteome reveals the involvement of specific E2 ubiquitin-conjugating enzymes in fruit ripening. Genome Biol 15, 548 (2014).  
+
+65. Kumar S, Stecher G, Li M, Knyaz C, Tamura K. MEGA X: molecular evolutionary genetics analysis across computing platforms. Mol Biol Evol 35, 1547-1549 (2018).
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1. Engineering the \(N\) -hydroxy pipecolic acid (NHP) biosynthetic pathway in Arabidopsis. </center>  
+
+(A) Phenotypes of NHP producing Arabidopsis plants. WT, wild-type Arabidopsis plant; pNHP, Arabidopsis plants overexpressing the NHP biosynthetic pathway genes. Two different pNHP independent lines are shown (marked as 1 and 2). Red arrows indicate yellow leaves in pNHP plants. (B) Small size of leaves and earlier senescence phenotypes in pNHP plants. (C) and (D) Abundance of NHP and SA in leaves of WT and pNHP plants. (E-H) Transcript levels of defense-related genes in the leaves of WT and pNHP plants. Leaves of WT and pNHP plants (pNHP-1 and pNHP-2) at 28 days were subjected to metabolite analysis by Quattro LC Triple Quad mass spectrometer and transcript analysis by RT-PCR. AGD2-LIKE DEFENSE RESPONSE PROTEIN 1, ALD1; SAR DEFICIENT 4, SARD4; FLAVIN MONOOXYGENASE 1, FMO1; PATHOGENESIS-RELATED GENE 1, PR1. In A, and B, bar = 1 cm. Error bars represent the SDs of three independent biological replicates. ND, not detectable. Asterisks indicate significant
+
+<--- Page Split --->
+
+750 changes compared to WT samples as calculated by Student's \(t\) test ( \(*p\) value \(< 0.05\) ; \(**p\)  
+
+751 value \(< 0.01\) ; \(***p\) value \(< 0.001\) ).
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2. Proteomics analysis reveals a potential regulator of NHP biosynthesis. (A) Label-free proteomics identified 724 proteins showing differential abundance in pNHP plants as compared to WT ones. A total of 4587 proteins (pink color) were identified and quantified. Of these, 656 (yellow color) and 68 (brown color) proteins were up-regulated or down-regulated in pNHP relative to WT plants. (B) Venn diagram showing overlapping transcripts/proteins between the differentially accumulated proteins found in this study and NHP response genes obtained from available transcriptomics data (16). The blue, orange, and brown colors represent the number of NHP-induced genes (NHP+), differentially accumulated proteins in pNHP plants (pNHP), and NHP-repressed genes (NHP-), respectively. (C) A heatmap displaying 725 differentially accumulated proteins in pNHP plants relative to their expression in WT samples. The color scale represents log2 fold change. Each row in the color heat map represents a single protein. Two different pNHP independent lines are shown (marked as 1 and 2) (D) GO annotation of proteins detected as differentially accumulating in pNHP plants as compared to the WT ones. (E) Relative </center>
+
+<--- Page Split --->
+
+abundance of the NAC90 protein in WT and \(pNHP\) plants. Values showed here were the average of three biological replicates and error bars represent the SDs of three independent biological replicates. Asterisks indicate significant changes compared to WT samples as calculated by a Student's \(t\) test (\\*\\*\\* \(p\) value \(< 0.001\) ).
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3. Pathogen-dependent induction of NAC90 is mediated by NHP, SA, and SARD1. (A) </center>  
+
+Transcript levels of NAC90 in 4-week-old WT and pNHP plants. (B) Transcript levels of NAC90  
+
+in WT Arabidopsis plants in response to Psm ES4326, NHP, and SA treatments. Leaves of 4-  
+
+week- old WT Arabidopsis plants were treated with \(10\mathrm{mM}\mathrm{MgCl}_2\) (Mock), Psm ES4326 (OD600 =
+
+<--- Page Split --->
+
+0.001) (Psm), 1 mM NHP (NHP), and 200 μM SA (SA) solution, harvested at 24 hours post treatments, and subjected for qRT- PCR analysis. (C- D) Transcript levels of NAC90 in different SAR deficient mutants. Leaves of 4- week- old WT and mutant plants (AGD2- like defense response protein 1, ald1; flavin monooxygenase 1, fmol; salicylic acid induction- deficient 2, sid2; non- expression of PR1, npr1; enhanced disease susceptibility 1, eds1; phytoalexin- deficient 4, pad4; sar deficient 1, sard1; Calmodulin- binding protein 60g, cbp60g) were treated with 10 mM MgCl2 (Mock) and Psm ES4326 (OD600 = 0.001), harvested at 24 hours post treatments (hpi), and subjected for transcript levels analysis. Ten mM MgCl2 was used as mock control (Mock). (E) Transactivation of NAC90 promoters with SARD1 and CBP60g. All the values showed here were the average of three biological replicates and error bars represent the SDs of three biological replicates. Asterisks indicate significant changes compared to WT or Mock samples as calculated by a Student's t test (\*p value < 0.05; \*\*p value < 0.01; \*\*\*p value < 0.001).
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4. Characterization of NAC90 mutants and overexpression plants. (A) and (B) Transcript levels of NAC90 in leaves of 3-week-old wild-type (WT), nac90 mutants (nac90-1 and nac90-2) and NAC90 overexpression (NAC90-1 and NAC90-2) plants. (C) Western blot (Anti-Flag) showing NAC90 protein abundance in the leaves of 4-week-old WT and NAC90 overexpression plants (NAC90-1 and NAC90-2). (D) Phenotype of WT, nac90-1, nac90-2, and NAC90 </center>
+
+<--- Page Split --->
+
+overexpression (OE- NAC90- 1 and OE- NAC90- 2) plants. Bar = 1 cm. Red triangles indicate senescence leaves in both nac90- 1 and nac90- 2 mutants. Bar = 1 cm. (E) Go annotation analysis of differentially expressed genes in both nac90 mutants (nac90- 1 and nac90- 2). (F) Heatmap showing differentially expressed genes identified in the transcriptome data of both nac90 mutants (nac90- 1 and nac90- 2). The color scale represents log2 fold change. Each row in the color heat map represents a single gene. AGD2- LIKE DEFENSE RESPONSE PROTEIN 1, ALD1; NAC DOMAIN CONTAINING PROTEIN 61, NAC061; NAC DOMAIN CONTAINING PROTEIN 36, NAC036; SAR DEFICIENT 1, SARD1; SENESCENCE- ASSOCIATED GENE 13, SAG13; FLAVIN MONOOXYGENASE 1, FMO1; CHITINASE, CHI; ENHANCED DISEASE SUSCEPTIBILITY 5, EDS5; CALMODULIN- BINDING PROTEIN 60g, CBP60g; PATHOGENESIS- RELATED GENE 1, PR1. Values showed here were the average of three biological replicates and error bars represent the SDs of three biological replicates. Asterisks indicate significant changes compared to WT samples as calculated by Student's \(t\) test (\\*\\*\\* \(p\) value \(< 0.001\) ).
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Figure 5. NAC90 negatively regulates plant immunity by controlling the NHP and SA biosynthetic pathways. (A-D) Transcript levels of NHP (ALD1, SARD4, and FMO1) and SA (ICS1) biosynthetic genes in the leaves of 4-week-old WT, nac90-1, nac90-2, and NAC90 overexpression (NAC90-1 and NAC90-2) plants. (E) Chromatin immunoprecipitation-qPCR assay indicated that NAC90 directly binds to the upstream region of NHP biosynthetic genes. (F) Dual luciferase promoter activity assays showed that NAC90 represses the expression of NHP biosynthetic genes. (G-H) Abundance of NHP (G) and SA (H) in the leaves of 4-week-old WT, nac90-1, nac90-2, </center>
+
+<--- Page Split --->
+
+and NAC90 overexpression (NAC90- 1 and NAC90- 2) plants. (I) Growth of Psm ES4326 in leaves of WT, nac90- 1, nac90- 2, and NAC90 overexpression (NAC90- 1 and NAC90- 2) plants. (J) Symptoms of leaves treated with Psm ES4326 observed in WT, nac90- 1, nac90- 2, and NAC90 overexpression (NAC90- 1 and NAC90- 2) plants. (K) Transcript levels of PR1 in the leaves of 4- week- old WT, nac90- 1, nac90- 2, and NAC90 overexpression (NAC90- 1 and NAC90- 2) plants. Values showed here were the average of three biological replicates and error bars represent the SDs of three biological replicates. Asterisks indicate significant changes compared to WT or control samples as calculated by Student's \(t\) test ( \(*p\) value \(< 0.05\) ; \(**p\) value \(< 0.01\) ; \(***p\) value \(< 0.001\) ).
+
+<--- Page Split --->
+![](images/Figure_6.jpg)
+
+<center>Figure 6. FMO1 is required for constitutive activation of defense response in the nac90 mutant background. (A) FMO1 determines the size of the rosette in nac90-1 plant. Four-week-old wild-type (WT), fmo1, nac90-1, nac90-1 fmo1, and nac90-1 sid2 plants was examined for their morphology. Bar = 1 cm. (B) and (C) Abundance of NHP (B) and SA (C) in the leaves of 4-week-old WT, fmo1, nac90-1, nac90-1 fmo1, and nac90-1 sid2 plants. (D) Growth of Psm ES4326 in the leaves of WT, fmo1, nac90-1, nac90-1 fmo1, and nac90-1 sid2 plants. (E) Symptoms detected in leaves of WT, fmo1, nac90-1, nac90-1 fmo1, and nac90-1 sid2 plants treated with Psm ES44326. </center>
+
+<--- Page Split --->
+
+Leaves were treated with \(Psm\) ES4326 (OD \(_{600} = 0.001\) ) and photographed 3 dpi. Bar \(= 1 \mathrm{cm}\) . (G) Transcript levels of PR1 in leaves of WT, fmo1, nac90- 1, nac90- 1 fmo1, and nac90- 1 sid2 plants. Values showed here were the average of three biological replicates and error bars represent the SDs of three biological replicates. Different letters above the bars denote significant differences among treatments ( \(p< 0.05\) , ANOVA).
+
+<--- Page Split --->
+![](images/Figure_7.jpg)
+
+<center>Figure 7. A triad of NAC transcription factor cooperate in the regulation of NHP biosynthesis. (A) </center>  
+
+Co- expression analysis identified potential transcriptional regulators of NHP biosynthesis. Each triangle in the color nightingale rose charts represents a single gene family. (B) and (C) Phenotype
+
+<--- Page Split --->
+
+of 4- week- old wild- type (WT), nac36 mutants (B, nac36- 1 and nac36- 2), and nac61 mutants (C, nac61- 1 and nac61- 2). Bar = 1 cm. (D) Growth of Psm ES4326 on leaves of 4- week- old WT and nac36 mutants. (E) and (F) Abundance of NHP (E) and SA (F) in leaves of 4- week- old WT and nac36 mutants (nac36- 1 and nac36- 2). (G) Co- immunoprecipitation analysis analyzing the interactions between the NAC90, NAC61, and NAC36 proteins. Total protein (input) was extracted from agroinfiltrated N. benthamiana leaves and co- immunoprecipitated with anti- Flag. The immunoprecipitated proteins (IP) were analyzed by immunoblotting using anti- HA antibody (α- HA). (H- J) Split luciferase complementation assays showing the interactions between the NAC90, NAC61, and NAC36 proteins. Images were acquired at 2 days post infiltration. Red circles indicate infiltrated areas. (K) A schematic depicting the ‘gas- and- brake’ model of NHP and SA biosynthetic pathways regulation including negative regulation by the NAC protein triad as revealed in this study. Values showed here were the average of three biological replicates and error bars represent the SDs of three biological replicates. Asterisks indicate significant changes compared to WT samples as calculated by Student’s \(t\) test ( \(^{*}p\) value \(< 0.05\) ; \(^{**}p\) value \(< 0.01\) ; \(^{***}p\) value \(< 0.001\) ).
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+Supplementarydata.xlsx  SupplementaryFigure.docx
+
+<--- Page Split --->